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.
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.
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.
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.
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.
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.
They don't need a business model to have value here.
They sell a few cloud services, but the most valuable thing in their portfolio is live, engaged accounts from pretty much every person directly interested in running AI models.
If you were an org positioned to do sell better cloud services on a wildly better margin, wanted to promote consumer-binned AI hardware from your manufacturing chain, or wanted to promote new AI enterprise or research-support initiatives, those accounts would seem to be an absurdly fantastic collection of warm leads that you could gradually mine and upsell (and critically: that your competitors could no longer buy/use themselves).
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.
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 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.
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.
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.
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?
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.
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.
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.
Well congrats to Clem and the team. I remember when huggingface was doing things like coreference resolution models on spacy.
I hope nvidia does right by the community.
Edit to add: $13B should cover the S3 egress fees for a couple months :D
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.
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.
Yes. Great for devs, for a while. Awful for overall competition and industry's health.
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).
Except for during Biden, they've been asleep for almost 50 years.
6 replies →
> 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.
The era of VC-funded home-delivery recipe boxes is still my favourite.
Yeah this is the "early netflix" era. Take advantage while you can and plan for the future.
Remember just 6 months ago that "Ggml.ai joins Hugging Face to ensure the long-term progress of Local AI" (https://news.ycombinator.com/item?id=47088037)
(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...
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.
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.
What is the business model of hugging face? My understanding was they were basically just a file hosting platform.
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.
Didn’t we just witness it? Become a critical ecosystem player then get acquired based on the value of that position.
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.
6 replies →
Isn't this pretty much the model for any tech company? Rush to get MVP released, stoke the hype, cash in on FOMO.
They don't need a business model to have value here.
They sell a few cloud services, but the most valuable thing in their portfolio is live, engaged accounts from pretty much every person directly interested in running AI models.
If you were an org positioned to do sell better cloud services on a wildly better margin, wanted to promote consumer-binned AI hardware from your manufacturing chain, or wanted to promote new AI enterprise or research-support initiatives, those accounts would seem to be an absurdly fantastic collection of warm leads that you could gradually mine and upsell (and critically: that your competitors could no longer buy/use themselves).
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.
I see the opposite — NVIDIA is aiming to commoditize the model layer and push open models that are tuned to run on NVIDIA hardware.
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.
File hosting isn’t that hard to replicate though. Even if they start charging people will just set up torrents for the files or lists of mirrors.
Unless I’m missing something, this feels like Nvidia having more money than they know what to do with.
Maybe everyone will migrate to Hugging Bay for downloading models via torrent?
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.
ondemand cloud? b2b?
I have never seen an acquisition be good for the users. I struggle to believe this one will be different
Acquisition of GitHub let to increased downtimes, but also to unlimited private repos for the free tier.
The lowest hanging fruit that hf could offer is community monitored commands for running local models.
Right now it is a pain to find the correct incantation.
So is the next step to just basically create a cloud business.
they basically own chips - data centers - discovery etc?
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.
There are always torrents, which would seem to be a fine way to distribute models.
We also have modelscope.cn, I'm sure that one won't get acquired
The companies have aligned incentives so I don't think this is a terrible acquisition for customers.
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 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.
I wonder if OpenRouter being acquired was a catalyst for this deal.
I'm more worry than positive tbh.
I hope they're good stewards. This reminds me of the Microsoft GitHub acquisition in terms of the importance to the broader community.
As someone who've played video games and obsessed over graphics cards long before this all started, don't get your hopes up.
as gamers, we got burnt by nvidia way too many times. I'm still holding grudge over shutting down 3dfx.
2 replies →
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.
Handing out digging spots if you are the shovel manufacturer sounds reasonable.
1 reply →
Fix your title it's not finalized yet!
According to The Information, it seem it's finalized? https://www.theinformation.com/articles/nvidia-agrees-buy-op...
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 Torvalds: 'Nvidia has been the single worst company we've ever dealt with'
His other comments on Nvidia often contain expletives.
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.
In fairness, most of those comments were made before Nvidia capitulated in the EGLStream war and released open kernel modules.
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?
Not just a repo/index but they also offer reasonably-priced compute, with a free tier not to sniff at, on which to run the models.
Network effects. It's like a social network for LLM nerds as well as hosting and rating models, etc.
It also means that finding models that aren't "approved" will be dealt by other websites.
good or bad, it comes down to a financial decision. Nvidia was willing to offer the highest bidding price, and HF was willing to sell.
What does Linus’s hatred of NVidia drivers imply for this situation?
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.
That is a very good point. I will set aside some storage and download those models that I think are most at risk from a corporate clean-up action.
I don't think nvda has ever pushed the idea that ai is too powerful for regular people.
That's the anthropics and openais of the world.
s-1.site
What’s with all the big acquisitions recently? Is it advantageous for some reason to announce all of this stuff now?
The VC capital for AI is probably drying up. Now the question is: who will buy OpenAI and Anthropic?
The public
1 reply →
Huggingface sold out. Sad day. All the good companies lose their 'dont do evil' clause as soon as they make bank
You've sold out when you became VC Funded... It's just a matter of how long the free punch lasts.
To be fair, this is the best thing that can happen for the founders. I don't blame them.
What, are you expecting them to just work for good vibes?
I’ve asked ChatGPT etc a few times how Hugging Face makes money. Never quite got to understanding . Good for them.
Gemini did a good job of explaining it. fwiw, it may have helped that i accused Gemini of leading me to a "scam site" like "mega dot com"
You could just go to their website and look at the pricing pages? Why are you asking an LLM?
Oh jeez
Big Wolf buying the sheep farm. Nice
Its becoming impossible to track all the acquisitions.
Hold on, these say no deal has been reached yet.
I really wonder why Nvidia would do this.
Owning the roots, the trunk, the branches and the tip of the leafs.
Just like that they lost their ethos. AI oligarchy in the US is now cemented.
whoa
This is huge!
um um um um do we want nvidia to be on the same network as hugging face given the recent incidents?
the title of this post is misleading… “in talks” != not “agrees”
It’s not misleading according to the The Information piece (second link)
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.
[dead]
[dead]
Anyone old enough to remember anti trust laws or even just laws?
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.
https://news.ycombinator.com/newsguidelines.html
p.s. Also, please don't be snarky on HN. That's also in the guidelines.
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.
sometimes it is comments like these that make me wish hn had a block button
I mean you can easily write a browser plugin using AI to do that, there might already have been one.
what? why? seems pretty apt
The AI judge doesn’t have that feature yet
Laws are for poor people
[dead]
[dead]
tldr: the marketplace for opensource models became wide open