FR version is available. Content is displayed in original English for accuracy.
Advertisement
Advertisement
⚡ Community Insights
Discussion Sentiment
71% Positive
Analyzed from 8127 words in the discussion.
Trending Topics
#nvidia#models#open#more#source#hardware#face#don#hugging#huggingface
Discussion (533 Comments)Read Original on HackerNews
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.
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...
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...
Why is it alarming for a shovel salesman to also sell plants?
Always like that with Microsoft apologists
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.
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.
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.
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.
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.
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?
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.
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.
(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...
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 is the article you are referring to is this but your facts are wrong: https://techcrunch.com/2026/08/24/hugging-face-reportedly-in...
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 :)".
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.
Just now I got my hands wet with local LLMs before this was announced. Can somebody from llama.cpp/ggml.ai confirm this?
–Some guys in every bubble I’ve witnessed.
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.
> 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
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.
Very optimistic to say this will not have profound effects on the whole industry.
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.
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.
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.
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.
good old torrents
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.
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.
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.
>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.
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.
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.
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.
Just a little bit longer and this whole "AI" insanity might finally be over! :)
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?
So instead they sold themselves to the same investor completely ?
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.
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.
https://openai.com/index/hugging-face-incident-and-the-road-...
We're definitely better at maths than at promotion.
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.
I feel like if Nvidia ends up turning into a bad actor, in terms of restricting/censoring models... another HF will spring up.
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.
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.
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.
My primary concern is that Nvidia will bow to the pressure and restrict abliterated/uncensored models on HF.
That's the anthropics and openais of the world.
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
they basically own chips - data centers - discovery etc?
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??
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.
Not just on this HF acquisition, but on the whole ecosystem monopoly.
It's different for their local counterparts, but the original companies?