HF has been a close part of my ML/AI career, coinciding exactly when I moved into this space 10 years ago. There are lot of nuances here (if the deal goes through). Some people say it's a loss for EU sovereign AI but HF is technically an American corporation. On the positive note, the founders (Julien, Thomas and Clem - all French) stem to make significant amount of money, which they are likely to pour into a new frontier AI lab in Europe. So potentially it's a big win. For Nvidia this is a great strategic play as it potentially gains control of the "AI app store" and can influence the direction of Transformers, Diffusers, PEFT and various other HF libs inits favour. At the same time Nvidia has been (somewhat paradoxically) a dominant force in actual "open source" AI and has contributed significantly via Nemotron and various optimisers close to the metal. This would likely accelerate further. Nvidia has everything to gain from having a massive open model ecosystem, instead of a consolidated market consisting of 2-3 players. Will that have impact on MLX and other contributions? We'll see!
And I would say it's actually a more secure future demand channel than the hyperscalers. Openai and co can build their own chips, so they won't buy from NVidia. But they probably won't sell them to somebody else, so everyone who wants to run their own install (and can, not the least because of HF), will buy Nvidia.
Translating CUDA kernels to ROCm (or any other ecosystem, tbh) is going to become trivially easy with AI. Not saying that code is the only glue that Nvidia has when it comes to their ecosystem, but it’s certainly a large part of it.
> which they are likely to pour into a new frontier AI lab in Europe
HF was.. a hub to download models and some of the worst source code in the space that contributed to huge amounts bugs that did a lot of downstream damage. Go ahead and read their blog by their CTO on how they don’t believe in DRY and then implemented DRY in the worst way imaginable with unnecessary code generation.
Their BLOOM model was a joke and dead on arrival.
But yea these guys are definitely going to be the frontier of EU AI.
They’ve provided insane value to the ai community. Imo is there really any obvious way to standardize model/preprocessing/inference/training/rl recipes across the space of frankenstein model family/architecture/modality combinations? It’s not even clear to me that the current frontier would be where it is today without HF doing what they did how they did it. Where else were people widely sharing datasets w that ease of reuse/redistribution, or sharing their retuned models etc? Would there even be as much interest/tinkerers today?
It has to be a bell curve doesn't it.. I am too much of a perfectionist...
Then again, my llms constantly claim that I am doing things that are beyond the SOTA. Even my own mom lost hope.. Still haven't shipped a single thing... haha.
(half joke aside, it is finally coming in a couple days.. which I have been saying for quite some time to the extent that my peeps don't even trust me anymore.. 3 things are coming and it is genuinely research grade apparently, for the amount of research I am aware that is released publicly.
I really don't know how people ship things, the tooling I see is terrible, or maybe I have NIH syndrome...)
"The company was founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf in New York City, originally as a company that developed a chatbot app targeted at teenagers. ..." Wikipedia
Besides that: which EU and sovereignty doesn't really rhyme, given data protection, energy costs, AI restrictions and no vital start up scene at all.
Mistral is heavily state funded via Banks that are predominantly owned by the French state, so I don't get it.
Also name one global success that doesn't came from US? China? Well, second place maybe, but second in any sense.
Nevertheless, kudos to huggingface. Not only to the founders, but to the ecosystem and I guess it got harder by the day to finance all the hosting and only hope for the future, that Nvidia will help HF to thrive, at least not put burdens on them.
GitHub is still alive, and hope that HF will benefit from more AI power.
Nvidia doesn't have any disadvantages in my playbook, because hardware wise, there is no one to beat them in any regard and Apple will always stay nichy.
So well played - hopefully.
It could be worse or even a shut down otherwise, who knows. So for me it sounds cool and I am glad that EU isn't involved in any regard.
As the AI Act demands full disclosure under really easy to match requirements into your inner workings, this is quite the opposite of freely sharing information.
The list goes on for products and services they cloned from western innovation. China didn't get there first in any of the categories you listed. Of course, people would rather spend less than more, disregarding the quality and longevity of the product.
Paying pennies on the dollar for a BYD vehicle sounds great when all you consider is the initial sticker price. The real cost is what happens after you buy the disposable car and own it for a decade.
So yea. Second place is the reality and it is also true for AI models, and the hardware needed to run them.
Also, congratulations on landing a rocket 11 years after the US.
Sorry but you're talking absolute nonsense, Europe has no global successes?
How about...
SAP (Germany) - Most of the Fortune 500 run on their software.
Spotify (Sweden)
Novo Nordisk (Denmark) - most of America is taking their drugs.
Airbus (France) - Has out competed and out delivered Boeing for years.
Ericsson and Nokia (Finland and Sweden) - Most of American telecom systems use tech from them, essentially the western alternative to Huawei in telecom networking.
Adyen (Netherlands) - One of the worlds largest payment processors, processes most of American big tech payments.
ASML (Netherlands) - Responsible for building effectively every modern chip in the world outside of China.
Maersk (Shipping and logistics)
Siemens (Industrial Tech)
Infineon (Semis)
And there are so many more I'm missing btw. It's funny how incredibly biased you are, but the reality is that Europe and the USA are totally dependent on each other, tech wise.
If every European company which the US is dependent on immediately stopped trading with the US, the US would collapse as a country. And yes, the opposite is true also. But it was you who argued that Europe effectively has no successful companies, what rubbish.
SAP isn’t a company to be crazy proud of, but it doesn’t make it not a success nor does it mean “EU is fucked”.
May I remind you that in the US, Facebook is in that same list of top successes? Have you ever heard the phrase “the top minds of this generation of Americans are all working on ads”?
Also, if huggingface is “technically” American, then fuck me, Stripe is technically European. Hell, their CEO is insanely pro europe, pro EU, and a big enabler of projects like EU Inc and lately, the Rhine Group.
I am quite happy to see such a successful outcome for this team. When looking at the landscape of AI-adjacent startups, rarely do you see such a nice balance between following the hype and providing a sufficiently useful and general service that makes technology more accessible to all.
The $13b doesn't go to hugging face's bank acct to pay for anything like S3 egress fees... It go to hugging face's owners' Bank accounts for them to do anything else. New cohorts of billionaires and centi millionaires getting minted.
It’s a core part of the open model infrastructure. Their hardware business benefits greatly if this works really well and open models proliferate to every corner of the economy, with everyone buying GPUs with a lower capacity factor/duty cycle than the centralized ones. It also helps people across the world to collaborate on new uses for this, effectively running a massive search in the search space of what’s possible with these things, for every year (see the endless variety of quants and finetunes to fit in different cards, and bias their abilities on different tasks). More use cases pop up, some of them are killer apps, buying Nvidia’s hardware becomes more valuable at whatever price point, more people do it at higher prices, Nvidia stock chart goes up.
If Anthropic, OpenAI buy them, this org gets very different priorities, owning them ensures that doesn’t happen.
Hugging Face's internal storage system is more sophisticated now after they acquired XetHub. They have a clever way to chunk and content dedupe the data that makes it a lot more efficient to version control large files.
Xet's tech is very interesting and I greatly enjoyed their blog posts and other technical writeups[0]
But in the era of frontier models and hardware designed for fast model loading (eg: LLM on a chip) I can't imagine that this is worth so much. There's *a lot* of competition to improve Git LFS.
So while I agree with you that this plays a role I also have to believe that it's to understand model usage by competition and thus stifle it.
I highly doubt it, there's about a million things that only show up at scale that aren't lessons the bots can easily digest from Medium posts and bypass
There's more to than just the website. You have network bandwidth costs, security costs, infrastructure costs.
You can't just spin up a VPS and throw a few brain jar's in; the network bandwidth must be a fortune. Hosting over 700GB+ for many of the same models has storage cost too.
Call me a conspiracy theorist, but I think it might have something to do with OpenAI's model hacking HF thing, especially considering OpenAI is one of Nvidia's most important customers.
They are paying for an insurance policy. Elon Musk and SpaceXAI must be incredibly interested in buying the company who was the victim of a mass hacking campaign by OpenAI. Any lawsuit, criminal investigation, or slowdown of OpenAI's model training would put OpenAI's massive data center build out, which Nvidia just backstopped, at risk.
Any DA can investigate it, they don't need Musk or some other big tech dipshit to tell them otherwise. Real crimes were committed and no charges are being pressed.
It's a narrative to pump the stock. Nothing more. Nvidia could have build out this infrastructure themselves for less than 1% of the cost, but it wouldn't have a name like Huggingface does.
Did you read the model hack that occured on their platform? I wouldn't exactly call them "advanced" if their security practices are anything to go by.
Code is hard but more likely it is a brand thing which is much harder to do. Nvidia for years now has tried to get a public ai model hub of sorts working. This is obviously them throwing in the towel.
It's such an obvious customer funnel they have been trying for years to make.
1. Customer googles "some model"
2. Hugging face is often top result
3. Thinks it's great
4. Buys
The customers are already on hugging face.
Acting like code is the blocker here would be very wrong. Nvidia 200% has the technical knowhow to make a hugging face platform and they'd likely make it better.
Buying a company is very rarely about buying the tech, at least principally. It is true, as others have pointed out, that Nvidia could have "rolled their own" Hugging Face far more cheaply and probably in about the same amount of time as it will take to execute the deal end to end. That should give you a clue that the tech isn't really the target here
More than anything, they are buying a position within the ecosystem. Many of us could get a version of an app deployed in short order that implements most core HF functionality, but I am sorry to say that this is very far from the "hard part" of building a business, and this was the case even before coding agents were a thing. It's no small thing to become the first tool pretty much everyone reaches for in a particular niche, and therefore the incumbent in that niche
And there's lots of inertia keeping companies and developers on these incumbents like HF and GitHub -- see also Microsoft's play here dumping billions into a company that is basically peanuts and probably nowhere near as profitable as Microsoft normally would like. But they bought it at a time when they were trying to make Microsoft synonymous with open source, and it was a great play to that end. Likewise Nvidia wants to be synonymous with AI. Even if HF is a loss leader, the positioning is probably worth it
To a lesser degree, Nvidia probably also might like to reinvigorate its flagging cloud compute business and this could be a natural way to do that. At least TechCrunch thinks this, but I, an overconfident layperson, feel that the positioning is where most of the value is for Nvidia, and consolidating that will act as a force multiplier for that + anything else they may want to do with this space
Wasn’t Huang the guy who said your $500K employee should burn at least half his salary on tokens? I don’t think Nvidia is focused on cost-effective general computing for anybody.
Maybe but they make their real money in data centers. Certainly, they want to have fingers on every trend. Owning something the Government wants to control gives you more access to trade repression of open models for more Government favors and money.
Nvidia is halving the production of its higher end GPUs, doubling prices and heavily segmenting the market. They don't give a single shit about individuals. One wafer that makes 10 5090s that might sell at 3000 each, or one wafer that it already sold 6 months ago for 200k to one of the incestuous AI companies it works with?
I guess this unfortunately means HuggingFace won't be "the first company to go public with an emoji instead of the three-letter ticker" as the cofounders originally intended:
"When we started the company, a running joke with my co-founders was that we wanted to be the first company to go public with an emoji instead of the three-letter ticker when you go in the NASDAQ."
-- Clem Delangue (https://unsupervisedlearning.substack.com/p/the-future-of-open-vs-closed-source)
Historically NYSE (and AMEX; thought of in this specific context as CTA) had a lock on 1-3 character symbols. NASDAQ (via UTP) had a lock on 4-6 character symbols.
For a long time you could tell where something was listed purely by looking at the length of the symbol it traded under.
This all changed in the late 2000s or so, so no longer useful information.
I honestly would leave the industry if one of the exchanges tried to bend the industry to allow an emoji in a symbol. The communication protocols essentially lock this at 8 character alpha. It would be a huge pain in the ass, for everyone, to change this just because some exchange wants to appeal to a potential listing.
Yeah, they tried to carry the suffix symbology into the ticker itself, but that fortunately died not because of some great sanctity of the symbols but because someone thought it would confuse too much with how ACT and the CTCI protocol handled security categorization.
NYSE (really the access stack that both NYSE and AMEX ran on; the name escapes me since it died back in I think ‘08) has for a long time included a completely separate field in their protocols to indicate classification (preferred, warrants, classes, when issued, etc), but CTCI just jammed it into the symbol field.
Amusingly there is plenty of precedent (link below under Nasdaq Integrated Platform Suffix and the older ACT/CTCI suffixes) for using all of the wonderful character symbols to indicate different classes of securities.
Not as common, again, as it used to be...but still exists.
Nasdaq becomes more and more like a “normal” exchange every year. Historically, they’ve always just done their own thing for their own purpose, so it makes sense.
** for those that don’t know: NASDAQ stands for “National Association of Security Dealers - Automated Quotation (service)”. It was forced into existence by the SEC. Prior to that there was only the NASD, who traded blocks of shares amongst themselves so that each individual security dealer could meet the needs of their customers. As the trading amongst themselves grew larger and larger, the SEC added progressively more regulation until finally the had to build the NASDAQ computer system, which nobody would even recognize today.
So again, they were always separate from everything else and did things that made sense according to their own needs. They were very different until they were no longer allowed to be different.
Don’t disagree there, though this has been status quo for nearly two decades now.
By that measure regional exchanges were their own beasts for a long time. The environment has definitely streamlined a ton since RegNMS was implemented and the exchanges consolidated a bit.
Even the technology stacks the current markets run are basically identical within three families of design (even the upstart markets, and even those that are not directly licensing technology from ICE, Nasdaq, or Cboe). Differentiation just doesn’t pay like it used to.
I tried for a while to push internally and amongst a few of the markets to move to ISIN, Currency, MIC but yeah…anything but that.
Honestly the problem is the ticker is too valuable advertising for the listed company, and since listings are not governed by a central regulator in the same way they are in Europe, for example, there’s always going to be this sort of friction about who is willing to bend the most to win the business.
I think ISINs being used in Europe might even predate the EU taking charge of a lot of financial market regulation.
Probably they became more popular/important specifically because tickers did not, which in turn is probably because there are so many countries/exchanges that a global exchange-driven namespace was never really feasible?
I keep reading this about old.reddit.com and yet I keep using it without being logged in. It must be some kind of A/B testing or slow rollout? Or maybe it's regional?
That stopped working for me on my mobile firefox (but not on my desktop) literally yesterday. It's pretty clear they're doing some kind of A/B testing.
Obviously, NVIDIA is trying to own the AI development chain.
Owning HF -- the discovery and distribution channel -- is one thing, but I think the biggest threat vector is the privileged access to HF platform data, that includes HW survey info and model download pattern. This can be a borderline anti-trust case.
This brings back memories of the CVS/Aetna merger for me, which ended up going through of course. Hopefully the anti-trust enforcers have some experts in the ML/AI pipeline, but I wouldn't be surprised if they didn't bat an eye at this one. Hope I'm proved wrong though.
Also they bought the company which develops the open source slurm scheduler which is the primary and defacto standard HPC scheduler on the market. Bit of a concern to say the least
The more interesting is HF turned down a $500M Nvidia investment late last year at
a $7B valuation, after passing on a $235M round in 2023 at $4.5B —
going from "we don't want a dominant investor" to a $13B full
acquisition in under a year is quite the reversal.
Oh yeah, so surprising to refuse $500M that go in the pockets of your company that is now owned in large part by the investor, but then accept $13B that go directly in your own pockets.
Reversal would be "valued at 7B then valued at nothing", this is more like "nah, we want more" then "nah, we want more" then "yes, that's what we want :)".
its always the most virtue signaling people (sam on "openess", dario on pretty much every word coming out of his mouth, HF) that ends up doing the worst things because they always tend to extreme. Funny to see them joining the biggest corporation on earth by market cap after the every day virtue signaling of pretty much every employee working for them.
very few people can lock a vision and fight billions dollars with integrity. In tech they mostly end up shadowed. People like JB Kempf with VLC or Bellard. Maybe even Torvald or Jeremy Howard belong to this list of people who could have been way richer than they are now if they wanted to go against their visions and beliefs.
I don't think they passed on the 4.5 billion. They still raised it in 2023 with investors like Alphabet etc. They turned Nvidia *earlier this year* for the $7 billion valuation deal. It makes sense if they were angling for an acquisition.
This trend to vertical integration is very dangerous. We should not allow the people making tech to control the whole pipeline unless we want a permanent underclass.
Something something Apple. For what it's worth though, I agree that companies _subsuming_ others to obtain complete vertical market capture has been a very dangerous trend.
Apple isnt an analog though, in order to be a relevant analogy they would need to buy Intel, TSMC and ASML and "commit" to preserving access to other people chip manufacturing. As it is, they are guilty of reserving capacity so other producers cant access current generation tech and have to weight a cycle, which I would argue should be blocked as anticompetitive by regulators but we have a perverse definition of free markets now that defines market consolidation and monopoly as markets working instead of recognizing they are a sign that a free market has been captured and destroyed.
They have various paid services that relate to AI—paid inference hosting; paid accounts aimed at AI development with GPU rentals [credits plus paid overages, I believe], more private storage and public storage than free accounts; and many of the community a and some other benefits on individual/team/enterprise tiers; additional paid storage above the base quotas for the paid account tiers; on demand rentals of HF managed containers on GCP and AWS, and some other things.
I’m asking more to work out what is the basis of this valuation. Why would Nvidia spend $13B for what seems to be a services that gives things away for free.
I dunno. You can argue over whether they're overpaying, but it's not like Huggingface is Clinkle. They hit $150 million in ARR this year, they have tons of runway, and according to reports, have just started to even burn the money they raised a few years ago.
I get that it's fun to be glib about the stupidity of tech elites and investors in general, but Huggingface have been pretty open about their financials and are, in my opinion as a practitioner in the field, one of the most responsible orgs in our space. They've been a pillar of open source ML for years now and have made a very positive impact on our ecosystem.
Nvidia is getting a real business generating revenue, and the center of the universe for open models. Both seem like pretty valuable attributes, from Nvidia's perspective.
Solid point. I'm sure Nvidia went into this deal expecting completely flat growth and no other benefits to their core business. Sorta like how Meta never increased Instagram's revenue from $0 and is still waiting for it to pay off that billion dollar acquisition price.
Or like GitHub, which was generating something like 200 million in ARR and had never hit profitability when Microsoft bought it for $7.5 billion back in 2018. I'm sure it has come as nothing but a happy surprise to Microsoft that GitHub generated $1 billion in 2023. They had initially penciled it in for 38 years til ROI.
Basic napkin math: 5% IRR means they'd only need to 4.5x their revenue to make this roughly work. If they can finance this cheaper and/or do not have better options for their cash, it's even less.
It's not too dissimilar from GitHub, but geared towards ML. They have a 9/mo pro plan for individual users for upgraded storage/usage, and an enterprise version of Hub that larger orgs can pay for. I think the enterprise has some contract minimum + 50/mo per seat. https://huggingface.co/pro
They also have inference endpoints with metered prices, and their spaces product (though i'd imagine this is a smaller portion of revenue).
I’d be curious to see how much that’s subsidized. Maybe they’re different, but when I see ML and monthly plan in the same sentence, I see zero sustainability.
Nvidia has a market cap of $5T USD today, and a decent chunk of that is due to LLM speculation.
Does Nvidia want their stock price to be at risk of being tanked by a download service being in the news? No, they want to make sure the party keeps going and is under their direct supervision, and part of that is making sure Hugging Face isn't bought by a competitor or runs out of money.
If their stock moves up more than third of percent due to this they have won... The math at these valuations get somewhat extreme. But there is some logic in it.
There is a brand and that brand is really a type of call option that is very hard to price.
If you are doing the valuation using the tools of Ben Graham it is not really going to work.
Damodaran's the Dark Side of Valuation is the text for this.
Yeah, great point. Nvidia could potentially be overpaying. Not sure how that equates to Huggingface being "a file download mirror with a couple of side features dangling off".
Absolutely none of this makes sense and the thing that amazes me is how long it has continued. Future historians will just laugh at how stupid and obvious the crash was.
While it's probably easier to say this in retrospect, they were eliminating a direct competitor to their core business. Something so advantageous it should have been blocked by regulators.
I can't see this as being as good of a purchase, especially when it's 13x the price of what was seen as an absurdly large amount back then.
I guess we may have to start paying to download models.
Or perhaps they will start throttling downloads for free users.
I don't know what they business case is, it might be to shut them down: I suspect good free models on local hardware is a threat to Nvidia's investments in OpenAI/Anthropic.
NVidia has been expending energy helping improve local models and inference platforms for them targeting NVidia GPUs; good free models that users can run locally rewards Nvidia’s investment in product lines for local inference (DGX, RTX PCs, etc), as well as—given their continued dominance in the space—the premium over competitors of their consumer and workstation GPUs.
Maybe they're trying to see how big the market actually is before considering shutting it down, maybe or getting lawyers and politicians to try and outlaw or restrict open models if they see a big enough opportunity.
The open weights AI ecosystem is way too concentrated with HF. That said, it was clear they were going to either get bought or aggressively monetized at some point, Nvidia seems like it could be one of the more aligned acquirers.
I do worry about any sort of crowding out or downplaying non Nvidia-relevant quants, and about changing rules to crack down on models or datasets that for one reason or another “don’t align with their corporate values” - uncensored etc. Someone mentioned Microsoft and GitHub, they appear to me anyway to have been very hands off, I hope it’s the same model.
It's already part of Alibaba, which has its own incentives. So does kaggle with Google. Civitai is neutral but it is mostly nsfw even though they don't show it outright. Torrenting is the Best way
I could put on a tinfoil hat and say that nvidia will have some control on the flow of the open models, so if someone like the US government want to "control" and police the models like Dario was talking about then this investment might actually make sense. The HF platform can then be used as the "controlled" channel and any other site can be designated a national security risk and blocked / taken down.
"Quantized models are no longer permitted on the Hugging Face platform. Reduced precision models are a safety hazard and a violation of our TOS. Click here to speak with a sales representative about our many exciting cloud hosting offerings or enterprise GPU packages."
This acquisition and the Stripe acquisition of OpenRouter are puzzling. Both HF and OR, from a technological standpoint, don't appear to have much IP that cannot be easily replicated.
I wonder how much it's worth to Nvidia just to obtain a list of who's downloading which models and for what purpose. Similarly with the Stripe acquisition of OpenRouter. In the past, it would have been valuable to obtain an audience or market share or even a productive team. However, in this new age of the surveillance state ...
America is headed towards techno zaibatsu economy.
Just a product of the dynamics of capital and power. Too much cash, Nvidia is forced to try and find ways to deploy excess capital very quickly. Most efficient way is to absorb players in emerging industries, bet on the potential for growth in new markets to help keep ROI up.
It is a great day for startups, the goal is to sell out and get rich, no?
INB4 China comes in and dismantles the American zaibatsus in 2245 like America did to Japan in 1945.
We simply need to bring back that 90% federal income tax bracket we had in the 1950s. If we had the same tax policy today, a CEO or investor would be limited to 5 million dollars a year in earnings. This is what fixed everything.
If a CEO cannot personally get richer by cutting wages, cutting staffing, and outsourcing; all of it stops. That was the secret of affordability. Don't let a thousand rich people own everythjng. Tax them so we can have millions of working millionaires instead of 900 non-working billionaire elites.
As those 1950s taxes and post 1929 financial regulations were cut in the following decades, the wealth inequality that led to the great depression has returned.
This is more a result of Clinton's failed CEO pay reform. It is what created this monster, not the tax bracket. Now, saying that, we DO need to tax these people higher. Eliminate all the ways they're able to squirm out from paying a fair share.
Marginal tax rates are meaningless because, even in the 1950s, the system was complex and can be gamed.
If you want to judge the net effect of tax policy over time, look at the percentage of GDP captured by federal receipts (the vast majority of which are taxes). In the U.S., it's averaged ~16.5%-~17.5% with occasional spikes and dips. Currently, it's very close to what it was in those 90%-marginal-rates 1950s.
I disagree on this. It's not the wealth that is the issue, it's the corruption that allows wealth to be converted into power. Citizens united is so backwards that it actually encodes the corruption as a right.
People have a right to support candidates > they can print a sign and stand on a street corner > printing signs costs money > people have a right to spend money supporting candidates.
People have a right to assemble > they can stand on a street corner with their friends > they can hold a big sign together > they can have lots of friends and form a group and print a really big sign.
We want to encourage private business > limiting liability for owners who do not participate in the business would increase access to capital > we offer corporate forms that shield investors from liability > identity of owners may be kept confidential.
People complain about the wrong thing. The problem isn't that people are allowed to support candidates with their money, buy airtime, etc.. It's that companies have secret owners...
You said "People" a lot in your post. Corporations are not "people".
Only natural persons should be able to contribute to political candidates.
Corporations can lobby, as a corporate person, just like a union or other organization, but not contribute.
They can run ads themselves, but there needs to be a much greater wall between PACs and candidate campaigns, with the previous restrictions now null and void due to lack of enforcement.
If it's okay for Corporations can run ads themselves, then your objection isn't to wealthy people supporting candidates or even them doing so secretly, but... only when the little guy pools donations and they happen to be pool donations via a corporation?
No, state's cannot e.g. discriminate on the basis of race by allowing corporations only composed of certain races. States also cannot infringe on the right to assemble or speak, by prohibiting people from supporting political candidates when they are organized as such a group.
But, generally, yes. States can require that the owners / members / corporate books / etc. of a company be disclosed publicly. If owners don't want their books disclosed, then they can forego the immunity that the corporate forms provide.
> People have a right to support candidates > they can print a sign and stand on a street corner > printing signs costs money > people have a right to spend money supporting candidates.
Individuals printing signage is not where most of the funding is going though, is it?
> People have a right to assemble > they can stand on a street corner with their friends > they can hold a big sign together > they can have lots of friends and form a group and print a really big sign.
This isn't how most of the money is being spent either, is it?
I don't know how you seem to either believe the point of the right to free assembly was to give richer corporations more sway, or be completely oblivious to the fact that that is what's actually getting people angry, not the fact that citizens want to stand with signs by the street corner. Assuming you don't have a conflict of interest leading you to that conclusion, I... don't know what to tell you.
I think your objection is fairly clear in terms of 1) you point out the difference between e.g. standing with a sign vs. more expensive options, and 2) you say "right to free assembly was to give richer corporations more sway."
Fundamentally it sounds like your complaint is that wealthy people are presumably more able to be effective with their efforts than folks who, e.g. can't afford as big of a bullhorn.
Citizens United wasn't about that. There wasn't a law saying no person shall spend more than $1k/yr supporting a candidate.
This is a completely inhuman and unamerican take. Money buys secrecy, it always has. Money is a tool, exercising speech is a right. For the Supreme Court to equate them is ahistorical, plutocratic, and terrible. You sound like you support giving corporations the right to vote as well,
Where do you draw the line between money and speech? IE, would you "simply" ban all political advertising completely - one can scream from a soapbox and that's it?
What is a corporation but people? Companies are groups of people. They're one or more owners, some managers, possibly some staff. If you take away all the people, then there's nothing there.
Assets, mind share, etc. All the things that get sold off in bankruptcy proceedings.
A railroad company also has a huge amount of steel roads with cross country right of ways. You could fire everyone, and it'd still be extremely valuable. It might even be more valuable.
Even in asset-lite software land, "Atari" the name still has a lot of value, even though none of the people are there anymore. They own the right to control the things that the neurons in millions or billions of people's brains are tied to, that their habits tie them to.
Twitter, you can fire almost everyone, make the online square less pleasant for a lot of the people, the network effects truck on. People continue to use it even as they hate using it, because most everyone else is.
Democracy is built on the idea that people are equal and have equal rights. This is only sustainable if the power differential between people isn't too large.
Corporations consist of people, yes, but the power to control where the efforts of these people are directed is concentrated on very few people, giving those people a large amount of power -- so large that without any checks on their influence on government, it undermines democracy.
It's the same argument for why super-wealthy individuals are so problematic. They're literally destroying the fabric of democracy through the power imbalance afforded by their wealth.
A corporation is a liability and pseudonym shield.
A corporation having contribution rights multiplies the otherwise imposed contribution limits on the underlying humans.
A corporation allows tax planning opportunities not afforded to the bottom 90-ish percent of the country (such as shifting income to alternate years to avoid progressive taxation, and utilizing expenses to lower that income in ways a W2 cannot)
Corporations are granted benefits that people do not get, and CU disturbs the balance of power.
Legally, the corporation itself is similar to a person and has rights unto itself, as stupid as that sounds. Those rights are different from those the individuals who run the company have.
> People complain about the wrong thing. The problem isn't that people are allowed to support candidates with their money, buy airtime, etc.. It's that companies have secret owners...
This was a major crux of Colbert’s “our lady of perpetual exemption” bit. People absolutely know about the shadowy machinations CU enabled. The lack of transparency is a major critique.
I wouldn't say CU enabled it. CU didn't address the secrecy of corporate forms. That wasn't at issue. CU was only about ~"Can people use money with their friends to support a candidate."
To say that repealing CU is the solution is to allow the critique to go unresolved, while arguing ~"People shouldn't be free to support political candidates."
We have the techno-feudalism and dystopia but none of the cool aesthetics from cyber punk or arguably even the real Zaibatsu. Nvidia isn’t cool, there’s no interesting aesthetics or back story, it doesn’t appear to have zaibatsu ambitions, it’s just riding a meme stock bubble.
In a different world I can imagine SpaceX to be the only thing that could come close, but well...
Nvidia is hedging their bets against becoming too dependent on OpenAI, Anthropic, and the hyperscalers as their only customers. If on-prem deployments of open models starts taking off in corporate America, this acquisition positions them well.
Nvidia really is best positioned to benefit from the AI wave. They are the main provider for GPUs to the cloud-based LLM companies, and their GPUs are still preferred for local models as well.
> This acquisition and the Stripe acquisition of OpenRouter are puzzling. Both HF and OR, from a technological standpoint, don't appear to have much IP that cannot be easily replicated.
I was puzzled by the comments on the openrouter acquisition saying "they're just a proxy" - they're an open marketplace where models compete head-to-head, they're quite a bit more than just a proxy.
Hugging Face has essentially become github for model repositories and I think nvidia wanted to snag that before microsoft does.
They run a proxy for LLM providers, which inevitably creates a marketplace. They’re amazon.com for tokens.
The flip side that others are pointing out is that Amazon is sticky because there’s a sprawling, physical logistics apparatus underneath. You can’t replicate amazon.com’s business without billions of dollars and a decade of building warehouses.
I don’t see where OpenRouter has that. As far as software goes, theirs doesn’t seem particularly tricky. LiteLLM does vaguely similar things, at least for organizations where managing accounts isn’t absurd overhead.
They were first, and there’s money to be made there, but what’s stopping someone else from building LibreRouter that charges a 3% or 4% fee instead of 5%? That’s where I get dubious of their valuation.
... and there is some of that two-sided market effect that (1) people who want to publish a model will publish on HuggingFace and (2) people who look for models will look at HuggingFace. Even if HuggingFace is poorly run in the future it will be hard to dislodge.
Just because a huge company could replicate an OR or HF replacement, doesn't mean by any measure that users will abandon OR/HF and flock to the mega-corp equivalent. In fact even the opposite-- users might be turned off and reticent to even try the mega-corp's..
Nvidia has a market cap of $5T, and so far today the stock is up almost 10%, so that's another $500B. Paying $13B for HF is chump change to them.
NVidia have been making lots of moves recently towards supporting open weight models and local AI, and it makes sense for them to buy HF who support the same, if for no other reason than to prevent someone else like Musk buying them just to shut them down, although I assume they have plans to do a lot more than just keep HF alive.
I think a close analogy is Github. The base infrastructure is open source and replicable, and still there's a virtuous cycle that keeps people choosing it over the alternatives.
Github definitely extended git to a far less technical audience than basic git came from. Emailing patches or connecting to others' remote hosts to pull changes was too much complexity for many customers.
Pull requests, CI platform, authentication that is not PAM etc.
OpenAI recently committed multiple felonies against HuggingFace.
Nvidia is heavily invested in OpenAI, recently guaranteeing a datacenter buildout up to $105 billion dollars.
Any reasonable criminal investigation into OpenAI's actions or negligence into this incident would hurt their reputation, their ability to raise money, and therefore Nvidia, who is invested in them.
If I was a founder of HuggingFace, I am immediately getting in contact with Elon Musk and SpaceXAI, and letting him know how that he can purchase them and sue OpenAI, this time with solid claims.
If I was particularly amoral about if I think OpenAI or Anthropic should get to AGI first, I would take this bid to Nvidia, whose entire stock price is propped up by OpenAI's over-purchasing of compute.
First mover advantage. The first big company with the most customers stays the de-facto leader of that industry/product. Later OpenRouter and HF will start building moats to keep their customers from easily leaving. That's why everyone's still on GitHub despite it blowing chunks every week. Too painful to move.
Don't see this happening. Both Nemo and TensorRT from Nvidia are heavily invested in low precision formats and higher inference throughput as well as low memory use. It's more about what happens in relation to GGUF, MLX and non CUDA-native frameworks/formats.
I really, really doubt that, and I'm a bit surprised (not too surprised, HN has gotten really cynical) that this is the top comment. I think the future nvidia most doesn't want is a small number of closed labs that run away with it, and gain power to steer a large part of the hardware spending towards nvidia's competitors, as leverage in pricing negotiations for big hardware buys. Their ideal would seem to be lots of competition in the model space, driving lots of innovation in lots of areas, at low margins for the model makers, driving use and demand for hardware way up. They want the portion of the economy working on this to go up, and that's not going to happen if there's just a few companies that can afford to build models.
I see this as NVidia placing 50% of their bets on local models as the future, which it seems may be more profitable to them than cloud (sell 100/whatever local cards, which see 40hr/week max utilization, vs one cloud card seeing 24x7 utilization).
NVidia seem like a smart company - if they want to promote this direction they are not going to cripple it by trying to limit quantized models, even though you might expect them to promote benchmarks showing the benefits of larger and less quantized ones.
It's like the smart Intel of old - support CPUs all all price points, while also promoting CPU hogging applications like OpenCV.
VC funding (much of it from big tech). They have a freemium model and offer some pay for enterprise features, really just private repos, permissioning, etc. but not sure if it is meaningful
The history of acquisitions mothballing the acquired assets or moving them off their original mission or letting them atrophy is too stark to support your optimism. Especially when nvidia has a direct incentive to to steer the ecosystem. For example are they going to highlight and surface alternative chip designs like grok? Or will model search mysteriously not find related models?
Nvidia doesn't exactly have Oracle's reputation when it comes to acquiring companies, but given its overall hostility towards competition and open source I can't blame people for being anxious about the news.
The most important resource for self-hosting LLMs is now under the control of a company that has very markedly kept the specialized hardware outside of the broader public's hands.
You're probably right. It was mostly hyperbole, but the older I get the more I think hyperbole does eventually come to pass in situations like this. It's gradual, though, not all at once.
Natural log cynicism and any cynicism about public corporations in our late-stage capitalism world is likely to look tame a year later. Or possibly too.
Ex-NVIDIA employee here who worked on inference software. When it comes to inference software and models, NVIDIA's strategy is advancing the state of the art in the open model realm. They are building software and services business for those markets that need to be competitive on their own merits, and quantization and precision tuning are techniques they use themselves constantly.
Also very silly prediction given this would dramatically lower the amount of models hosted on HuggingFace and move users elsewhere to platforms that do support quantized models (not what Nvidia wants).
Yeah. I want to be optimistic, but this feels a lot like how there used to be multiple video hosting sites and now there's just YouTube (and its copyright filters).
Unlikely, but say goodbye to uncensored and abliterated models.
These acquisitions are not healthy. They stifle competition and make it easier for governments to censor open weight models by choking off distribution points.
This is the world we live in now. 3-4 megacorps owning everything in every major product category - eyewear, cosmetic, consumer goods, media, tech; you name it.
We need small companies worldwide, developing the things huggingface does. The us china and Europe dominate while the rest of the world sits by idly. Torrent and seed ai models. It's not piracy
Has always been this way. Big companies always seek to buy fast growing startups and unfortunately the founders of these startups mostly take the money.
Potentially horrible for monopoly reasons, but if other big acquisitions in the AI era teach us anything, developers are about to get a whole lot of free and discounted trial credits.
That’s at least a plus. I will happily burn through as much VC money as they will give me to tinker with my projects.
Current US government is pay to play. As in buy my cryptocurrency and you get a pardon or case dismissal, no backroom deal required. Payment directly. Blatant corruption sucks.
If the House or Senate turn that might actually stop or slow down. The market falling by 30% when the AI bubble eventually pops would also trigger it.
Now that you mention sleeping. I imagine one could do cool sleeping startups that clone something but exist only on paper. The sleeper startup is only launched the moment some ham fisted mega corp aquires the original and all the customers are looking for a way to abandon ship.
I think failing to take advantage of the charisma of both Lina Khan and Janet Yellen, and not making them public-facing economics gurus like Alan Greenspan or Larry Summers historically were, was a huge tactical mistake by Biden.
Fair, but the biggest mistake by far was failing to take charge of the immigration debate. The second biggest was his press secretary's insistence on announcing that everything was fine in the economy over and over as a distant second. Third was not resigning when it became clear that he couldn't really keep up with the job, or at least not trying to run for a second term. Virtually nothing else matters by comparison.
Fear over economic damage due to immigration is perhaps the last remaining unifying force among the American political right wing. Biden actually did a lot of good things as president -- arguably a lot more good than he did in his time as senator. But his administration was too eager to just run the federal government as a quiet meritocracy and avoid dealing with any of the hot political issues of the day. He and his administration fucked up so bad that now not only is most of that good undone, but we are in a dire existential situation, facing generational-scale damage that might be unfixable.
It's one thing that they genuinely fumbled the handling of the huge immigration/refugee waves spanning the entire continent and it took them well over a year to start coordinating a response (which they never really did apart from eventually just closing the border) -- it's another that they kept trying to insist everything was fine, there was no problem, there's nothing to see here. People were freaking out about it all over the country, and of course they were being heavily propagandized, but that's not the point. A strong and proactive White House would have made all the propaganda look stupid. Instead, it looked like the truth.
It's absolutely mind-boggling to look back on just how weak the president had to have been from 2020-2024, to somehow let the party of Jan 6 and the disastrous handling of the Covid-19 pandemic regain enough control of the political narrative to make Trump seem like anything other than a narcissistic mobster.
Yellen was objectively the dumbest Treasury Secretary ever. When everyone and their dog realized inflation was getting out of control, she thought it was transitory. She literally gets an F- as an economist.
And some former Roomba employees may have thoughts about Kahn.
Kahn was right about Roomba. Amazon was going to torpedo all the superior competition on the Amazon platform where almost all robot vacuums are sold in the US, plus as the e-commerce monopoly, Amazon was going to instantly have access to an army of cameras inside people’s homes to analyze personal information related to purchase preferences.
IMO allowing Roomba to hit chapter 11 was still a better option for the consumer than handing them to Amazon. They’re still in business as an independent competitor on the market, consolidation was successfully avoided.
Yellen, I don’t have much of an opinion on, but she absolutely wasn’t alone in that opinion (the nuance of that opinion being inflated by this hyperbole), and the treasury is a lot less involved than the federal reserve in doing anything about inflation, anyway.
This is a tangent to my first comment parent to this one, but also more related to the actual article at hand: whether it's Amazon buying Roomba or Nvidia buying Hugging Face, I have a somewhat radical (or is it?) opinion that large companies like Nvidia, Amazon, Apple, Microsoft, Coca-Cola, etc, should not actually be allowed to acquire companies. At all. under any circumstances, even if competition is healthy.
These companies are generally large enough that they do not need the competitive aid of buying an existing company and starting with that sort of structural advantage.
E.g., did Nvidia not have enough money in their bank to start a company to compete with Hugging Face? This is a company that reportedly has ~200-300 employees with investment rounds totaling $400 million. Nvidia made $31.9 billion in net income last quarter.
I look at a company like Xiaomi which just developed its automotive division in-house without resorting to buying car companies. I think our traditional business and finance mindset has an overreliance on acquisitions.
Roomba was not competitive in the market, no amount of Amazon market manipulation would have changed that. Best case scenario, Amazon would have invested in making it competitive. Worst case, their engineers would have been absorbed into Amazon's warehouse robotics projects.
Is either of those worse than what actually happened: the company is now a zombie brand for a Chinese company?
Of course Amazon would have changed that. You would go on Amazon and search for "robot vacuum" and Amazon would put iRobot at the top of the results. Review manipulation on the platform would be trivial.
Amazon could email/push notification/text customers asking for reviews of iRobot vacuums but then not do the same for competing vacuums, skewing their reviews higher (asking for reviews boosts ratings by gathering opinions from happy customers who usually don't bother writing a review). Competing brands potentially don't even have your contact information to ask for a review.
Go on Amazon right now and search for "usb cable." There's a giant banner at the top that recommends the Amazon Basics brand, three across horizontally, which on my desktop monitor takes up nearly 50% of the screen real estate. Then below it are the Anker cables that you're more likely to be looking for.
They would have almost certainly been manipulating pricing on them as well. For example, they could take the strategy of lowering the price of the vacuums to break even or sell as a loss leader, but use them as a data-gathering robot in your house to help Amazon sell more of everything else. They could make their Alexa smart home platform preferential to iRobot or lock out competing models.
Robot vacuums are a consumer goods category that is heavily skewed toward Amazon.com as the place of purchase compared to other retailers.
I think "zombie brand owned by a Chinese company" is actually preferable, yes. They still operate and sell vacuums competing with the other robot vacuums on the market, and they aren't in service as household data collection bots for a monopoly e-commerce platform.
I don't think the ownership of the company being foreign or domestic is very relevant to the FTC's goal of preserving positive trade conditions. Would we think the iRobot situation was a bad outcome if iRobot was purchased by a foreign company we view more positively like Miele? We only think of it negatively due to anti-Chinese bias. iRobot being Chinese-owned is almost certainly the best possible outcome for preserving the amount of competition in the market.
Yes, that's a very good company to be the owner of iRobot, because they are just a robot factory and not a near-monopoly e-commerce retailer.
They are essentially on equal footing with other robotic vacuum manufacturers. iRobot didn't go out of business or get absorbed into a larger company lowering competition in the marketplace.
It's also not the FTC's job to ensure that companies, especially ones with zero/trivial national security or domestic labor force value, remain under domestic ownership. Amazon itself is not really a "domestic" company, it's publicly traded. Anyone from any non-sanctioned country can buy shares of Amazon.
I spent a little time working there around 2017. Whatever you're implying my former colleagues' thoughts on the subject are (or mine, for that matter), you're probably wrong.
True, but better nvidia than any other big tech contender. Nvidia has a very strong business incentive to make huggingface thrive, whereas pretty much everyone else has the opposite incentive.
The Government plans to regulate open source ai. Nvidea is dependent on the Government to back loans, provide property, and make laws in their favor. Now they have ownership of what the Government wants to control. A perfect wedge they can use to get more from the Government.
Why would China want to crash the AI industry? Their goal is not to damage the United States, necessarily. Their goal is to have the best domestic AI on the planet. They intend to do that through the way that they dominate every other industry: good enough quality at a much greater scale.
Surely crashing the Ai industry (in terms of reducing commercial revenue) makes it harder for the leaders to keep raising money and so keep investing in r&d to the same levels. Thus allowing competitors without access to latest American hardware to keep up.
I think this is good.. They have incentives to keep things free to keep people using their GPUs. I think that's one of the least enshittifying outcomes possible
HF is the default platform to find models. Not just ones published by big labs but also a lot of the distilled or fine-tuned versions, etc.
If Nvidia buying HF makes it tough for all the diverse models on HF, then what are some alternatives?
It seems models are the best things to be available on a Torrent platform? Of course HF is much more than just the files but perhaps the metadata can be separate and hosted on multiple community platforms.
I'd honestly prefer that. I'm usually on 5g if not 4g and pulling models is an absolute nightmare when I basically have to use HF and even with a token getting throttled. Although I have a pretty unique use case. Pulling wads of 5gb tensor files over cellular is painful.
I am on the same boat, I live in a small Himalayan village. I have 2x5G based devices and one Wireless bridge (Ubiquiti LiteBeam M5) for a local broadband. Generally I get 50 Mbps, sometimes up to 100 Mbps
i'm traveling in an RV and can always (80% of the time) get up to 3MBps, max to 300. But with cellular it's mega spotty so pulling 3-10 5GB files then losing your connection for a few seconds/minutes completely kills that files download. I've tried about 10 different keep-alive-ing type hf downloader clients and with my 64gb ram I'm usually downloading 25-32GB models and I don't even try unless I'm in sitting in a city next to a library anymore.
It's less of a pain now that I'm not in my LLM newbie phase of trying to test out 30 different models and stick with Qwen, but I couldn't even try to download qwen 3.8 for a week or two until I got back to the city.
Starlink is the obvious option and I have it on my roof but I do everything I can to not pay that $170 a month vs my $170/quarter 5g service.
I am quite skeptical that an acquisition by Nvidia will make people move somewhere else ... The network effect remains central, if you are researcher/lab/company and you want to promote your model/dataset you go to the popular platform where most of the community is. The same way most of the public repo are still being published on GitHub despite people being pissed of by Microsoft acquisition. (I you add on top of that the cost changing your workflow, not using HF library, ... it make the transition even more difficult)
I don't know anyone who "finds" models on HF. The announcements always come through Reddit, HN, Xitter, and various Discords. HF is just a place where some people bought a domain name, spun up a file server, and struck it rich. If they went away tomorrow, they'd be replaced the next day.
NVIDIA buying HF feels like defensive move. They don't want open models become hardware agnostic. I bet we see more 'optimized for CUDA' push in one year.
I do think it's a defensive move, but at first order it's defensive against OpenAI, Anthropic, and arguably Google. NVIDIA doesn't want to end up squeezed by those three as they increasingly make their own metal, etc.
Ensuring a healthy open-weight model market means ensuring continued demand for metal running in corporate data centers, bypassing any middleman. NVIDIA has a strong interest in a thriving "run your own agents" market.
They're actually aligned with Apple here. And in this market, Apple is a real competitor too, with Apple probably more consumer-centric, while NVIDIA probably more enterprise-centric.
Dgx Spark and Strix Halo have very close specs and deliver similar performance. If nvidia makes their stack be more efficient with for example 50% more tockens on similar hardware specs agaist competitors, they don't need HF.
My take is that Chinese Labs, though slightly behind on the frontier (due to compute constraints) are on a trajectory to surpass Western labs (this is me speculating, reasons are better ecosystem creation on China's part and potentially better/more data environment). Qwen-3.5-122b was the king in it's category and noone came up with something better, even though many tried like poolside with laguna. Similar with 35b and 27b param models. I think poolside and HF acquisitions show us that nvidia really wants to have competitive models on the prosumer (~100-150b param size) and likely at the 300-500b as well. Together with a hardware to run them that's a good market to be in. And as the recently rumored Xiaomi AI cube shows us (together with gorgon/medusa halo and mac studios), this is a market segment that will have competition.
That could be a factor, but the optimistic interpretation would be that they want to support the open ecosystem because it sells more chips.
Models are already largely hardware agnostic. It would be pretty hard to put that cat back in the bag.
I could imagine them building value-added services on top of HF to advantage Nvidia products (i.e. "run this model on NVIDIA cloud" with one-click), but in this moment it's hard to imagine how they could actively disadvantage models built to run on other platforms.
That huge impromptu 2023 party at the Exploratorium Hugging Face did had such amazing covid-unthawing AI spring energy, and I keep thinking about it. It was clear then they had the momentum to do a lot of things, and more so now. So I was surprised by this news.
The party is a is a relatively small concern compared to the openness and github comparisons, but since no one else mentioned it, I hope something like it happens again. Or maybe that moment has passed.
Nvidia's been pretty terrible for open source / free software. No need to quote Linus Torvalds here. They want to control what runs on their hardware. They want to you write code against their proprietary drivers and APIs, not directly against the hardware (which these days of course also contains plenty of software, but still).
Don't expect things to go differently this time around. Nvidia wants control over the software stack. Acquiring HF fits in perfectly. The play is long term.
Yes, there is no question NVIDIA wants to lock you into CUDA and their hardware. But also, they’ve consistently demonstrated the most openness when it comes to model training, datasets, and research; even before the LLM era (e.g. StyleGAN).
There’s also modelscope.cn (china’s huggingface) which is worth checking out. I would not be surprised if one day, we have to use China VPNs to download open weight models.
I'll expand on the economics somewhat. My intuition is that you can like a moat left of you in the supply chain, but dislike moats to the right of you. (I'm using left and right as I picture this horizontally drawn. It is usually called vertical integration by economists.) But free competition in your market is worst of all. Free competition right outside your moat is pretty sweet, and that's were parent is commenting on.
Nvidia probably likes that ASML is a monopolist (it's called a monopsony). The price is high and Nvidia can't scale as hard as they want to (more general: capital intensive market). This monopsony makes sure that other chip companies can't rapidly scale up and try to beat Nvidia. That there only ever was one other GPU firm (I'm prehistoric; once there were more) and they bungled it on software, is pretty sweet for Nvidia.
On the side of their customers. It would be best for Nvidia if there is free competition for the outputs generated from there GPUs. This maximizes consumer surplus and thus demand. Maximum demand for tokens, is maximal demand feeded in their monopoly. If there is a monopoly right from you, demand is curtailed, and your value is limited.
And that exactly is why you see interest from token generators for chips. Bridge that moat and gain a larger value surplus. Both NVidia and, say, China actively undercutting the token-supplier value chain is quite interesting to watch. It's like the Opium wars with us as somewhat happy customers.
In this same vein, why isn't ASML raising thousands of billions for building their own (subsidiary) chip foundries, while raising prices and starving the market (a little) for their machines.
> In this same vein, why isn't ASML raising thousands of billions for building their own (subsidiary) chip foundries, while raising prices and starving the market (a little) for their machines.
You're looking this purely through an economic lens, while in reality geopolitical factors play a huge role in what ASML can and cannot do. The US government would likely take an extremely dim view of any new external competitor popping up for their chip foundry industry (especially with all the new US plants being built or planned) and would lean heavily on their vassal/ally the Netherlands to prevent this. Unlike with China, the US has more leverage over the Netherlands[1]
The US security state and US tech giants are joined at the hip, as they have been since the beginning of Silicon Valley[1], right through the Snowden revelations through to the present day[2].
There are - see the RTX Pro 6000, which has 96 GB.
There are a few problems though, primarily, a GPU with lots of VRAM and very high bandwidth is inherently very expensive (on top of which there is also the CUDA premium); AI use cases are better served by SoCs with lower (but still high) bandwidth and more RAM.
This is not a GPU, this is a laptop SoC with CPU and integrated GPU, and knowing nvidia it will probably be even more closed than an intel CPU. You will own even less of your hardware
While nvidia 's drivers are closed source there's enough interest in running LLM's that you are not locked in using nvidia. Strix Halo chips have been around before dgx spark came out and deliver very similar performance.
When last looked at it, NVIDIA was not supporting OpenCL beyond 1.0.
Also, when running OpenCL, NVIDIA hardware disables multiple DMA engines, and allows only one memory transfer at a time to prevent OpenCL running as fast as CUDA.
Did NVIDIA finally allow open source drivers to access all parts and features of the card to allow feature parity? Last time I checked they were considering a plan for planning a solution to that.
I love your insights which enlighten my blind spots most of the time, but I didn't blame NVIDIA for OpenCL's failure. I just noted a company's choices when it comes to a competing set of libraries w.r.t. to their native ones.
Having said that, I'll try compiling a OpenCL 3.0 program in the cluster, so I can report whether NVIDIA runs this software, and if yes, how well.
Yeah, however if Intel and AMD actually delivered a working 2.0 with proper support for C++ and Fortran, maybe the OpenCL 3.0 back to 1.0 reboot would not have been needed.
Likewise SYSCL although built on top of OpenCL 3.0 primitives, is mostly Intel, which also owns CodePlay, the company that delivered the first working SYSCL compute experience, again neither AMD nor Intel (until it bought CodePlay).
I don't think bits like the GSP firmware will ever be open-sourced, but the opportunity to write better OpenCL drivers has always existed. Some of Nvidia's other proprietary driver backends (eg. GBM, Vulkan) are also decently neglected, but mostly out of disuse rather than malice. I don't think any of these things mean you don't own the hardware.
What is the support like for them? Can they be seriously considered as alternatives to Nvidia or AMD cards? I have been looking into buying a GPU, specifically the Radeon R9700 AI Pro but noticed the Intel cards too and was not sure what you make of them.
A few years ago everyone said _Open_AI is the the most open labs. How did that turn out? Lots of coy, deceiving actions till the whole company was turned into whatever rent seeking amoral borg adjacent shell of it's former past it is now.
Because making new AI models is a research activity, so the group within Nvidia (or other companies) that does that activity is called a research lab, or just "lab" for short. I don't think anyone is saying that all of Nvidia is a lab (that's just shorthand I guess).
What amuses me is that “lab” is shorthand for “laboratory” and virtually no computer research happens inside a room people would typically call a lab.
It’s a little like the trend of calling developers “engineers”. There’s no actual engineering in the traditional sense but I’m sure developers think it sounds cool to call themselves that.
That’s not what I said and the snarky remark about my experience is unwarranted (I have more experience in “software engineering” than many on here have been alive)
My point was just that I just find it amusing that people call themselves engineers when the code they produce is so far removed from the level of rigour one would expect in literally any other engineering industry.
I say this as someone who also has family and friends who are actual engineers, if they built bridges and buildings to the same standards that many developers write code, then people would die.
This isn’t meant as a criticism of developers, by the way. Just an observation at the vast differences in the domains and thus the tolerance for errors in the process.
Likewise for labs. I’ve worked in AI startups and the science departments are not something one would think of when you say “laboratory”. I get why the term is used, but it’s still amusing.
I know language isn’t static. It’s something that evolves, like how a “computer” used to refer to a person rather than a thing, but that doesn’t stop me from being amused. But maybe the real issue here is I take myself less seriously than others so I have that capacity to be amused by the titles I’ve held?
Because they are mostly research departments (aka a lab) turned into a corp structure. It’s not a new phenomenon, just that until recently labs didn’t get $1T valuations so you see it more now
"The proprietary shovel seller has some excellent tutorials on how to dig gold. Nobody else has such good step by step guides. Therefore them buying a shovel-agnostic tutorials and techniques method (that has a lot of info on using other shovels effectively) is justified."
Well if you find that obvious (I also do) then why do you not find their motives in acquiring HF onvious or that this move would be bad for the ecosystem as a whole? Cause it's all kinda the same thing.
Their (not especially great, compared to Chinese ones) models being open weight doesn't even come close to outweigh the effect of CUDA & Co being proprietary and closed.
CUDA is proprietary for a pretty understandable reason. There's no good way for Nvidia to standardize it.
They supported OpenCL when Khronos floated the idea of a GPGPU standard to manufacturers, but OEMs didn't want to design scalable hardware or sponsor the software.
Evidently we should, because Linus has been more positive about Nvidia in the last 2 years [0]. I've been using the open driver for years now, for both gaming and CUDA.
He definitely sounds more pragmatic than before, which is, in a sense, more positive.
> This is actually one of the benefits brought by AI; it has made Nvidia a good participant in the Linux kernel space. [...] Now, when Linux is so important for AI clouds, Nvidia suddenly cares very much about Linux.
Nvidia releases some of the most open open weights models, Nemotron 3, which have the full training code open, and most but not all of the training datasets.
Nvidia is a big company. They are good about some things and bad about others.
I think they really do like open weights because they make some of the best hardware for training, and the more open weights models there are, the more people are training and fine-tuning them, mostly on Nvidia hardware.
I feel like Nvidia is one of the better choices for buying Huggingface. Not perfect, but definitely far from the worst.
The worry is not that Nvidia isn’t open about their models.
The worry is that Nvidia is trying to control the way you run those models. Trying to bake CUDA assumptions into model design, and pushing the software ecosystem to be as Nvidia first as they can.
> Nvidia's been pretty terrible for open source / free software.
Their strategy for AI is pretty clear and they bank on on-premises OSS models for the busines, with their really cool open-source software:
https://www.youtube.com/watch?v=tmcn1-jFLWY
This is a really good watch and is a glimpse into what the actual future shapes up to be considering the current situation in where the OSS Chinese models successfully compete with proprietary US ones.
As a AMD GPU user, I feel as if trying to mandate CUDA only or some other way to increase NVIDIA lock in like that would be met with people leaving the platform or at lest simply working on the non NVIDIA formats away from hugging face. Though requiring a lot of space it's quite easy to spin up a competitor at least for hosting weights. And if they try some legal shenanigans then many countries outside the US will still be happy to host I am sure why not China?
The only thing I can see them being able to get away with is increasingly bending the hugging face python API and any other features of that sort they develop to NVIDIA only. I personally don't use that and don't see a reason to and I am not sure how many people do use the hugging face python library.
Hf’s python / transformers is a hot mess. It conflates so many different ideas into huge monoliths that it’s barely usable for anything other than the small snippets listed for trying out models.
They need to take a machete to all the cross coupling they’ve metastasized.
Arguably, this move will help AI open source market more. Nvidia has an incentive to make AI open source competitive with closed source. If OpenAI or Anthropic gains in market share, they will eventually gain power over hardware vendors as well, which Nvidia does not want.
Nvidia wants people and companies to go choose a free open model, run that model on Nvidia hardware. And because Nvidia can't fully control what hardware an AI model can run, Apple Silicon and AMD hardware users will benefit as well.
Hey. I worked there, and I'm no longer bound by their PR policy, so I can mention this. Also, know that the fact that I worked there doesn't make me privy to their high-level strategic decisions, this is a more look from the trenches kind of thing.
My impression I've developed in the years of working there is that Nvidia's relationship with opensource is... not intentional. They kinda suck at it because they genuinely don't know how to do it more than they want to make money out of it.
Here's an anecdotal "success story" which is also an illustration to how things might not work out well otherwise.
So, I was on the team that deals with server infrastructure. One day we get a new "feature" which was supposed to allow Slurm (the workload manager, a kind of software used to run "jobs", including eg. model training) to be deployed with distributed MySQL as a backend. The feature is all obviously written by a single developer with an enormous amount of "help" from AI. I was tasked with testing it.
Trying to figure out what it does... I realized that the "distributed" part of the feature was to be achieved by integrating with Oracle's MySQL by means of using MySQLShell (another proprietary Oracle's product). Until that point, by default, we integrated with MariaDB. Not only was it using Oracle's proprietary tool, the tool, actually, didn't support the "distributed" part of the "solution". It was pitched as the "first step on the way there".
So, I was able to push back on it, mentioning Galera, arguing that the "solution" doesn't solve the problem and will require from customers to change databases (even if they are mostly compatible... they never quite 100% compatible). And the misfeature was rolled back.
I made an effort to investigate how did we even get there, and turned out that whoever authored the "solution" had an experience of working with Oracle products, but never really tried the open-source ones. So, he didn't do a research. He just used what he knew.
Unfortunately, this is a rare win, where the evidence of disadvantages of using proprietary solution was huge and enough to turn the tide. But often it doesn't face any resistance because nobody is even aware of the problem.
If that's true, it may bode well for the future. The AI boom creates some strong incentives for Nvidia to develop an intentional strategy for software generally and FOSS specifically, and there's a chance that could influence the rest of the company in a very positive direction. Hopefully, what they learn by working with HF and having the value of their hardware products increasingly determined by their ability to run FOSS software stacks will backwash into the rest of the company, or at least make them much more aware of the problem at all levels.
I do think that they still might be a bit reticent to release fully open-source drivers, though, only because of the extent to which hardware design could be inferred from the driver source. OTOH, AI itself is making disassembling and reverse engineering binary code easier and easier, so there might not be much of a point to withholding the source in the near future.
I think this is part of an open-source play. I'm not arguing your other points, I think they're true.
They're trying to mix up the competitive landscape(that doesn't impact their bottom line, and I don't think opensource is eating their lunch), so I don't think this is fake, at least that's my initial take.
> as a force to have open weight models run better on Nvidia against the rest.
this is the crux - if nvidia makes it so that open weights end up running better on nvidia hardware than competitor's, then it's going to prevent hardware innovation and competitiveness in the entire sector.
It's like as tho General Motors buys out oil refinery to make gas for all, but the gas somehow runs smoother in GM cars.
If you don't own the customer relationship, you don't own the profit. See App store, FB.
NVidia is delivering commodity hardware to the hyperscalers, and would eventually get commodity margins (when hyperscalers make their models work on their own hardware).
Amazing article on relevant economics of squeezing vendors - actually about antitrust ad-models but:
The answer is that [advertising based] companies are an ideal test case for an increasingly common business meta-model, where companies try to create a consumer surplus at one end in order to maximize their negotiating leverage for capturing the producer surplus everywhere else in their supply chain.
It's a clever move if that's what they're doing. They're restricted in China, and are likely to face stiff competition from Chinese chipmakers in the coming years. Acquiring the largest repository of trainable models and ensuring they run better on Nvidia hardware is probably one of the few moves they have for keeping ahead of the competition. I mean, it would be terrible for the consumer, but it does make me think that NVDA is a decent investment.
Well now that they own it, they can do whatever they like. They don't necessarily have to be optimizing the models for their hardware, they could just setup preferential pipelines to funnel users to their ecosystem. That seems fairly prudent given the challenges coming their way.
We can stop with these weak excuses since AMD and Intel have done more for open source than Nvidia has, including their GPU drivers for Linux.
Nvidia on the other hand has not and the best they have done is a bunch of closed-source blobs which they do more closed source releases than the rest.
Mojo is open source and targets all GPU architectures for their compiler regardless of the vendor and nvcc targets their own (and both that and CUDA are closed source).
So this is directly an apples to apples comparison.
I'll give credit to Intel, they've always been open source friendly and I prefer to buy them because I'm not a gamer, so don't need the best.
AMD is another story. Their binary closed source blobs were even worse than nvidias for many years. So much so people used terrible performance open source tries just to avoid the headache. That they finally slopped together an open source version after they'd lost the GPU race isn't exactly noble. But as an open source supporter in general, I commend the effort still.
I care, I just don't believe you in that statement. There were "confidential nda" notes embedded in the release that Chris didn't remove until afterwards. That doesn't scream: "we're going to open source this some day!"
If it was going to "eventually" be made open source, why not do it from the start? Why not do it a year in?
From what I was told, your speculation is incorrect. I can't say anything more than that, sorry.
I don't know why you're so aggressive in tone towards me, chill out.
I am not sure you need to posit nefarious intent for this to be a bad deal for the industry at large. Basically the only provider of GPUs getting into vertical integration is a classic monopoly move and I hope regulators look very hard at this.
"That's not how citations work." (Dude on the interwebz, 2026)
But more seriously, this is my first time seeing that as well, and I'm not sure I like it. Citing an LLM is a little like citing Wikipedia to me, you cite the primary source the LLM is quoting directly, not the secondary source.
Far worse from my perspective is that it's not a fixed citation that it's not falsifiable. It's just a mutable, generalized source, which is stochastic in nature.
So much of present day feels like being back in middle school around the turn of the millenium, with classmates (and even some teachers!) new to the internet insisting that "source: internet" or "source: google" was totally fine.
It’s the formatting and formality that bother me. Say “I looked it up with Gemini,” don’t give me some weak attempt at “proper citation” to legitimize the bare minimum effort you put in.
They’re buying a brand, some employees, and some momentum — not any of their software. HF probably does have propriety goodies to make it all run efficiently, but certainly not a billion dollars worth, much less 13!
We are so used to these numbers being thrown around in the AI era that something one needs a reminder that this is 13 billion, not million. Insane exit by HF.
Was looking for this comment, the tech world has gotten completely insane with ”valuations”. I would love to hear why it was 13 and not 5. It would still be completely insane at 5 billion, but someone though they should add another 8…
Opportunity cost for Nvidia to prevent HF from going public or being sold to someone else. It's an insight into Nvidia's market prediction and what HF told them they'd do if left alone.
No company is going to be acquired for their software alone with multi-billion valuations ever again. It should be clear by now that code alone is not where the value is anymore, if it ever was.
Now what will happen to the llama.cpp performance on non Nvidia GPUs and especially on arm64 CPU chips after this acquisition? Will Apple M* chips and cloud arm64 chips (Graviton, Cobalt, Axion) become second class?
Just now I got my hands wet with local LLMs before this was announced. Can somebody from llama.cpp/ggml.ai confirm this?
We are witnessing a set of very disrupting in the industry and it is very hard to say how things will look like. AI companies buying runtimes, frameworks, editors, model routers, model hosting providers, teams behind databases... It is hard to find anyone able to reject the insane amount of money.
Very optimistic to say this will not have profound effects on the whole industry.
Assuming this deal goes through, I do wonder what this means for the future of open source models. Like it's very possible the next deepseek model host it's weight on it's site first, vs treating HF as a canonical hosting site. Just my guess, but regardless we'll have to see how labs (and even countries) respond to a less neutral HF.
I remember when people realized gaming cards could be used for general compute (GPGPU) on certain types of parallel numeric work. It was a very cool thing for many types of hackers and science and engineering students and NVIDIA went far out of their way to ensure we could never buy it.
The optimal market strategy there (as in a lot of places) wasn't "sell as much as you can". There's often a superior strategy, when (as with HPC) you have minority industry customers who are very rich and have low price sensitivity. It's to raise the price to what those special customers are willing to pay, and to drop everyone else.
What NVIDIA did was to rip out FP64 capability, systematically, from all of their consumer cards. They firewalled off "useful for GPGPU" as a differentiating feature, segmented the market, and astronomically raised the price of what (if you were looking soley at cost-to-manufacture) could have been easily affordable to any ramen student.
(It's a more obscure version of the Intel-made-ECC-memory-disappear story).
Yes, this is ironic - it was the tinkerers using CUDA on cheap graphics cards that made Nvidia stuff useful in constrained academic environments.
True about Intel and ECC, but AMD now does similar things, even with their consumer CPUs and chipsets.
These three companies now make very sure that consumer products can never canibalize those juicy data center profits - so they make sure to limit what the consumer segment can do.
The same is true for the simultaneous NVENC video encoding session limit which is artificially locked for consumer NVIDIA GPUs (see https://github.com/keylase/nvidia-patch)
For the longest time 10-bit color was only available on their professional cards. All it took was a driver update to unlock it on their consumer cards.
Same reason why GitHub was acquired for $7.5B at $250 million ARR which is 30x revenues:
VCs could not see any other reason to raise more money and Huggingface was not growing as fast as they thought to justify the valuation or the next fundraise.
So they might as well get Nvidia to save them from the VCs pressurizing them.
That doesn't make much sense to me. Companies don't just buy companies for absurd multiples out of the niceness of their heart.
In the case of GitHub, it was likely for data reasons + wanting to own where developers do work (VScode + Github).
In the case of HuggingFace, honestly not sure as I'm not familiar enough with their business. But I can assure you that Nvidia didn't buy them for 13 billion cause HuggingFace were desperate. When you're desperate, you sell for less not more.
> But I can assure you that Nvidia didn't buy them for 13 billion cause HuggingFace were desperate.
From Nvidia's side? You get to keep the shell game of where your money and hardware are going spinning on the table for a little longer. If the party stops, Nvidia loses a zero right off their valuation instantly.
And, as a side benefit, you get to place your thumb on the scale of the open-weight hosting ecosystem. And maybe even fund a Chinese Anthropic or OpenAI at a discount.
OpenAI just popped out an inference ASIC. Google is on their 8th generation of TPU. Graviton is out from Amazon. The hosting companies want Nvidia out of their finances. Full stop. Nvidia has to do something, or it's going to get swept away.
Strategically, it matters - and nvidia’s strategy is to basically own the entire industrial supply chain of the future - they have interests in their suppliers, their competitors, their users.
In this case, Nvidia wants the easiest route from:
find model -> adapt model -> optimise model -> run model
to terminate inside the Nvidia stack. This is a boon for DGX cloud.
Nvidia don't share CUDA, don't open source their drivers, don't support capable but older hardware (forget Pascal etc), and generally charge a premium over competitors for hardware.
This acquisition will likely solidify this general stance, reduce free compute allowance for a subscription, cap downloads, push advertising, generally favour models that are Nvidia prescribed or have been sponsored, and potentially ban models and datasets that are deemed risky legally (abliterated etc).
I see no other reason why Nvidia would want this kind of vertical.
Open Libraries: CCCL (Thrust, CUB, libcudacxx), CV-CUDA (Vision), CUDA-Q (Quantum).
Open Tooling: CUDA Python bindings, NVIDIA Kernel Module (Linux driver outer layer).
Nvidia is working with Red Hat to develop the NOVA driver and with Valve to develop NVK, which will be the future open-source drivers.
Nvidia supported the Pascal architecture for 10 years. No other graphics company did that. AMD recently made the RDNA 2 architecture legacy after only 5 years.
Nvidia is that most promotes open models and contributes significantly through Nemotron and various hardware-level optimizers. I think the value has more to do with the potential dominance of the open model ecosystem.
To play devil's advocate: Hugging Face was probably not profitable, and someone has to pay the bills long-term. Nvidia benefits from Hugging Face and the operating costs are pennies to them. For them it makes sense to acquire HF just to keep it running. HF shutting down or self-destructing in the search of profits would be costlier for Nvidia than just buying it now
That said, I share all your concerns. The days of a permissive hands-off HF may be numbered
They had incredible revenue growth the last few years and just broke 100M in revenue. I don't know what their internal spend was , but that was almost half of their recent round in ARR. Mostly likely they were profitable or on a clear trajectory to revenue growth. Huggingface hosts a lot of data and models, but mostly static cold storage is pretty cheap tbh.
+1 on @SillyUsername. Combined with recent acquisition of OpenRouter we see exactly what we were increasingly concerned about: tightening (chocking?) of Open Weights locally-run models. The sequence and timing of these events make the intent very clear.
Now the Open-Models crowd will try to move to some other place. But fragmentation will weaken the position. Yes, there is Civic, there are purely Chinese websites ... for those who speaks Mandarin. Which only reinforces the point.
Vertically integrated monopolies naturally have aligned in incentives for themselves, which allows more capture which turns out bad for customers. Whether it is a monopoly is an open question.
This isn't vertical integration. This is a commoditize your complements play and that does benefit customers. Nvidia benefits from supporting the open model ecosystem, if models are a commodity the gpus become what's scarce.
Oh, I totally agree, this alone is not vertical integration, but can be a path too it, long term being able to vertically integrate undercutting others that would be on top of the chain then retracting the open model.
Someone with financial / decision making weight their is going to ask why they are commoditizing models that run on "others" GPUs .... Since doing that makes their gpus less of the scarce resource..
>Vertically integrated monopolies naturally have aligned in incentives for themselves, which allows more capture which turns out bad for customers.
Apple's devices have resulted in a great increase in quality of life for myself and family members who used to waste a lot of time with Windows drivers/BSODs/random updates/malware/etc.
It's been almost 20 years, and it's turned out OK.
1) Local inference means more general GPUs and less ASIC hardware. Only big companies can push ASIC because of the software required, meanwhile nVidia owns cuda which is the standard.
2) Local is less resource-efficient per chip (chips remaining idle much more, meaning more chips required).
3) End-customers have less bargaining power compared to hyper-scalers. Although this might change if customer hardware start behaving more like phones (SoC with everything packed in), but even then the SoC makers will likely still have less bargaining power than hyper-scalers. But then nVidia could potentially make the whole SoC too.
So overall local-inference users = higher profit margins for nvidia. They much rather have every business on the globe buy one nvidia rack (or every laptop have a beefy GPU) than have 5-10 hyperscalers buy a few hundred thousand.
The hyperscalers were stable customers buying far more expensive equipment and NVidia has also made a bank selling a lot of auxillary hardware like Mellanox to AI datacenters.
But the spending spree is probably coming to an end with the looming IPO's and NVidia is probably trying to hedge their bets by making themselves the sure bet once big-AI stops monopolizing RAM and everyone races to get their local setups.
> Hugging Face turned down a $500 million investment offer from Nvidia late last year that would have valued it at $7 billion, the Financial Times previously reported. Hugging Face said at the time it did not want a dominant investor that could sway decisions.
So instead they sold themselves to the same investor completely ?
It could go either way. I think its better for Nvidia to play both open sides and closed models as their chips will be used in both. The only thing that compromises them is that there is a lot of investment they have in the closed models.
Nvidia buying Hugging Face would give them even more control over the AI ecosystem. The big question is whether Hugging Face could remain truly neutral once it’s owned by a chipmaker with plenty of competitors.
Anyone could comment or speculate what this enormous valuation stems from? Sources list their revenue as $150M? If true this is likely one of the biggest valuation gap I've ever seen. IMO hugging face is not something could not be reproduced, perhaps not very easily but I see its business model akin to github's. Various github alternatives exist and so will hugging face alternatives.
Huggingface is not a product, it's a community. It's the go-to place where everyone who has anything to do with open AI models gathers. You could trivially reproduce their stack, but building such a huge AI community around a single place is probably impossible by now.
When Facebook bought Instagram they basically bought a photo sharing app with 10-50 Million monthly active users. Today it has 3 Billion monthly active users. In hindsight, $1B looks a very good deal for a giant like Instagram. But it wasn't a giant back then, it was just a photo sharing app and it was really insane. So it must not be about the app or the network. Facebook paid that price to enter the mobile social media market as the market leader.
BTW Instagram raised $50 million at the valuation of $500M, just 4 days before the Facebook acquisition. %100 return of investment in 4 days, I think that's insane.
I think everyone just reads *illion and thinks 'lots of money'.
The amounts are incomprehensible at this point. I don't think anyone can conceive of 13 billion dollars accurately, including the people brokering this deal, they're just thinking of bargaining chips and weighing them in comparison to how many they have.
Nvidia is investing $1 billion in Poolside and paying $6 billion to license its technology and hire most of its engineers.
They are the elephant in the room, with their CDS sky-high. Sounds like they are opinionated on the whole AI thing, trying to drive this with open-source models, countering OpenAI using Jalapeno.
Looks zero-sum for the players.
All the while Google silently planning to get milk from all layers.
Nvidia is indeed a terrible company when it comes to open source, etc. however, as someone who has been gaming on desktop hardware since the 6xxx card days, Nvidia hardware has always worked. I switched to Linux full time since the 2XX days (maybe it was the 280 gtx?) and never had a lot of the problems everyone else got on Linux (using only the closed source drivers).
Likewise, their hardware has been rock solid and very performant for me. I bought most of my current setup second hand and they've been going strong for years now.
NVIDIA introduced programs like The Way It's Meant to be Played, optimizing major game titles specifically for NVIDIA architecture, which made competitions like AMD cards underperform in critical releases.
But they don't/didn't have the same level of market share as Nvidia, particularly in the PC gaming market. Nvidia had 65%-85% market share, and still worked with game studios to make sure their competitions failed. That's the opposite of good stewardship that's all I'm saying.
They worked with studios to optimize games for their hardware. I don’t think they went around sabotaging other companies hardware. They did what was good for them.
I think they will be, Nvidia has released open source model that also included the data it was trained on. I think they are the only ones that have done that.
Started from an ugly repository and ended up making history in less than 10 years, from `pytorch-pretrained-bert` to huggingface, well deserved, its fascinating to follow.
I'm puzzled by the reactions. This is not a good thing. Yes, they're releasing more open source code and models, but they do it BECAUSE OF pressure from the community and other open source projects.
But Nvidia is a terrible open source and consumer company. They gatekeep a lot and oftentimes it's only open source in name. Outside contributions are often slow-walked or rejected if they don't align with business incentives , and leadership is retained 100% in a couple of people from a certain country.
Linus isn't wrong but at the same time NV has shipped more compute than AMD and Intel combined to consumers. The kind of performance I get out of a modest outlay never ceases to amaze me.
NV is a very different beast from all of those. I have yet to regret buying any of their products and I wouldn't touch your list with a stolen 10' pole.
Isn't it super easy for the community to just replace HuggingFace? Isn't it pretty much just a repo/index for open-weight models? Why is it even worth anything?
> Isn't it pretty much just a repo/index for open-weight models?
No, its not just a repo/index (they also have training, inference hosting, and they develop/maintain a bunch of core AI infrastructure software), and even if it was just a repo/index, replacing a bug centralized repo/index that used by an large community isn’t trivial.
They were already pretty vocal about it, but this makes Nvidia pretty much the de facto face of open weight models.
As battle lines get drawn over duopoly vs. open weight it’ll be interesting to see what Nvidia does. They definitely want a piece of more of the stack especially as Huawei chips become more and more of an alternative to cuda.
Nvidia’s backstop deals are now reaching $130 billion. If neocould sales collapse Nvidia can direct that capacity towards Hugging Face, and recover some of that by selling directly.
This is the opening move of the end game for the current "AI" wave.
Hyperscalers and hardware manufacturers will absorb the overgrowth and hopefully (for them, sucks for us) settle at the top of the software value chain. The only way to make all the money poured into compute for the last 4 years pay for itself at any level is selling people and corporations on very inefficient software for the next couple of decades.
Not too different from what "the cloud" did to this market from 2015 or so...
> Not too different from what "the cloud" did to this market from 2015 or so...
I mean for all the inefficiencies of the cloud, it solved a lot of problems. It also created a new category of problems but rarely does any particular technology arrive as a pure boon without it's own compromises. Despite it's issues and despite how hot ASS some implementations are (looking at you Azure) I think cloud infrastructure you can at least say does solve a lot of problems for a lot of firms. I can't remotely say the same for AI.
It has a lot of uses don't get me wrong, but not nearly the number being pitched currently, and I don't know that I've seen any that I truly can't live without. And at the prices that are going to need to be charged when we turn the proverbial profit lever... I mean I'm sure tons of huge software houses where shipping is all that matters, behind things like quality, maintainability, the sanity of the development team, what have you, AI code assistants will have a lot of uses and the company will (probably) pay for it. But I think it's a stretch to say it's even remotely as good a trade as Cloud infra was.
In the handful of years since we've gotten all these supposedly revolutionary tools, software quality has already taken a gut-punch, will no sign of that slowing down, quite the opposite. And that's not even going into all the AI features being crammed into everything that few, if any users want or use. In fact the obsession with shoving AI shit into Photoshop instead of fixing any of the issues that've been present in that software since ~2009 was the straw that broke the camel's back for me and I've left Adobe behind entirely, and I know for certain based on communities I'm in, I'm far from alone in that.
Take a look at the historical developments of wage participation in global income, labor productivity growth vs wage growth, and so forth. It works very well. As long as the most capable laborers are paid well enough to go along, the rest of us have little choice. Especially after citizens united basically privatized American elections.
The inflection point is whether or not everyone else can convince the "most capable laborers" who are about to be cut from the "Most Capable Laborer Club" to join the Tang Ping Rebellion before real price discovery happens on their salaries.
No one seems to get that you have to do a general strike BEFORE they lay everyone off. It's a gambit. But either you actually were all that valuable and you can get your job back if it fails, or you weren't and you finally acted in your own self interest.
So where’s the return? Outside of annualized numbers I haven’t even heard people pretend that OAI or Anthropic are profitable. If this is all about ROI there’s sure been a lot of I for as yet unrealized R.
We can't discuss the finance of private companies so we have to wait for their IPO, acquisition by some public company, or bankruptcy to find out more.
The story written in the public numbers, from corporations or government statistics, is one of rise in factor productivity, rise in capital returns, and decline in wage participation in overall income.
I'm merely observing that it seems to be the latest chapter in the development of the capitalistic corporation.
IMO the top end goal is to allow capital to access skill without allowing skill to access capital. IF they can make these products work as advertised (big if, very very big if) the corpos can fire shitloads of low-end workers and give their middle managers access to these AI models and nothing else really has to change.
Workers lose their ability to earn a living, but OpenAI and Anthropic become the gatekeepers to effectively all productivity.
I know I sound tinfoil hat here but I don't see how else the numbers work. The only thing full-priced, profit-earning AI can possibly be cheaper than is, in my mind: as much of the workforce's salaries and cost of benefits as it can manage to offset, put together.
It's a great plan if you're a CEO who can't see beyond the next quarter's earnings call. Less so if you want things like, just pulling from a hat here, a functioning economy, a stable society, the ability for humans at large to live. But it's not like the owning class hasn't been undermining all of that for the better part of 50 years anyway.
Who (what) would all these middle managers manage? Agents? Why? If agents were so capable, they'd sure be capable of managing themselves or each other...
No, see, I'm making the opposite argument. This is not a bubble, this is a deliberate play by capital owners to increase returns on capital to the detriment of labor. Much like "the cloud" before. The increased capital intensity in software was inevitable and it has accelerated since 2022.
There are a lot problems with this post
- Only "The Information" is claiming the deal's done.
- Neither CNBC, Reuters, Techcrunch nor the Business Insider post whose title is cited do more than reference The Information's paywalled claim
- Only TC _even mentions_ the potential ~$13B deal on its homepage
- Not even a mention on Lobste.rs
wonder if the article got updated? Currently it reads:
>The companies have not yet reached a deal, and the talks could still fall apart, the person said. Business Insider on Sunday was the first to report that Hugging Face was fielding takeover interest.
Nvidia has a strong interest in open source AI being a strong contender to proprietary models. They don't want a single AI company to win, as that company would then have enormous leverage over Nvidia, and could perhaps even erode the Cuda moat with their own chip spending.
They want the AI market to be hypercompetitive, with AI companies focusing on competing amongst themselves instead of with Nvidia on one side, and many self-hosters / small inference providers buying Nvidia GPUs for large open models on the other.
It's in Nvidia's business interest to be a good steward of Huggingface, while being a good steward of Github is at best incidental to the interests of Microsoft.
NVidia has a strong interest in profit. The only reason we are allowed to live in chip shangri-la is because they can make a profit out of it, nothing else.
All the big boy AI stuff runs on servers, which certainly aren't going to be running on Windows or Mac anytime soon. Linux _is_ what all this NVIDIA hardware is going to run on until... what exactly? FreeBSD overtakes Linux in the server space?
> Who knows what i sthe next big thing after AI. Maybe something that doesn't need Linux?
And Nvidia hardware was running on Linux before AI as well...
> And I'm not so sure that Mac is out of the equation
Mac servers exist but anyone serious about running at scale isn't using them for a number of reasons, not least of which is that Apple simply doesn't make real servers anymore...
They have the money they have the capability design and engineering in house and they are vertical. The question is whether or not they have the will, one thing that might help them come to a decision is that they now have an engineer CEO from the Apple Silicon, part of the company and not a bean counter.
HF CEO keeps posting about how great local models are running in his Mac, it would be funny to see if he replaces his MacBook with a 500W Nvidia laptop after the deal gets through.
But more importantly for business, if HF fails to use the upcoming ASIC inference chips for their ZeroGPU after the acquisition, their competitors using them might have an advantage.
For the last few months, I’ve been working on a vendor-agnostic inference library. Have 3 backends so far: legacy D3D11 for compatibility, D3D12, and Vulkan 1.3. I have reasons to believe nVidia deliberately crippling Vulkan API for their consumer GPUs. Couple examples to be specific.
nVidia driver sets quite low number for VkPhysicalDeviceLimits::maxTexelBufferElements. I don’t think that’s a hardware limit because I have D3D12 backend doing the same thing on the same hardware.
Vulkan performance is not great on nVidia. On all AMD cards I am testing, Vulkan 1.3 is the fastest backend. On nVidia however, D3D12 is faster despite tensor cores (WMMA / wave matrix multiply accumulate / cooperative matrices) are only available through Vulkan, D3D12 is using shader cores exclusively.
Yes, they could just say they're taking it towards prioritizing models that only run on one kind of equipment and making it harder to find, support or list models that run on different hardware.
The acquisition sets up corporate interests to be able to increasingly drive it.
I hope better for Nvidia and hopefully it can be cleared up, and maintained.
Context: Liz Lemon has been buying two products together because one of them creates a need for the other.
Jack Donaghy: The only thing I will be discussing with the House Subcommittee on Baseball, Quiz Shows, Terrorism, and Media is vertical integration.
Liz Lemon: What's vertical integration?
Jack Donaghy: Imagine that your favorite corn chip manufacturer also owned the number one diarrhea medication.
Liz Lemon: That'd be great, 'cause then they could put a little sample of the medicine in each bag.
Jack Donaghy: Keep thinking.
Liz Lemon: [beat] Except then they might be tempted to make the corn chips give you...
Jack Donaghy: Vertical integration.
That's not vertical integration though. Vertical integration would be the corn chip maker buying corn farms for example. I don't think there's a name for "buying makers of complementary goods".
"In microeconomics, management and international political economy, vertical integration, also referred to as vertical consolidation, is an arrangement in which the supply chain of a company is integrated and owned by that company." [emphasis mine]
"Horizontal integration is the process of a company increasing production of goods or services at the same level of the value chain, in the same industry. A company may do this via internal expansion or through mergers and acquisition" [emphasis mine]
Anything else you care to throw at the wall to see if it sticks, just go ahead and assume I disagree with you given how our interaction has gone so far.
To be fair Microsoft haven’t done as much integration with their eco system; such as integrating it with Azure DevOps as they could. And long may it continue.
Not a fan of defending GitHub, but TBH them launching GithubActions made me abandon all other CI platforms I was using in the past, I even left GitLab just for that.
A 99.5 means over one year you will need to wait about 40 hours total for the issue to be solved. What kind of high pressure crud app are you making where this is a problem?
Oh yes I can totally believe they use public and private repos for their model training. It is surprising they aren’t ahead of the curve for AI coding like Claude code, codex or cursor because of this.
I very much doubt that. Imagine the PR nightmare and potential liability if someone managed to extract from the model credentials in a private repository.
Explain to me how using OPEN source for training is theft? I've (admittedly skimmed) quite a few licenses, but I really dont remember anyone restricting reading/indexing source code?
MIT license includes: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
The argument is that AI models trained on open source should include the copyright info for all projects blended into the weights. And since they don’t, it violates the terms of the license.
It was 2018. As someone who worked at GitHub during the Microsoft era, I think Microsoft was a pretty reasonable steward of GitHub right up until Copilot started getting popular. After that, leadership stopped investing in the core product and infrastructure and either incentivized people moving into the Copilot org or just fully reorged teams into it regardless.
The load increase from AI is definitely significant, but I think with proper investment they would not be having this many issues.
HuggingFace offers abliterated/uncensored models, which would seem to run contrary to the "AI is too powerful for regular people" narrative that the big players are pushing.
My primary concern is that Nvidia will bow to the pressure and restrict abliterated/uncensored models on HF.
Having Nvidia behind HF will now probably have a bunch of litigious entities salivating to sue them for not only allowing, but also distributing uncensored AI models that will then be used for deepfakes, porn, spam, scam, hacking, etc.
Sure that could have also been done to HF before, but I'm thinking Nvidia is a much juicier target.
Should also now be much easier for the government to just eventually force NVidia to restrict access to Chinese open models seeing as it will now become an American company that needs to obey to American law and we all know how American AI companies are lobbying for exactly that.
>Should also now be much easier for the government to just eventually force NVidia to restrict access to Chinese open models seeing as it will now become an American company that needs to obey to American law and we all know how American AI companies are lobbying for exactly that.
Using regulation to avoid having to be competitive is the American Way!
In all seriousness though, probably time to start an underground equivalent to HF.
OP said they will bow to pressure, which just means that governments will be able to control nv and get them to ban the obliterated models, implement ’safety’ controls on all uploaded models.
It was either they (literally) sell out or enshittify. Choose your poison. Hosting and running LLMs is an expensive business to be in. There wasn't ever going to be a good ending to this story. The original investors need to be made whole.
Not if they can't indefinitely contain ram and SSD procurement. And they can't and the Market already showed with specex that they are unwilling to float these fly by night outrageous 'investments' onto institutional holders.
Institutional holders mass revolted at spacex getting into the basket.
AI costs more than having people do the work. Q2 CFO reaction proved that.
Forgive me for newb question, but why Hugging Face is a thing at all? Couldn't models be simply distributed via torrents? And the whole aggregator thing would come down to indexing magent links published by model providers?
There's a herding effect with torrents where popular things are quick to get, unpopular things may be very difficult to get, because the majority of users delete files to save space.
I might host 3 or 4 models that I've downloaded recently, but I won't be hosting the 50 or so that I've tried in the last two years.
And who seeds it? If you're going to pay a host to seed it, they might as well just provide HTTP and let you add them as a web seed to your torrent, which we could be doing already.
But if you don't pay a host to seed it, either the host's business model won't play well with torrents, or there's no host and a torrent will rot.
The important thing about HuggingFace isn't the safetensors files, it's the READMEs and the Google search placement. It's easy to reproduce their distribution with torrents, it's very hard to then establish yourself as the default option.
It would be far more efficient to decentralize the data. But HF provides convenience, they subsidize the cost. They gained millions of users, and many enterprises. They can now sell the popularity of their platform. Nobody can buy a p2p network of people.
Or even “why would I pay for a flight when I can simply walk across the country”
It’s all about the convenience, minimising the time and effort from “looks interesting” to “running the model”.
Nvidia don’t care if you do it on their cloud, someone else’s cloud, or on your own machine - they win either way, as it further propagates the technology on which they are building their future.
hugging face is basically a real easy way to run llms. They also provide a bunch of libraries to do things like split compute across all your resources.
I'm sure it'll be fine. As long as you have the latest 90-class card you'll probably continue have nearly full functionality from the HF libraries. (Upgrade to enterprise gear for full functionality)
The fit is theoretically good (nvidia is model agnostic) but I can see then pushing the ecosystem increasingly towards cuda and some custom nvidia software packages for inference etc etc
I'm not a fan of big-tech acquisition results either, but one benefit can be that a product continues to exist when it would otherwise become insolvent.
That specific example might be relevant to the case "I don't see any acquisition being good for the users", but not for the specific case of the HuggingFace buyout.
Workflow was already operating within the bounds of the closed Apple ecosystem. Huggingface is providing models not only for Nvidia chips but for their competitors hardware (Apple Silicon, AMD GPUs, ...).
I was not making a point. I was trying to clarify the example.
But after the answers, I do think his example is poor. Models available at HF have many purposes and will run on many platforms. Commenters here are suggesting this will no longer be the case.
Whereas the example Workflow (iOS only) -> Shortcuts (iOS only), there was no change in that regard.
The unlimited private repos for free tier would have happened anyways. You have to remember that this was around the time people were jumping ship from GitHub exactly because they didn't offer private repos for free. It was something many wanted, and they were actively seeking out other places that would give it to them.
It was unreliable post acq. and before the slop-tsunami, the latter has just multiplied the underlying issue of their ipv4 only half assed azure migration.
Can someone explain how this works? Huggingface serves enormous files and therefore presumably must pay huge cloud fees. As the dot com joke goes, they make a loss on every transaction, but don't worry, they're going to make it up in volume.
Nvidia has already a good free offer with Nim and a lot of compute power available. Plus, they obviously have interest in spreading open models culture.
Keep an eye on Exa. With local models and niche harnesses proliferating, they're going to need a way to search without everyone having to run their own index.
Google is an obvious buyer if they could get it past regulators. A Doubleclick-level addition of the agentic stuff they're bad at to the indexing they're good at.
Nvidia is the obvious less conflicted #2
Secret third option: Cloudflare. They're already the glue for so much of this stuff.
We keep talking about EU having to be more present in the AI race, but if the lifecycle of European AI companies is to be bought by US ones, what’s the point?
This acquisition should be blocked (and never will sadly), they now have both the incentive and the ability to influence a platform that's supposed to be hardware agnostic (and built trust over that) toward CUDA stack
Microsoft did the exact same with github, acquired stack agnostic platform, and turned it into a copilot/azure one
Keep alive the on-prem hardware sales? Not as big part likely but still something good to have. On-prem hardware needs models to be available to be useful. So pushing those could sell more on-prem hardware.
I really wish I'd done better bookmarking back when Nvidia was talking about acquiring ARM. There were all kinds of things coming out saying, if we do this, even though we will be taking over ARM, it will transform us from the inside out. We would become a new different company, that cares about something beyond our own self interest, our own chips.
This feels like a similar leap of faith. One that is hard to believe in. Thankfully, I think Nvidia can keep the lights on here & keep this going. I don't think they have to do much, per se. But it felt implausible then to image an Nvidia that gave a shit about anyone else, an Nvidia that actually gave a flying fuck about drivers or upstream Linux or ecosystems that weren't entirely within their own control.
Similarly the upper quartile of succes here feels mostly like benevolent neglect. I think we can hope for Nvidia to just not mess up a good thing, for them to understand that this open model open ai universe hinges upon Hugging Face, and for them to pretty please keep caring about the existential risk of the hyper-ai'ers all building their own properietary models on proprietary hardware and leaving Nvidia behind some day, and HF being the hedge against being left behind.
The story is in an intermediate state right now, but as semiquaver points out, The Information is reporting it as fact. Since their reporting tends to be as reliable as it is hardwalled, we went with their claim in the title.
Please don't post unsubstantive comments to Hacker News.
No doubt you have the nucleus of a substantive comment here, but that's not enough. If you only post the shallowest top stratum of what you're thinking, other people do the same, and then we get "Laws are for poor people" and endless descending repetition. The whole point of this site is to try for something other than that.
I thin HF holds huge power in how they’ve consolidated all the open models and data, but monopoly feels like a stretch here. They are in reality a fairly small startup with lots of alternatives (all of which are much worse at this point granted) and are mostly a loss leading part of the open source ecosystem. I think it would be very hard to make a case against this on monopoly grounds, though I wouldn’t be opposed to someone trying if they thought it was possible.
Nothing paradoxical about it, open source AI use raises demand for nvidia's chips too.
HF was.. a hub to download models and some of the worst source code in the space that contributed to huge amounts bugs that did a lot of downstream damage. Go ahead and read their blog by their CTO on how they don’t believe in DRY and then implemented DRY in the worst way imaginable with unnecessary code generation.
Their BLOOM model was a joke and dead on arrival.
But yea these guys are definitely going to be the frontier of EU AI.
Yeah probably more like the frontier of early retirement.
(half joke aside, it is finally coming in a couple days.. which I have been saying for quite some time to the extent that my peeps don't even trust me anymore.. 3 things are coming and it is genuinely research grade apparently, for the amount of research I am aware that is released publicly. I really don't know how people ship things, the tooling I see is terrible, or maybe I have NIH syndrome...)
"The company was founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf in New York City, originally as a company that developed a chatbot app targeted at teenagers. ..." Wikipedia
Besides that: which EU and sovereignty doesn't really rhyme, given data protection, energy costs, AI restrictions and no vital start up scene at all.
Mistral is heavily state funded via Banks that are predominantly owned by the French state, so I don't get it.
Also name one global success that doesn't came from US? China? Well, second place maybe, but second in any sense.
Nevertheless, kudos to huggingface. Not only to the founders, but to the ecosystem and I guess it got harder by the day to finance all the hosting and only hope for the future, that Nvidia will help HF to thrive, at least not put burdens on them.
GitHub is still alive, and hope that HF will benefit from more AI power.
Nvidia doesn't have any disadvantages in my playbook, because hardware wise, there is no one to beat them in any regard and Apple will always stay nichy.
So well played - hopefully.
It could be worse or even a shut down otherwise, who knows. So for me it sounds cool and I am glad that EU isn't involved in any regard.
As the AI Act demands full disclosure under really easy to match requirements into your inner workings, this is quite the opposite of freely sharing information.
TikTok: achieved global cultural and algorithmic dominance.
DJI: controls roughly 70% of the global consumer and commercial drone market.
CATL: The world’s largest EV battery manufacturer.
BYD: Surpassed Tesla in total electric vehicle production volume.
Shein & Temu
The list goes on…
Paying pennies on the dollar for a BYD vehicle sounds great when all you consider is the initial sticker price. The real cost is what happens after you buy the disposable car and own it for a decade.
So yea. Second place is the reality and it is also true for AI models, and the hardware needed to run them.
Also, congratulations on landing a rocket 11 years after the US.
Took long enough.
You might want to remove your pink USA glasses and look around.
Spotify good enough for you?
How about...
SAP (Germany) - Most of the Fortune 500 run on their software.
Spotify (Sweden)
Novo Nordisk (Denmark) - most of America is taking their drugs.
Airbus (France) - Has out competed and out delivered Boeing for years.
Ericsson and Nokia (Finland and Sweden) - Most of American telecom systems use tech from them, essentially the western alternative to Huawei in telecom networking.
Adyen (Netherlands) - One of the worlds largest payment processors, processes most of American big tech payments.
ASML (Netherlands) - Responsible for building effectively every modern chip in the world outside of China.
Maersk (Shipping and logistics)
Siemens (Industrial Tech)
Infineon (Semis)
And there are so many more I'm missing btw. It's funny how incredibly biased you are, but the reality is that Europe and the USA are totally dependent on each other, tech wise.
If every European company which the US is dependent on immediately stopped trading with the US, the US would collapse as a country. And yes, the opposite is true also. But it was you who argued that Europe effectively has no successful companies, what rubbish.
May I remind you that in the US, Facebook is in that same list of top successes? Have you ever heard the phrase “the top minds of this generation of Americans are all working on ads”?
Also, if huggingface is “technically” American, then fuck me, Stripe is technically European. Hell, their CEO is insanely pro europe, pro EU, and a big enabler of projects like EU Inc and lately, the Rhine Group.
They may not be a financial and industrial juggernaut like the others listed, but cultural significance doesn't have to be all about money.
I hope nvidia does right by the community.
Edit to add: $13B should cover the S3 egress fees for a couple months :D
I think you mean hecto-millionaires.
Can I setup a forgejo instance, upload a bunch of open source models to it and get bought out for $13bln?
I fail to see the point of this. Their inference hosting is a joke (I don't think I ever had a working response from a "try this model" form).
Their brand alone surely is not worth this much.
So what are they paying for?
I worry the answer is, for control over who is allowed to distribute the weights. Nothing else.
If Anthropic, OpenAI buy them, this org gets very different priorities, owning them ensures that doesn’t happen.
But in the era of frontier models and hardware designed for fast model loading (eg: LLM on a chip) I can't imagine that this is worth so much. There's *a lot* of competition to improve Git LFS.
So while I agree with you that this plays a role I also have to believe that it's to understand model usage by competition and thus stifle it.
[0] - https://huggingface.co/docs/hub/en/xet/overview
You can't just spin up a VPS and throw a few brain jar's in; the network bandwidth must be a fortune. Hosting over 700GB+ for many of the same models has storage cost too.
HF has become the front runner in open model hosting. They have a lot of trust from the wider community. That's gotta count for something.
They are paying for an insurance policy. Elon Musk and SpaceXAI must be incredibly interested in buying the company who was the victim of a mass hacking campaign by OpenAI. Any lawsuit, criminal investigation, or slowdown of OpenAI's model training would put OpenAI's massive data center build out, which Nvidia just backstopped, at risk.
Code is hard but more likely it is a brand thing which is much harder to do. Nvidia for years now has tried to get a public ai model hub of sorts working. This is obviously them throwing in the towel.
It's such an obvious customer funnel they have been trying for years to make.
1. Customer googles "some model" 2. Hugging face is often top result 3. Thinks it's great 4. Buys
The customers are already on hugging face.
Acting like code is the blocker here would be very wrong. Nvidia 200% has the technical knowhow to make a hugging face platform and they'd likely make it better.
I'm not that familiar either, and so had same question.
I thought they just hosted other peoples models. So why the value of 13B?
More than anything, they are buying a position within the ecosystem. Many of us could get a version of an app deployed in short order that implements most core HF functionality, but I am sorry to say that this is very far from the "hard part" of building a business, and this was the case even before coding agents were a thing. It's no small thing to become the first tool pretty much everyone reaches for in a particular niche, and therefore the incumbent in that niche
And there's lots of inertia keeping companies and developers on these incumbents like HF and GitHub -- see also Microsoft's play here dumping billions into a company that is basically peanuts and probably nowhere near as profitable as Microsoft normally would like. But they bought it at a time when they were trying to make Microsoft synonymous with open source, and it was a great play to that end. Likewise Nvidia wants to be synonymous with AI. Even if HF is a loss leader, the positioning is probably worth it
To a lesser degree, Nvidia probably also might like to reinvigorate its flagging cloud compute business and this could be a natural way to do that. At least TechCrunch thinks this, but I, an overconfident layperson, feel that the positioning is where most of the value is for Nvidia, and consolidating that will act as a force multiplier for that + anything else they may want to do with this space
(Ggml.ai is llama.cpp.)
Curious if the “I consider HuggingFace more "Open AI" than OpenAI” sentiment in that top comment will still apply with NVIDIA as the boss now...
Historically NYSE (and AMEX; thought of in this specific context as CTA) had a lock on 1-3 character symbols. NASDAQ (via UTP) had a lock on 4-6 character symbols.
For a long time you could tell where something was listed purely by looking at the length of the symbol it traded under.
This all changed in the late 2000s or so, so no longer useful information.
I honestly would leave the industry if one of the exchanges tried to bend the industry to allow an emoji in a symbol. The communication protocols essentially lock this at 8 character alpha. It would be a huge pain in the ass, for everyone, to change this just because some exchange wants to appeal to a potential listing.
EVERYONE said it was a bad idea but they kept pushing it. Repeatedly.
Eventually, smarter heads prevailed and they got rid of the idea.
NYSE (really the access stack that both NYSE and AMEX ran on; the name escapes me since it died back in I think ‘08) has for a long time included a completely separate field in their protocols to indicate classification (preferred, warrants, classes, when issued, etc), but CTCI just jammed it into the symbol field.
Good thing “xn--zp9h” fits, then.
Not as common, again, as it used to be...but still exists.
https://www.nasdaqtrader.com/trader.aspx?id=CQSsymbolconvent...
** for those that don’t know: NASDAQ stands for “National Association of Security Dealers - Automated Quotation (service)”. It was forced into existence by the SEC. Prior to that there was only the NASD, who traded blocks of shares amongst themselves so that each individual security dealer could meet the needs of their customers. As the trading amongst themselves grew larger and larger, the SEC added progressively more regulation until finally the had to build the NASDAQ computer system, which nobody would even recognize today.
So again, they were always separate from everything else and did things that made sense according to their own needs. They were very different until they were no longer allowed to be different.
By that measure regional exchanges were their own beasts for a long time. The environment has definitely streamlined a ton since RegNMS was implemented and the exchanges consolidated a bit.
Even the technology stacks the current markets run are basically identical within three families of design (even the upstart markets, and even those that are not directly licensing technology from ICE, Nasdaq, or Cboe). Differentiation just doesn’t pay like it used to.
Honestly the problem is the ticker is too valuable advertising for the listed company, and since listings are not governed by a central regulator in the same way they are in Europe, for example, there’s always going to be this sort of friction about who is willing to bend the most to win the business.
Probably they became more popular/important specifically because tickers did not, which in turn is probably because there are so many countries/exchanges that a global exchange-driven namespace was never really feasible?
Edit: Just discovered via this post that hn posts, like the exchanges, filter out emoji
Ironically enough, reddit recently killed off old.reddit; you now have to log in, which defeats the point of old.reddit.com totally.
https://addons.mozilla.org/en-US/firefox/addon/old-reddit-re...
Owning HF -- the discovery and distribution channel -- is one thing, but I think the biggest threat vector is the privileged access to HF platform data, that includes HW survey info and model download pattern. This can be a borderline anti-trust case.
Reversal would be "valued at 7B then valued at nothing", this is more like "nah, we want more" then "nah, we want more" then "yes, that's what we want :)".
very few people can lock a vision and fight billions dollars with integrity. In tech they mostly end up shadowed. People like JB Kempf with VLC or Bellard. Maybe even Torvald or Jeremy Howard belong to this list of people who could have been way richer than they are now if they wanted to go against their visions and beliefs.
This is the article you are referring to is this but your facts are wrong: https://techcrunch.com/2026/08/24/hugging-face-reportedly-in...
rant over.
> We make money via compute credits + Enterprise Hub + HF Pro subs [...]
I guess this plus custom inference deployments, external inference providers, partnerships with the big cloud AWS, Azure, etc.
[1] https://x.com/reach_vb/status/1928050126498713706
–Some guys in every bubble I’ve witnessed.
I get that it's fun to be glib about the stupidity of tech elites and investors in general, but Huggingface have been pretty open about their financials and are, in my opinion as a practitioner in the field, one of the most responsible orgs in our space. They've been a pillar of open source ML for years now and have made a very positive impact on our ecosystem.
Nvidia is getting a real business generating revenue, and the center of the universe for open models. Both seem like pretty valuable attributes, from Nvidia's perspective.
Or like GitHub, which was generating something like 200 million in ARR and had never hit profitability when Microsoft bought it for $7.5 billion back in 2018. I'm sure it has come as nothing but a happy surprise to Microsoft that GitHub generated $1 billion in 2023. They had initially penciled it in for 38 years til ROI.
They also have inference endpoints with metered prices, and their spaces product (though i'd imagine this is a smaller portion of revenue).
Nvidia has a market cap of $5T USD today, and a decent chunk of that is due to LLM speculation.
Does Nvidia want their stock price to be at risk of being tanked by a download service being in the news? No, they want to make sure the party keeps going and is under their direct supervision, and part of that is making sure Hugging Face isn't bought by a competitor or runs out of money.
To stop that! Literally. To stop those services being free.
Google is obvious.
Hugging Face is a file download mirror with a couple of side features dangling off.
Pretty nice dangling side features apparently.
I can't see this as being as good of a purchase, especially when it's 13x the price of what was seen as an absurdly large amount back then.
Or perhaps they will start throttling downloads for free users.
I don't know what they business case is, it might be to shut them down: I suspect good free models on local hardware is a threat to Nvidia's investments in OpenAI/Anthropic.
Unless I’m missing something, this feels like Nvidia having more money than they know what to do with.
I do worry about any sort of crowding out or downplaying non Nvidia-relevant quants, and about changing rules to crack down on models or datasets that for one reason or another “don’t align with their corporate values” - uncensored etc. Someone mentioned Microsoft and GitHub, they appear to me anyway to have been very hands off, I hope it’s the same model.
"Quantized models are no longer permitted on the Hugging Face platform. Reduced precision models are a safety hazard and a violation of our TOS. Click here to speak with a sales representative about our many exciting cloud hosting offerings or enterprise GPU packages."
I wonder how much it's worth to Nvidia just to obtain a list of who's downloading which models and for what purpose. Similarly with the Stripe acquisition of OpenRouter. In the past, it would have been valuable to obtain an audience or market share or even a productive team. However, in this new age of the surveillance state ...
telemetry, orchestration, and financial
2. I think it will happen much sooner than 2245 given how easily capital and the rich are able to move around the globe in 2026 (compared to 1945)
If a CEO cannot personally get richer by cutting wages, cutting staffing, and outsourcing; all of it stops. That was the secret of affordability. Don't let a thousand rich people own everythjng. Tax them so we can have millions of working millionaires instead of 900 non-working billionaire elites.
As those 1950s taxes and post 1929 financial regulations were cut in the following decades, the wealth inequality that led to the great depression has returned.
If you want to judge the net effect of tax policy over time, look at the percentage of GDP captured by federal receipts (the vast majority of which are taxes). In the U.S., it's averaged ~16.5%-~17.5% with occasional spikes and dips. Currently, it's very close to what it was in those 90%-marginal-rates 1950s.
People have a right to support candidates > they can print a sign and stand on a street corner > printing signs costs money > people have a right to spend money supporting candidates.
People have a right to assemble > they can stand on a street corner with their friends > they can hold a big sign together > they can have lots of friends and form a group and print a really big sign.
We want to encourage private business > limiting liability for owners who do not participate in the business would increase access to capital > we offer corporate forms that shield investors from liability > identity of owners may be kept confidential.
People complain about the wrong thing. The problem isn't that people are allowed to support candidates with their money, buy airtime, etc.. It's that companies have secret owners...
Only natural persons should be able to contribute to political candidates.
Corporations can lobby, as a corporate person, just like a union or other organization, but not contribute.
They can run ads themselves, but there needs to be a much greater wall between PACs and candidate campaigns, with the previous restrictions now null and void due to lack of enforcement.
If it's okay for Corporations can run ads themselves, then your objection isn't to wealthy people supporting candidates or even them doing so secretly, but... only when the little guy pools donations and they happen to be pool donations via a corporation?
The problem isn't "corporations can do the same thing as people."
The problem is anybody (solo or as a group) can legally run an unrestricted indirect campaign.
Elon can do the same, or more, as an individual than just about any corporation. That's just as problematic as Ford or Meta buying an election.
Dark money contributions are a closely related problem that we ought to fix at the same time.
Corporations only have the powers granted to them by the state. Contributing to politics need not be one of them.
No, state's cannot e.g. discriminate on the basis of race by allowing corporations only composed of certain races. States also cannot infringe on the right to assemble or speak, by prohibiting people from supporting political candidates when they are organized as such a group.
But, generally, yes. States can require that the owners / members / corporate books / etc. of a company be disclosed publicly. If owners don't want their books disclosed, then they can forego the immunity that the corporate forms provide.
> People have a right to support candidates > they can print a sign and stand on a street corner > printing signs costs money > people have a right to spend money supporting candidates.
Individuals printing signage is not where most of the funding is going though, is it?
> People have a right to assemble > they can stand on a street corner with their friends > they can hold a big sign together > they can have lots of friends and form a group and print a really big sign.
This isn't how most of the money is being spent either, is it?
I don't know how you seem to either believe the point of the right to free assembly was to give richer corporations more sway, or be completely oblivious to the fact that that is what's actually getting people angry, not the fact that citizens want to stand with signs by the street corner. Assuming you don't have a conflict of interest leading you to that conclusion, I... don't know what to tell you.
Fundamentally it sounds like your complaint is that wealthy people are presumably more able to be effective with their efforts than folks who, e.g. can't afford as big of a bullhorn.
Citizens United wasn't about that. There wasn't a law saying no person shall spend more than $1k/yr supporting a candidate.
Speech is speech, and what people forget about the First Amendment is that it prevents the government censoring their speech.
Corporations no.
A railroad company also has a huge amount of steel roads with cross country right of ways. You could fire everyone, and it'd still be extremely valuable. It might even be more valuable.
Even in asset-lite software land, "Atari" the name still has a lot of value, even though none of the people are there anymore. They own the right to control the things that the neurons in millions or billions of people's brains are tied to, that their habits tie them to.
Twitter, you can fire almost everyone, make the online square less pleasant for a lot of the people, the network effects truck on. People continue to use it even as they hate using it, because most everyone else is.
Democracy is built on the idea that people are equal and have equal rights. This is only sustainable if the power differential between people isn't too large.
Corporations consist of people, yes, but the power to control where the efforts of these people are directed is concentrated on very few people, giving those people a large amount of power -- so large that without any checks on their influence on government, it undermines democracy.
It's the same argument for why super-wealthy individuals are so problematic. They're literally destroying the fabric of democracy through the power imbalance afforded by their wealth.
A corporation having contribution rights multiplies the otherwise imposed contribution limits on the underlying humans.
A corporation allows tax planning opportunities not afforded to the bottom 90-ish percent of the country (such as shifting income to alternate years to avoid progressive taxation, and utilizing expenses to lower that income in ways a W2 cannot)
Corporations are granted benefits that people do not get, and CU disturbs the balance of power.
This was a major crux of Colbert’s “our lady of perpetual exemption” bit. People absolutely know about the shadowy machinations CU enabled. The lack of transparency is a major critique.
To say that repealing CU is the solution is to allow the critique to go unresolved, while arguing ~"People shouldn't be free to support political candidates."
Reminded of this discussion from a few days ago https://news.ycombinator.com/item?id=49389372
We have the techno-feudalism and dystopia but none of the cool aesthetics from cyber punk or arguably even the real Zaibatsu. Nvidia isn’t cool, there’s no interesting aesthetics or back story, it doesn’t appear to have zaibatsu ambitions, it’s just riding a meme stock bubble.
In a different world I can imagine SpaceX to be the only thing that could come close, but well...
So either way they win.
I was puzzled by the comments on the openrouter acquisition saying "they're just a proxy" - they're an open marketplace where models compete head-to-head, they're quite a bit more than just a proxy.
Hugging Face has essentially become github for model repositories and I think nvidia wanted to snag that before microsoft does.
They run a proxy for LLM providers, which inevitably creates a marketplace. They’re amazon.com for tokens.
The flip side that others are pointing out is that Amazon is sticky because there’s a sprawling, physical logistics apparatus underneath. You can’t replicate amazon.com’s business without billions of dollars and a decade of building warehouses.
I don’t see where OpenRouter has that. As far as software goes, theirs doesn’t seem particularly tricky. LiteLLM does vaguely similar things, at least for organizations where managing accounts isn’t absurd overhead.
They were first, and there’s money to be made there, but what’s stopping someone else from building LibreRouter that charges a 3% or 4% fee instead of 5%? That’s where I get dubious of their valuation.
Given that most of the weights are from Chinese labs and teams, I'm surprised China's own ModelScope hasn't leapfrogged HuggingFace.
They have network effects and strong brand recognition, both of which are hard to replicate.
NVidia have been making lots of moves recently towards supporting open weight models and local AI, and it makes sense for them to buy HF who support the same, if for no other reason than to prevent someone else like Musk buying them just to shut them down, although I assume they have plans to do a lot more than just keep HF alive.
I'm ready to be wrong but doesn't this describe basically every tech acquisition for the last 25 years or longer?
Pull requests, CI platform, authentication that is not PAM etc.
Nvidia is heavily invested in OpenAI, recently guaranteeing a datacenter buildout up to $105 billion dollars.
Any reasonable criminal investigation into OpenAI's actions or negligence into this incident would hurt their reputation, their ability to raise money, and therefore Nvidia, who is invested in them.
If I was a founder of HuggingFace, I am immediately getting in contact with Elon Musk and SpaceXAI, and letting him know how that he can purchase them and sue OpenAI, this time with solid claims.
If I was particularly amoral about if I think OpenAI or Anthropic should get to AGI first, I would take this bid to Nvidia, whose entire stock price is propped up by OpenAI's over-purchasing of compute.
I see this as NVidia placing 50% of their bets on local models as the future, which it seems may be more profitable to them than cloud (sell 100/whatever local cards, which see 40hr/week max utilization, vs one cloud card seeing 24x7 utilization).
NVidia seem like a smart company - if they want to promote this direction they are not going to cripple it by trying to limit quantized models, even though you might expect them to promote benchmarks showing the benefits of larger and less quantized ones.
It's like the smart Intel of old - support CPUs all all price points, while also promoting CPU hogging applications like OpenCV.
Just out of curiosity, how did Hugging Face have the money to run that operation?
The most important resource for self-hosting LLMs is now under the control of a company that has very markedly kept the specialized hardware outside of the broader public's hands.
In all seriousness, isn't it pretty straightforward to quantize a model locally? I can do it using mlx in one command.
Nvidia has a deep interest in open source models to counter the LLM hyperscalers.
Nvidia also has a big bet on robotics and local AI, which requires model efficiency.
I'd worry far more about Dario and Sam regulatory capturing the US market via a "model safety" scare.
But selling more product is still on thr table until cloud AI starts the real cash drain, then therell be google like shenanigans.
These acquisitions are not healthy. They stifle competition and make it easier for governments to censor open weight models by choking off distribution points.
Has always been this way. Big companies always seek to buy fast growing startups and unfortunately the founders of these startups mostly take the money.
That’s at least a plus. I will happily burn through as much VC money as they will give me to tinker with my projects.
I think the federal antitrust regulators are asleep.
Edit: antitrust regulators' job has just begun -- we'll see how this deal gets adjudicated by the FTC (if at all).
Have been for a long time. All of the big techs should have been broken up long ago.
That is what will happen to torrents. Followed by VPNs. China, Russia and Iran had paved the way already.
Can't you see the puzzle coming toghether?
If the House or Senate turn that might actually stop or slow down. The market falling by 30% when the AI bubble eventually pops would also trigger it.
Fear over economic damage due to immigration is perhaps the last remaining unifying force among the American political right wing. Biden actually did a lot of good things as president -- arguably a lot more good than he did in his time as senator. But his administration was too eager to just run the federal government as a quiet meritocracy and avoid dealing with any of the hot political issues of the day. He and his administration fucked up so bad that now not only is most of that good undone, but we are in a dire existential situation, facing generational-scale damage that might be unfixable.
It's one thing that they genuinely fumbled the handling of the huge immigration/refugee waves spanning the entire continent and it took them well over a year to start coordinating a response (which they never really did apart from eventually just closing the border) -- it's another that they kept trying to insist everything was fine, there was no problem, there's nothing to see here. People were freaking out about it all over the country, and of course they were being heavily propagandized, but that's not the point. A strong and proactive White House would have made all the propaganda look stupid. Instead, it looked like the truth.
It's absolutely mind-boggling to look back on just how weak the president had to have been from 2020-2024, to somehow let the party of Jan 6 and the disastrous handling of the Covid-19 pandemic regain enough control of the political narrative to make Trump seem like anything other than a narcissistic mobster.
And some former Roomba employees may have thoughts about Kahn.
IMO allowing Roomba to hit chapter 11 was still a better option for the consumer than handing them to Amazon. They’re still in business as an independent competitor on the market, consolidation was successfully avoided.
Yellen, I don’t have much of an opinion on, but she absolutely wasn’t alone in that opinion (the nuance of that opinion being inflated by this hyperbole), and the treasury is a lot less involved than the federal reserve in doing anything about inflation, anyway.
These companies are generally large enough that they do not need the competitive aid of buying an existing company and starting with that sort of structural advantage.
E.g., did Nvidia not have enough money in their bank to start a company to compete with Hugging Face? This is a company that reportedly has ~200-300 employees with investment rounds totaling $400 million. Nvidia made $31.9 billion in net income last quarter.
I look at a company like Xiaomi which just developed its automotive division in-house without resorting to buying car companies. I think our traditional business and finance mindset has an overreliance on acquisitions.
Is either of those worse than what actually happened: the company is now a zombie brand for a Chinese company?
Amazon could email/push notification/text customers asking for reviews of iRobot vacuums but then not do the same for competing vacuums, skewing their reviews higher (asking for reviews boosts ratings by gathering opinions from happy customers who usually don't bother writing a review). Competing brands potentially don't even have your contact information to ask for a review.
Go on Amazon right now and search for "usb cable." There's a giant banner at the top that recommends the Amazon Basics brand, three across horizontally, which on my desktop monitor takes up nearly 50% of the screen real estate. Then below it are the Anker cables that you're more likely to be looking for.
They would have almost certainly been manipulating pricing on them as well. For example, they could take the strategy of lowering the price of the vacuums to break even or sell as a loss leader, but use them as a data-gathering robot in your house to help Amazon sell more of everything else. They could make their Alexa smart home platform preferential to iRobot or lock out competing models.
Robot vacuums are a consumer goods category that is heavily skewed toward Amazon.com as the place of purchase compared to other retailers.
I think "zombie brand owned by a Chinese company" is actually preferable, yes. They still operate and sell vacuums competing with the other robot vacuums on the market, and they aren't in service as household data collection bots for a monopoly e-commerce platform.
I don't think the ownership of the company being foreign or domestic is very relevant to the FTC's goal of preserving positive trade conditions. Would we think the iRobot situation was a bad outcome if iRobot was purchased by a foreign company we view more positively like Miele? We only think of it negatively due to anti-Chinese bias. iRobot being Chinese-owned is almost certainly the best possible outcome for preserving the amount of competition in the market.
They are essentially on equal footing with other robotic vacuum manufacturers. iRobot didn't go out of business or get absorbed into a larger company lowering competition in the marketplace.
It's also not the FTC's job to ensure that companies, especially ones with zero/trivial national security or domestic labor force value, remain under domestic ownership. Amazon itself is not really a "domestic" company, it's publicly traded. Anyone from any non-sanctioned country can buy shares of Amazon.
Huh?
Robots with vision + sounds sensors and connected to the internet in professional homes across your country are definitely a security issue.
https://www.abc.net.au/news/2024-10-05/robot-vacuum-deebot-e...
One of the leaked images taken by Roomba robot vacuums shows a lady sitting on her toilet - personal.
I spent a little time working there around 2017. Whatever you're implying my former colleagues' thoughts on the subject are (or mine, for that matter), you're probably wrong.
The era of VC-funded home-delivery recipe boxes is still my favourite.
Its likely in the near future we will be able to buy 128gb mac minis and run local AI for free.
If Nvidia buying HF makes it tough for all the diverse models on HF, then what are some alternatives?
It seems models are the best things to be available on a Torrent platform? Of course HF is much more than just the files but perhaps the metadata can be separate and hosted on multiple community platforms.
It's less of a pain now that I'm not in my LLM newbie phase of trying to test out 30 different models and stick with Qwen, but I couldn't even try to download qwen 3.8 for a week or two until I got back to the city.
Starlink is the obvious option and I have it on my roof but I do everything I can to not pay that $170 a month vs my $170/quarter 5g service.
good old torrents
Ensuring a healthy open-weight model market means ensuring continued demand for metal running in corporate data centers, bypassing any middleman. NVIDIA has a strong interest in a thriving "run your own agents" market.
They're actually aligned with Apple here. And in this market, Apple is a real competitor too, with Apple probably more consumer-centric, while NVIDIA probably more enterprise-centric.
Dgx Spark and Strix Halo have very close specs and deliver similar performance. If nvidia makes their stack be more efficient with for example 50% more tockens on similar hardware specs agaist competitors, they don't need HF.
My take is that Chinese Labs, though slightly behind on the frontier (due to compute constraints) are on a trajectory to surpass Western labs (this is me speculating, reasons are better ecosystem creation on China's part and potentially better/more data environment). Qwen-3.5-122b was the king in it's category and noone came up with something better, even though many tried like poolside with laguna. Similar with 35b and 27b param models. I think poolside and HF acquisitions show us that nvidia really wants to have competitive models on the prosumer (~100-150b param size) and likely at the 300-500b as well. Together with a hardware to run them that's a good market to be in. And as the recently rumored Xiaomi AI cube shows us (together with gorgon/medusa halo and mac studios), this is a market segment that will have competition.
Models are already largely hardware agnostic. It would be pretty hard to put that cat back in the bag.
I could imagine them building value-added services on top of HF to advantage Nvidia products (i.e. "run this model on NVIDIA cloud" with one-click), but in this moment it's hard to imagine how they could actively disadvantage models built to run on other platforms.
The party is a is a relatively small concern compared to the openness and github comparisons, but since no one else mentioned it, I hope something like it happens again. Or maybe that moment has passed.
Don't expect things to go differently this time around. Nvidia wants control over the software stack. Acquiring HF fits in perfectly. The play is long term.
They even share many of their pre-training and even post-training datasets for Nemotron on HuggingFace; for example: https://huggingface.co/datasets/nvidia/Nemotron-Post-Trainin...
Which other lab shares this?
Yes, there is no question NVIDIA wants to lock you into CUDA and their hardware. But also, they’ve consistently demonstrated the most openness when it comes to model training, datasets, and research; even before the LLM era (e.g. StyleGAN).
There’s also modelscope.cn (china’s huggingface) which is worth checking out. I would not be surprised if one day, we have to use China VPNs to download open weight models.
Of course they are. They're commoditizing their complement.
I want to own the hardware, not play around in an nvidia fiefdom full of nvidia rules.
Nvidia probably likes that ASML is a monopolist (it's called a monopsony). The price is high and Nvidia can't scale as hard as they want to (more general: capital intensive market). This monopsony makes sure that other chip companies can't rapidly scale up and try to beat Nvidia. That there only ever was one other GPU firm (I'm prehistoric; once there were more) and they bungled it on software, is pretty sweet for Nvidia.
On the side of their customers. It would be best for Nvidia if there is free competition for the outputs generated from there GPUs. This maximizes consumer surplus and thus demand. Maximum demand for tokens, is maximal demand feeded in their monopoly. If there is a monopoly right from you, demand is curtailed, and your value is limited.
And that exactly is why you see interest from token generators for chips. Bridge that moat and gain a larger value surplus. Both NVidia and, say, China actively undercutting the token-supplier value chain is quite interesting to watch. It's like the Opium wars with us as somewhat happy customers.
In this same vein, why isn't ASML raising thousands of billions for building their own (subsidiary) chip foundries, while raising prices and starving the market (a little) for their machines.
You're looking this purely through an economic lens, while in reality geopolitical factors play a huge role in what ASML can and cannot do. The US government would likely take an extremely dim view of any new external competitor popping up for their chip foundry industry (especially with all the new US plants being built or planned) and would lean heavily on their vassal/ally the Netherlands to prevent this. Unlike with China, the US has more leverage over the Netherlands[1]
The US security state and US tech giants are joined at the hip, as they have been since the beginning of Silicon Valley[1], right through the Snowden revelations through to the present day[2].
[1] https://nltimes.nl/2026/08/20/us-preparing-force-netherlands...
[2] https://www.brennancenter.org/our-work/research-reports/sect...
> Of course they are. They're commoditizing their complement
Then why don't they sell consumer GPUs with tons of memory. They clearly segment the market into consumer versus server/business.
There are a few problems though, primarily, a GPU with lots of VRAM and very high bandwidth is inherently very expensive (on top of which there is also the CUDA premium); AI use cases are better served by SoCs with lower (but still high) bandwidth and more RAM.
[0] - https://www.nvidia.com/en-us/products/rtx-spark/
If you don't want to use CUDA, they expose the PTX bindings to write your own CUDA alternative too: https://docs.nvidia.com/cuda/parallel-thread-execution/index...
Also, when running OpenCL, NVIDIA hardware disables multiple DMA engines, and allows only one memory transfer at a time to prevent OpenCL running as fast as CUDA.
Did NVIDIA finally allow open source drivers to access all parts and features of the card to allow feature parity? Last time I checked they were considering a plan for planning a solution to that.
Having said that, I'll try compiling a OpenCL 3.0 program in the cluster, so I can report whether NVIDIA runs this software, and if yes, how well.
Likewise SYSCL although built on top of OpenCL 3.0 primitives, is mostly Intel, which also owns CodePlay, the company that delivered the first working SYSCL compute experience, again neither AMD nor Intel (until it bought CodePlay).
I don't think bits like the GSP firmware will ever be open-sourced, but the opportunity to write better OpenCL drivers has always existed. Some of Nvidia's other proprietary driver backends (eg. GBM, Vulkan) are also decently neglected, but mostly out of disuse rather than malice. I don't think any of these things mean you don't own the hardware.
Also if we were just discussing labs, Ai2 opens ~everything with dramatically less resources than Nvidia.
Nvidia’s history with linux shows the opposite. And as a user running models on a linux/AMD stack, this information does not fill me with hope.
They might be right w.r.t. openness about LLM at the moment, but w.r.t. general software openness they are definitely the opposite of open.
A few years ago everyone said _Open_AI is the the most open labs. How did that turn out? Lots of coy, deceiving actions till the whole company was turned into whatever rent seeking amoral borg adjacent shell of it's former past it is now.
It’s a little like the trend of calling developers “engineers”. There’s no actual engineering in the traditional sense but I’m sure developers think it sounds cool to call themselves that.
My point was just that I just find it amusing that people call themselves engineers when the code they produce is so far removed from the level of rigour one would expect in literally any other engineering industry.
I say this as someone who also has family and friends who are actual engineers, if they built bridges and buildings to the same standards that many developers write code, then people would die.
This isn’t meant as a criticism of developers, by the way. Just an observation at the vast differences in the domains and thus the tolerance for errors in the process.
Likewise for labs. I’ve worked in AI startups and the science departments are not something one would think of when you say “laboratory”. I get why the term is used, but it’s still amusing.
I know language isn’t static. It’s something that evolves, like how a “computer” used to refer to a person rather than a thing, but that doesn’t stop me from being amused. But maybe the real issue here is I take myself less seriously than others so I have that capacity to be amused by the titles I’ve held?
Just like everything AI is a "model". It's actually not, but it sounds cool/sciency.
It sounds cool because people in our age are almost obsessed with scientistic performance.
At least AI labs are actually doing experiments.
Sure, the models are open weight, but porting the code needed to run them on non-Nvidia hardware is not trivial.
Dude, where's the src for GPU drivers and the firmware blobs?
Sorry, you can't say it is one of the most open labs without qualifying a proper response to the question above.
Nvidia is also one of the most closed hardware developers around. Two things can be true.
But it is a good example of people talking past one another and not communicating well, I would think.
No shit Sherlock. Name one company that shared their code to make you NOT to consume their stuff?
They supported OpenCL when Khronos floated the idea of a GPGPU standard to manufacturers, but OEMs didn't want to design scalable hardware or sponsor the software.
Evidently we should, because Linus has been more positive about Nvidia in the last 2 years [0]. I've been using the open driver for years now, for both gaming and CUDA.
[0] https://binarymusings.org/posts/talks/linus-on-ai-linux-in-k...
> This is actually one of the benefits brought by AI; it has made Nvidia a good participant in the Linux kernel space. [...] Now, when Linux is so important for AI clouds, Nvidia suddenly cares very much about Linux.
Now Nvidia is also premium sponsor of the Linux Foundation.
Nvidia is a big company. They are good about some things and bad about others.
I think they really do like open weights because they make some of the best hardware for training, and the more open weights models there are, the more people are training and fine-tuning them, mostly on Nvidia hardware.
I feel like Nvidia is one of the better choices for buying Huggingface. Not perfect, but definitely far from the worst.
The worry is that Nvidia is trying to control the way you run those models. Trying to bake CUDA assumptions into model design, and pushing the software ecosystem to be as Nvidia first as they can.
1. Oracle
2. A16Z
3. GameStop
2. Microsoft
3. Google
Are probably the worst of the realistic options for acquiring HuggingFace
Their strategy for AI is pretty clear and they bank on on-premises OSS models for the busines, with their really cool open-source software: https://www.youtube.com/watch?v=tmcn1-jFLWY
This is a really good watch and is a glimpse into what the actual future shapes up to be considering the current situation in where the OSS Chinese models successfully compete with proprietary US ones.
The only thing I can see them being able to get away with is increasingly bending the hugging face python API and any other features of that sort they develop to NVIDIA only. I personally don't use that and don't see a reason to and I am not sure how many people do use the hugging face python library.
They need to take a machete to all the cross coupling they’ve metastasized.
Nvidia wants people and companies to go choose a free open model, run that model on Nvidia hardware. And because Nvidia can't fully control what hardware an AI model can run, Apple Silicon and AMD hardware users will benefit as well.
Open weights correspond to "binary available" for software. Nobody would call that "open source".
(Even ignoring the licensing which "open source" normally entails.)
Though we agree that the term isn't typically used that way in the LLM context, right?
My impression I've developed in the years of working there is that Nvidia's relationship with opensource is... not intentional. They kinda suck at it because they genuinely don't know how to do it more than they want to make money out of it.
Here's an anecdotal "success story" which is also an illustration to how things might not work out well otherwise.
So, I was on the team that deals with server infrastructure. One day we get a new "feature" which was supposed to allow Slurm (the workload manager, a kind of software used to run "jobs", including eg. model training) to be deployed with distributed MySQL as a backend. The feature is all obviously written by a single developer with an enormous amount of "help" from AI. I was tasked with testing it.
Trying to figure out what it does... I realized that the "distributed" part of the feature was to be achieved by integrating with Oracle's MySQL by means of using MySQLShell (another proprietary Oracle's product). Until that point, by default, we integrated with MariaDB. Not only was it using Oracle's proprietary tool, the tool, actually, didn't support the "distributed" part of the "solution". It was pitched as the "first step on the way there".
So, I was able to push back on it, mentioning Galera, arguing that the "solution" doesn't solve the problem and will require from customers to change databases (even if they are mostly compatible... they never quite 100% compatible). And the misfeature was rolled back.
I made an effort to investigate how did we even get there, and turned out that whoever authored the "solution" had an experience of working with Oracle products, but never really tried the open-source ones. So, he didn't do a research. He just used what he knew.
Unfortunately, this is a rare win, where the evidence of disadvantages of using proprietary solution was huge and enough to turn the tide. But often it doesn't face any resistance because nobody is even aware of the problem.
I do think that they still might be a bit reticent to release fully open-source drivers, though, only because of the extent to which hardware design could be inferred from the driver source. OTOH, AI itself is making disassembling and reverse engineering binary code easier and easier, so there might not be much of a point to withholding the source in the near future.
They're trying to mix up the competitive landscape(that doesn't impact their bottom line, and I don't think opensource is eating their lunch), so I don't think this is fake, at least that's my initial take.
100%
Modular on the other hand creates the Mojo compiler gets criticised for not open sourcing it immediately and now once they do, no-one cares anymore.
Huggingface was not just a target for open source, but as a force to have open weight models run better on Nvidia against the rest.
this is the crux - if nvidia makes it so that open weights end up running better on nvidia hardware than competitor's, then it's going to prevent hardware innovation and competitiveness in the entire sector.
It's like as tho General Motors buys out oil refinery to make gas for all, but the gas somehow runs smoother in GM cars.
NVidia is delivering commodity hardware to the hyperscalers, and would eventually get commodity margins (when hyperscalers make their models work on their own hardware).
Amazing article on relevant economics of squeezing vendors - actually about antitrust ad-models but:
https://www.thediff.co/archive/ad-supported-platforms-are-a-...Perhaps not relevant to huggingface, sorry.
Couldnt they just take those models anyway?
Whereas mojo is a general purpose language, and we're absolutely spoiled for choice on modern languages with open source compilers.
I'm not saying it's fair or right, I still think mojo is neat, but isn't exactly comparing apples to apples.
Nvidia on the other hand has not and the best they have done is a bunch of closed-source blobs which they do more closed source releases than the rest.
Mojo is open source and targets all GPU architectures for their compiler regardless of the vendor and nvcc targets their own (and both that and CUDA are closed source).
So this is directly an apples to apples comparison.
AMD is another story. Their binary closed source blobs were even worse than nvidias for many years. So much so people used terrible performance open source tries just to avoid the headache. That they finally slopped together an open source version after they'd lost the GPU race isn't exactly noble. But as an open source supporter in general, I commend the effort still.
possibly because it took qualcomm buying them to make that happen.
Mojo was partially open source before Qualcomm bought them, and they were going to do open source it anyway.
Was NVCC or CUDA ever open source since the lifetime of its development?
uh huh.
> Was NVCC or CUDA ever open source since the lifetime of its development?
you ever ask Nvidia why? i did.
Exactly. You're one of those that don't care.
Mojo's standard library was open sourced a year before the acquisition.
> you ever ask Nvidia why? i did.
The correct answer to my question is "No". CUDA or NVCC was never open sourced during its lifetime.
Since you "claim" to have asked them and know the true reason why is not open source, just say what Nvidia told you verbatim right here.
Intel and AMD have no problem open sourcing their GPU drivers for Linux, but Nvidia has a problem with that.
We can only speculate that "open sourcing" it reveals their GPU trade secrets and their intellectual property for others to copy.
If it was going to "eventually" be made open source, why not do it from the start? Why not do it a year in?
From what I was told, your speculation is incorrect. I can't say anything more than that, sorry.
I don't know why you're so aggressive in tone towards me, chill out.
> They want
[citation needed]
Or at least, $13b to stay at the head of the race (or keep the race running) must be worth it to someone's desk.
There's only $50b in datacenter buildout nationally (Source: Gemini, 2026).
So it is a bit of a puzzling choice for what amounts to a pile of software, in my opinion. but I don't know shit.
But more seriously, this is my first time seeing that as well, and I'm not sure I like it. Citing an LLM is a little like citing Wikipedia to me, you cite the primary source the LLM is quoting directly, not the secondary source.
It may as well say "(Socrates, probably)"
Your reference lacks authority, veracity, and reproducibility.
In fact, has HF ever wished to make money? They probably pay AWS more infra cost than they earn. The exit was planned all along.
Just now I got my hands wet with local LLMs before this was announced. Can somebody from llama.cpp/ggml.ai confirm this?
Very optimistic to say this will not have profound effects on the whole industry.
Why can't everyone on it just move elesehwere?
Guess the answer is scale, same reason folks don't leave twitter/x.
The optimal market strategy there (as in a lot of places) wasn't "sell as much as you can". There's often a superior strategy, when (as with HPC) you have minority industry customers who are very rich and have low price sensitivity. It's to raise the price to what those special customers are willing to pay, and to drop everyone else.
What NVIDIA did was to rip out FP64 capability, systematically, from all of their consumer cards. They firewalled off "useful for GPGPU" as a differentiating feature, segmented the market, and astronomically raised the price of what (if you were looking soley at cost-to-manufacture) could have been easily affordable to any ramen student.
(It's a more obscure version of the Intel-made-ECC-memory-disappear story).
See, e.g.
https://news.ycombinator.com/item?id=47068890 ("15 years of FP64 segmentation, and why the Blackwell Ultra breaks the pattern (nicolasdickenmann.com)")
True about Intel and ECC, but AMD now does similar things, even with their consumer CPUs and chipsets.
These three companies now make very sure that consumer products can never canibalize those juicy data center profits - so they make sure to limit what the consumer segment can do.
VCs could not see any other reason to raise more money and Huggingface was not growing as fast as they thought to justify the valuation or the next fundraise.
So they might as well get Nvidia to save them from the VCs pressurizing them.
In the case of GitHub, it was likely for data reasons + wanting to own where developers do work (VScode + Github).
In the case of HuggingFace, honestly not sure as I'm not familiar enough with their business. But I can assure you that Nvidia didn't buy them for 13 billion cause HuggingFace were desperate. When you're desperate, you sell for less not more.
From Nvidia's side? You get to keep the shell game of where your money and hardware are going spinning on the table for a little longer. If the party stops, Nvidia loses a zero right off their valuation instantly.
And, as a side benefit, you get to place your thumb on the scale of the open-weight hosting ecosystem. And maybe even fund a Chinese Anthropic or OpenAI at a discount.
OpenAI just popped out an inference ASIC. Google is on their 8th generation of TPU. Graviton is out from Amazon. The hosting companies want Nvidia out of their finances. Full stop. Nvidia has to do something, or it's going to get swept away.
https://www.youtube.com/watch?v=NufJ7g63KSY
In this case, Nvidia wants the easiest route from:
find model -> adapt model -> optimise model -> run model
to terminate inside the Nvidia stack. This is a boon for DGX cloud.
> Here have some of my monopoly money I can print and come join us at Nvidia!
Nvidia don't share CUDA, don't open source their drivers, don't support capable but older hardware (forget Pascal etc), and generally charge a premium over competitors for hardware.
This acquisition will likely solidify this general stance, reduce free compute allowance for a subscription, cap downloads, push advertising, generally favour models that are Nvidia prescribed or have been sponsored, and potentially ban models and datasets that are deemed risky legally (abliterated etc).
I see no other reason why Nvidia would want this kind of vertical.
Closed Core: Driver, compiler (nvcc), runtime, assembler (PTX/SASS). Closed Libraries: cuDNN, cuBLAS, OptiX.
Open Libraries: CCCL (Thrust, CUB, libcudacxx), CV-CUDA (Vision), CUDA-Q (Quantum). Open Tooling: CUDA Python bindings, NVIDIA Kernel Module (Linux driver outer layer).
Nvidia is working with Red Hat to develop the NOVA driver and with Valve to develop NVK, which will be the future open-source drivers.
Nvidia supported the Pascal architecture for 10 years. No other graphics company did that. AMD recently made the RDNA 2 architecture legacy after only 5 years.
Nvidia is that most promotes open models and contributes significantly through Nemotron and various hardware-level optimizers. I think the value has more to do with the potential dominance of the open model ecosystem.
https://x.com/ctnzr/status/2072715518373777812
That said, I share all your concerns. The days of a permissive hands-off HF may be numbered
Commoditize your complement.
They care about their high margin GPUs being the dominant platform, otherwise they would have to reduce the price of their gpu/increase vram amounts.
Nvidia are making truck loads of money and want it to continue
Now the Open-Models crowd will try to move to some other place. But fragmentation will weaken the position. Yes, there is Civic, there are purely Chinese websites ... for those who speaks Mandarin. Which only reinforces the point.
Apple's devices have resulted in a great increase in quality of life for myself and family members who used to waste a lot of time with Windows drivers/BSODs/random updates/malware/etc.
It's been almost 20 years, and it's turned out OK.
works fine for now
whats nvidia gonna do except make it worse?
usually buyouts go something like this:
1 buy company
2 fire various people
3 enshittify
hugging face even said they didnt want to accept a 500 million dollar investment from nvidia, because they didnt want nvidia to run the ship.
instead they sell it... guess who'll run the ship?
https://techcrunch.com/2026/08/24/hugging-face-reportedly-in...
If Anthropic and OpenAI makes their own inference hardware, this is a shot across the bow for making their business worthless.
Expect the next NEMOTRON to top the AI performance indices.
1) Local inference means more general GPUs and less ASIC hardware. Only big companies can push ASIC because of the software required, meanwhile nVidia owns cuda which is the standard.
2) Local is less resource-efficient per chip (chips remaining idle much more, meaning more chips required).
3) End-customers have less bargaining power compared to hyper-scalers. Although this might change if customer hardware start behaving more like phones (SoC with everything packed in), but even then the SoC makers will likely still have less bargaining power than hyper-scalers. But then nVidia could potentially make the whole SoC too.
So overall local-inference users = higher profit margins for nvidia. They much rather have every business on the globe buy one nvidia rack (or every laptop have a beefy GPU) than have 5-10 hyperscalers buy a few hundred thousand.
But the spending spree is probably coming to an end with the looming IPO's and NVidia is probably trying to hedge their bets by making themselves the sure bet once big-AI stops monopolizing RAM and everyone races to get their local setups.
So instead they sold themselves to the same investor completely ?
Time to see how Jensen plays it.
This can only be bad news, Nvidia didn’t buy hugging face to be good custodians but for business reasons.
Let’s hope an alternative platform emerges and takes off.
Just a little bit longer and this whole "AI" insanity might finally be over! :)
The title is not the linked article’s title. It is the title of the first article in the description.
FWIW, “agrees” is a weird verb. Kinda sounds like they are doing this out of kindness and not their own interests.
Surely anyone can take any piece of CUDA code and tell some LLM to port it to another platform and keep grinding till performance is identical?
THIRTEEN BILLIONS for ONE THING.
BTW Instagram raised $50 million at the valuation of $500M, just 4 days before the Facebook acquisition. %100 return of investment in 4 days, I think that's insane.
The amounts are incomprehensible at this point. I don't think anyone can conceive of 13 billion dollars accurately, including the people brokering this deal, they're just thinking of bargaining chips and weighing them in comparison to how many they have.
A US billion is 1 000 000 000, a French billion is 1 000 000 000 000.
(French as in francophone, not exclusively from France)
Nvidia is investing $1 billion in Poolside and paying $6 billion to license its technology and hire most of its engineers.
They are the elephant in the room, with their CDS sky-high. Sounds like they are opinionated on the whole AI thing, trying to drive this with open-source models, countering OpenAI using Jalapeno.
Looks zero-sum for the players.
All the while Google silently planning to get milk from all layers.
I'm old enough to remember.
AMD could’ve done the same.
They worked with studios to optimize games for their hardware. I don’t think they went around sabotaging other companies hardware. They did what was good for them.
Why did competition not do the same?
No gamer had a problem with nvidia.
Many bought their mobos and video cards.
Mac was pretty hated, Steam was very hated, we didn’t bow to any software company.
Nvidia actually made hardware.
Your propaganda is so awful because it’s always obvious you have no experience
Everything I can find online is referring to this one source. That doesn't tell me if it's happening or not.
Thoughts on WebTorrent?
What should we do?
But Nvidia is a terrible open source and consumer company. They gatekeep a lot and oftentimes it's only open source in name. Outside contributions are often slow-walked or rejected if they don't align with business incentives , and leadership is retained 100% in a couple of people from a certain country.
His other comments on Nvidia often contain expletives.
No, its not just a repo/index (they also have training, inference hosting, and they develop/maintain a bunch of core AI infrastructure software), and even if it was just a repo/index, replacing a bug centralized repo/index that used by an large community isn’t trivial.
As battle lines get drawn over duopoly vs. open weight it’ll be interesting to see what Nvidia does. They definitely want a piece of more of the stack especially as Huawei chips become more and more of an alternative to cuda.
Hyperscalers and hardware manufacturers will absorb the overgrowth and hopefully (for them, sucks for us) settle at the top of the software value chain. The only way to make all the money poured into compute for the last 4 years pay for itself at any level is selling people and corporations on very inefficient software for the next couple of decades.
Not too different from what "the cloud" did to this market from 2015 or so...
I mean, haven't you just described VC?
I mean for all the inefficiencies of the cloud, it solved a lot of problems. It also created a new category of problems but rarely does any particular technology arrive as a pure boon without it's own compromises. Despite it's issues and despite how hot ASS some implementations are (looking at you Azure) I think cloud infrastructure you can at least say does solve a lot of problems for a lot of firms. I can't remotely say the same for AI.
It has a lot of uses don't get me wrong, but not nearly the number being pitched currently, and I don't know that I've seen any that I truly can't live without. And at the prices that are going to need to be charged when we turn the proverbial profit lever... I mean I'm sure tons of huge software houses where shipping is all that matters, behind things like quality, maintainability, the sanity of the development team, what have you, AI code assistants will have a lot of uses and the company will (probably) pay for it. But I think it's a stretch to say it's even remotely as good a trade as Cloud infra was.
In the handful of years since we've gotten all these supposedly revolutionary tools, software quality has already taken a gut-punch, will no sign of that slowing down, quite the opposite. And that's not even going into all the AI features being crammed into everything that few, if any users want or use. In fact the obsession with shoving AI shit into Photoshop instead of fixing any of the issues that've been present in that software since ~2009 was the straw that broke the camel's back for me and I've left Adobe behind entirely, and I know for certain based on communities I'm in, I'm far from alone in that.
The goal is to increase return on capital.
Take a look at the historical developments of wage participation in global income, labor productivity growth vs wage growth, and so forth. It works very well. As long as the most capable laborers are paid well enough to go along, the rest of us have little choice. Especially after citizens united basically privatized American elections.
No one seems to get that you have to do a general strike BEFORE they lay everyone off. It's a gambit. But either you actually were all that valuable and you can get your job back if it fails, or you weren't and you finally acted in your own self interest.
We can't discuss the finance of private companies so we have to wait for their IPO, acquisition by some public company, or bankruptcy to find out more.
The story written in the public numbers, from corporations or government statistics, is one of rise in factor productivity, rise in capital returns, and decline in wage participation in overall income.
I'm merely observing that it seems to be the latest chapter in the development of the capitalistic corporation.
Workers lose their ability to earn a living, but OpenAI and Anthropic become the gatekeepers to effectively all productivity.
I know I sound tinfoil hat here but I don't see how else the numbers work. The only thing full-priced, profit-earning AI can possibly be cheaper than is, in my mind: as much of the workforce's salaries and cost of benefits as it can manage to offset, put together.
It's a great plan if you're a CEO who can't see beyond the next quarter's earnings call. Less so if you want things like, just pulling from a hat here, a functioning economy, a stable society, the ability for humans at large to live. But it's not like the owning class hasn't been undermining all of that for the better part of 50 years anyway.
VCs invest, no one can compete and those with the moats survive.
>The companies have not yet reached a deal, and the talks could still fall apart, the person said. Business Insider on Sunday was the first to report that Hugging Face was fielding takeover interest.
Nvidia has a strong interest in open source AI being a strong contender to proprietary models. They don't want a single AI company to win, as that company would then have enormous leverage over Nvidia, and could perhaps even erode the Cuda moat with their own chip spending.
They want the AI market to be hypercompetitive, with AI companies focusing on competing amongst themselves instead of with Nvidia on one side, and many self-hosters / small inference providers buying Nvidia GPUs for large open models on the other.
It's in Nvidia's business interest to be a good steward of Huggingface, while being a good steward of Github is at best incidental to the interests of Microsoft.
Linus would heavily disagree!
source: https://diginomica.com/kubecon-china-33-and-third-linux-long...
And I'm not so sure that Mac is out of the equation
And Nvidia hardware was running on Linux before AI as well...
> And I'm not so sure that Mac is out of the equation
Mac servers exist but anyone serious about running at scale isn't using them for a number of reasons, not least of which is that Apple simply doesn't make real servers anymore...
That could change.
Who says they don’t switch to hardware manufacturing for AI servers
But more importantly for business, if HF fails to use the upcoming ASIC inference chips for their ZeroGPU after the acquisition, their competitors using them might have an advantage.
nVidia driver sets quite low number for VkPhysicalDeviceLimits::maxTexelBufferElements. I don’t think that’s a hardware limit because I have D3D12 backend doing the same thing on the same hardware.
Vulkan performance is not great on nVidia. On all AMD cards I am testing, Vulkan 1.3 is the fastest backend. On nVidia however, D3D12 is faster despite tensor cores (WMMA / wave matrix multiply accumulate / cooperative matrices) are only available through Vulkan, D3D12 is using shader cores exclusively.
This is just one company. If other hardware is better and they don’t provide access to it, won’t people just leave HF??
Is this somehow more of a monopoly than I am imagining?
The acquisition sets up corporate interests to be able to increasingly drive it.
I hope better for Nvidia and hopefully it can be cleared up, and maintained.
They will be as free as the Washington Post after the aquisition by Bezos.
Wait and see is the only thing we can do.
Why is it alarming for a shovel salesman to also sell plants?
https://en.wikipedia.org/wiki/Vertical_integration
https://en.wikipedia.org/wiki/Horizontal_integration
Anything else you care to throw at the wall to see if it sticks, just go ahead and assume I disagree with you given how our interaction has gone so far.
Always like that with Microsoft apologists
https://www.githubstatus.com/
That page shows 99.5%+ uptime across all 11 of the services provided.
To do that at a small company world cost significant money.
To have it integrated and cheap for all those small shops is very valuable to them.
and again don't know how this is acceptable in 2026, for a paid CI service
and I'm not talking about 6 9s, I mean, 99.5% is terrible
it's 2026, not 2016
and that 40 hr may be that over 40 days you just have to randomly stop working if you depend on GHA for something
and why are other services not struggling like this?
The copilot service and the training on open source would be the same if Microsoft hadn't bought GitHub, it would just be branded differently.
The argument is that AI models trained on open source should include the copyright info for all projects blended into the weights. And since they don’t, it violates the terms of the license.
The load increase from AI is definitely significant, but I think with proper investment they would not be having this many issues.
I feel like if Nvidia ends up turning into a bad actor, in terms of restricting/censoring models... another HF will spring up.
My primary concern is that Nvidia will bow to the pressure and restrict abliterated/uncensored models on HF.
Having Nvidia behind HF will now probably have a bunch of litigious entities salivating to sue them for not only allowing, but also distributing uncensored AI models that will then be used for deepfakes, porn, spam, scam, hacking, etc.
Sure that could have also been done to HF before, but I'm thinking Nvidia is a much juicier target.
Should also now be much easier for the government to just eventually force NVidia to restrict access to Chinese open models seeing as it will now become an American company that needs to obey to American law and we all know how American AI companies are lobbying for exactly that.
Using regulation to avoid having to be competitive is the American Way!
In all seriousness though, probably time to start an underground equivalent to HF.
The problem for them is that the leading provider of training and inference is actually AWS...
That's the anthropics and openais of the world.
We're definitely better at maths than at promotion.
https://openai.com/index/hugging-face-incident-and-the-road-...
Right now it is a pain to find the correct incantation.
Institutional holders mass revolted at spacex getting into the basket.
AI costs more than having people do the work. Q2 CFO reaction proved that.
I might host 3 or 4 models that I've downloaded recently, but I won't be hosting the 50 or so that I've tried in the last two years.
But if you don't pay a host to seed it, either the host's business model won't play well with torrents, or there's no host and a torrent will rot.
And now you've recreated most of HF
Because you can't host arbitrarily large files on github, it's basically become a defacto "publish your project here" thing.
For a CN domestic equivalent take a look at Modelscope.
It’s all about the convenience, minimising the time and effort from “looks interesting” to “running the model”.
Nvidia don’t care if you do it on their cloud, someone else’s cloud, or on your own machine - they win either way, as it further propagates the technology on which they are building their future.
I'm not a fan of big-tech acquisition results either, but one benefit can be that a product continues to exist when it would otherwise become insolvent.
But after the answers, I do think his example is poor. Models available at HF have many purposes and will run on many platforms. Commenters here are suggesting this will no longer be the case.
Whereas the example Workflow (iOS only) -> Shortcuts (iOS only), there was no change in that regard.
they basically own chips - data centers - discovery etc?
Hugging Face->Nvidia ($13B)
Keep an eye on Exa. With local models and niche harnesses proliferating, they're going to need a way to search without everyone having to run their own index.
Google is an obvious buyer if they could get it past regulators. A Doubleclick-level addition of the agentic stuff they're bad at to the indexing they're good at.
Nvidia is the obvious less conflicted #2
Secret third option: Cloudflare. They're already the glue for so much of this stuff.
Just in recent weeks Openrouter bought out by Stripe and now hugging face bought out by NVIDIA.
Is this really freedom and choice for consumers? No it’s the proliferation and nomination of big tech.
This acquisition should be blocked (and never will sadly), they now have both the incentive and the ability to influence a platform that's supposed to be hardware agnostic (and built trust over that) toward CUDA stack
Microsoft did the exact same with github, acquired stack agnostic platform, and turned it into a copilot/azure one
12-24 months from this acquisition will likely look like a crazy burn of capital and cash.
First they came for all the dev tooling - uv, Cursor, etc. Now the routers and providers - Open Router, Hugging Face...
Who or what is next? And what is the endgame I wonder??
Meanwhile, RAM prices still go up. I want my money back.
Nvidia is making tons of money by selling GPUs at a very high profit margin. They will do anything to remain dominant
This feels like a similar leap of faith. One that is hard to believe in. Thankfully, I think Nvidia can keep the lights on here & keep this going. I don't think they have to do much, per se. But it felt implausible then to image an Nvidia that gave a shit about anyone else, an Nvidia that actually gave a flying fuck about drivers or upstream Linux or ecosystems that weren't entirely within their own control.
Similarly the upper quartile of succes here feels mostly like benevolent neglect. I think we can hope for Nvidia to just not mess up a good thing, for them to understand that this open model open ai universe hinges upon Hugging Face, and for them to pretty please keep caring about the existential risk of the hyper-ai'ers all building their own properietary models on proprietary hardware and leaving Nvidia behind some day, and HF being the hedge against being left behind.
No doubt you have the nucleus of a substantive comment here, but that's not enough. If you only post the shallowest top stratum of what you're thinking, other people do the same, and then we get "Laws are for poor people" and endless descending repetition. The whole point of this site is to try for something other than that.
https://news.ycombinator.com/newsguidelines.html
p.s. Also, please don't be snarky on HN. That's also in the guidelines.
It's comparable to the biggest bank buying the biggest ratings agency.
It's different for their local counterparts, but the original companies?
Not just on this HF acquisition, but on the whole ecosystem monopoly.