Jensen's purchase of a new toy could reshape the entire AI industry

KETTLE AI repository Hugging Face will soon be an Nvidia property, and while everyone's promising things will remain the same we at The Register's Kettle desk aren't so sure the status quo is going to stay the status quo at the AI industry's leading open model host. 

You can listen to the latest episode of The Kettle right here on this page, as well as on Spotify, Apple Music, or YouTube. Those platforms also let you subscribe to Kettle, so you are always notified when the latest episode goes live.

Joining host Brandon Vigliarolo this week is reporter Dan Robinson, who broke the news of Nvidia's $12.9 billion acquisition for El Reg yesterday, weighs in on what he heard during the announcement call, while senior reporter Tom Claburn opines on what he sees coming. 

A lightly edited transcript is below:

Brandon (00:01)

Hello and welcome to another episode of The Register's Kettle podcast. I'm Reg reporter Brandon Vigliarolo, and if you've been keeping an eye on things the past few days, you probably already know what the big tech news story of the week is, and that's Nvidia's $12.9 billion acquisition of AI repository Hugging Face. With me to discuss this deal and what it might mean for the future of the AI sector for both companies and developers is our reporter Dan Robinson and senior reporter Tom Claburn. Thanks for coming on, guys.

Tom (00:31)

Thank you.

Brandon (00:31)

All right, so Dan, you broke this story, I think it was just yesterday now, and you were on the conference call that Nvidia and Hugging Face leadership held. So break it down for us. What was the agreement here? What's the deal?

Dan (00:45)

Well, like you said, it was a $12.9 billion purchase from Nvidia to acquire Hugging Face, which is the home of open AI, as I've been told.

Brandon (00:59)

Open source AI. 

Dan (01:01)

Well, open.

Brandon (01:03)

Open source and open weight. I just want to make sure it's not the home of OpenAI, the company, right?

Dan (01:08)

Yes, let's not get confused about that. They've entered into this agreement which is due to close next year, providing that they meet all the regulatory requirements, of course, and there are no objections from the authorities on that. It looks like Nvidia's agreed to keep a hands-off approach and allow Hugging Face to pretty much just do what they're already doing.

But as we've been seeing, there are quite a few concerns around that, as Nvidia is a very powerful company in the AI market. And Hugging Face is the go-to place for open models.

Brandon (02:15)

I'm sure most of our listeners know what Hugging Face is; it's essentially the GitHub for the AI space. It hosts models, and new open models tend to show up on Hugging Face first. There's also a pretty active community that discusses the development of these things on the platform.

Dan (02:36)

Right, yes sure.

Brandon (02:38)

So this is obviously a pretty consequential move in this space and I know both Nvidia and Hugging Face framed it as giving Hugging Face more resources to pursue its mission. I think Clement, the CEO of Hugging Face, said that they want to reach a hundred million developers in the next few years and that they were worried that they wouldn't be able to without more resources.

Let's be realistic. They're hosting a lot of very large AI models and all the data sets they contain. That's a lot of storage space and a lot of computing resources, not to mention serving and dealing with all the workloads that come with this. I imagine that could eat up a lot of cash and resources.

Dan (03:25)

Of course. And it's quite an ambitious goal given that they're at about 80 million users, I think it's estimated at the moment. And if you compare them with someone like GitHub, I think they are somewhere around 180 million since they've been taken over by Microsoft.

Brandon (03:50)

And we've seen how well that's worked in the past few months with the stability issues and everything. It kind of begs the question too whether or not, if you get that big, can you remain stable and would Hugging Face end up suffering from some of the same problems that GitHub does?

Tom (04:05)

I wonder how long Hugging Face is going to continue to be flooded with new models because the models are appearing daily at Hugging Face. But what are the economics for making them? A lot of these companies are presumably not generating the revenue to pay for the cost of these models. And at some point, what are the economics of just giving away these models? Nvidia I think clearly sees that there's a future in the enterprise business where companies are going to want to control their own AI and run it. But is the pace of model development going to continue on into the future because where does all the money come from to pay for this?

Brandon (04:55)

Well, probably Nvidia now, right?

Tom (04:58)

And it makes you wonder if this is part of the circular financing that we've seen throughout the industry. It raises a lot of questions. I was also surprised that one thing we haven't seen, which was talked about, was Nvidia purchasing a stake in Perplexity, which has been reported, but as far as I know hasn't actually happened. They were said to be looking around at investing some large sum in Perplexity. That hasn't been publicly announced yet, and I wonder if this deal changes that discussion.

Brandon (05:33)

Perplexity, just remind me, do they make models?

Tom (05:37)

Well, they started off as AI search and are now kind of an AI agent platform.

Brandon (05:41)

Right. Okay.

Tom (05:44)

And I think that they've been apparently doing pretty well, but all of these companies I think would like to have an exit before they can demonstrate that they have general market revenue.

Brandon (05:57)

Right. I mean, obviously Nvidia is probably well positioned to acquire more people in the AI space. But that again raises some antitrust concerns. I know Tobias, our systems editor, Tobias Mann, wrote an op-ed arguing that he basically thinks this is a ripe opportunity for antitrust regulators to step in and deal with this, right? Because he thinks that no matter what Nvidia claims, and no matter what Hugging Face thinks, the impartiality and the openness of the platform might not last given the situation and Nvidia's control of the AI space right now. What do you guys think about the potential conflicts of interest here?

Tom (06:40)

I think there's a huge potential conflict of interest. I think that the willingness of the administration to do any antitrust enforcement has been demonstrated, which is there's not very much willingness.

Brandon (06:50)

Right. And we saw that this week with Google, right? The government basically gave up on trying to break up Google and said that they can just continue operating as is with their ad tech.

Tom (07:02)

Can you point to the last time in the past twenty years there's been a substantive penalty that's changed the behavior of a large tech company? Even Meta, in the social media lawsuit, seventeen billion is a small percentage of its actual revenue.

Brandon (07:20)

Not only just seventeen billion, but seventeen billion over ten years, right? Seventeen billion is a substantial chunk of quarterly earnings, I think, but not when spread out over ten years. And even then, like thirty percent of that penalty, they're refusing to pay out unless YouTube and TikTok voluntarily conform to similar guidelines on age gating and restricting youth profiles that they do. So you can essentially say that deal is worth even less than they're reporting.

Tom (07:54)

Right. I haven't seen a substantive case of regulatory intervention, at least in the US. I think Europe is a little bit more assertive about this, but even they kind of dialed back their penalties.

Brandon (08:08)

And unfortunately, even though their founder might have a very French sounding name, Hugging Face is based in the United States, so I don't think that the EU is going to really have much of an antitrust investigation here. Dan, would the UK have any grounds to do any kind of investigation on this deal, or is this going to probably be purely US?

Dan (08:30)

I don't think so. It's difficult to say. They're both US companies really.

Brandon (08:35)

Right. Right.

Dan (08:37)

Obviously it could have worldwide implications. But I'm not sure that any other countries have any kind of regulatory claim on it.

Brandon (08:55)

So unfortunately it's probably going to fall to the Trump administration or whoever comes next. And even when Lina Khan was at the FTC and was doing a lot of antitrust stuff to try targeting big tech, a lot of that amounted to nothing, right? A lot of her big cases ended up again, like Tom said, in the past twenty years has there been any substantive ruling that affected a big tech company? And she was going after Amazon hard, and that amounted ultimately to not too much.

If we look at the antitrust landscape and say it's unlikely that this deal is going to be stopped, what is the likelihood you think that Nvidia honors its commitment to remain hands-off?

Dan (09:41)

Like Tobias said, the temptation is going to be pretty huge to have some of the models first support Nvidia hardware or to have better support or documentation on Nvidia. Even if you look at Microsoft and GitHub, you could perhaps argue that it's worked reasonably well apart from some hiccups. I don't know if all the developers would agree with that, but even there, Microsoft tried to infuse Copilot into it, which didn't go down well at all.

I do agree that the temptation to interfere is going to be too great for Nvidia and perhaps there should be some kind of antitrust investigation but we seem to be all of the mind that that probably isn't going to happen.

Tom (10:55)

I think it's likely that this will end up pushing other companies to develop their own repositories just to have a supply of models, and you already kind of see that with Apple. They have their own sort of MLX framework and it's probably not too far a stretch to imagine them pushing a set of optimized MLX models or things that work really well on their Metal hardware, and maybe AMD ... 

Brandon (11:21)

Nvidia, you mean?

Tom (11:22)

Apple.

Brandon (11:25)

Apple. Okay, I see what you're saying.

Tom (11:26)

And then AMD maybe does something similar. It would make sense for them strategically as a defensive measure

Brandon (11:36)

That feels like a net loss for the AI community, right? Hugging Face right now is kind of like the AI Switzerland; it's a platform that allows any and all comers to host their models on there, let people pick them apart, let people try them out. In the same way that a fracturing of GitHub would be an unfortunate loss, I feel like. You can argue that there's a centralized platform and that increases the chance of failure or loss of data.

If everyone decides to go their own direction with this, I wonder what kind of net loss that would have for the AI development space.

Tom (12:16)

Everyone is going to be looking for a sort of edge, and whether you find that in terms of the model distribution or in the harness or somewhere else in the technology stack, the open weight models have become commodified very quickly. So you're either going to have the Anthropics and OpenAIs of the world if they can sustain their financial positions selling the high-end stuff to government clients, and then a bunch of quickly commoditized models that are going to only have differentiation based on what can be added to them, like if they work really well with a particular harness or if they really are well suited for Nvidia's hardware.

Brandon (13:00)

Do you think that there's a chance that this gives Nvidia a more dominant position in the AI hardware space than it has right now?

Tom (13:08)

Once you have that feedback loop between software and hardware, it makes switching harder, which is why the other companies have to respond. They have to have a combined software-hardware stack to make the jump palatable.

Brandon (13:26)

Right. So basically what we're looking at is this could serve as a definitive moment in the commodification of open source and open weight AI models, essentially. Because you've got Nvidia basically staking its claim to it – this is all hypothetical, right? Because it's entirely possible that Nvidia could end up being like Microsoft and just kind of saying, "Yeah, we're just going to remain relatively hands off on this, except for providing the backbone for it, as failure-prone as that may be, and letting the community itself just function largely independent of us." Obviously, like you said, Dan, the Copilot situation, and Microsoft's made its changes to the platform over the years. But if Nvidia decides – and I agree with Tobias, this is a pretty big temptation for them now having direct financial control over everything that goes on at Hugging Face to basically prioritize their own stuff, to make this a platform that serves them first. I feel like you're right, Tom, this could end up fracturing the AI market into a whole bunch of different stacks where model A plays well with hardware A and model B plays well with hardware B. And you pick the stack you want to be part of.

Dan (14:48)

Yeah, and to be honest, people's perception will matter as much as anything. Even if Nvidia does manage to resist the temptation, if there's a perception that that's happening, people, users, developers will start to drift away. And with the rest of the open source world, people will start up their own alternatives and probably we'll get another Hugging Face perhaps.

Brandon (15:20)

Sure. Maybe one with a more sensible name. And the name Hugging Face has always kind of made me go, "Wait wut..." I know they started off as trying to appeal to teen users with some open source chatbots or something way back in 2016. Is that where they started? And then the name just kind of stuck even though they moved beyond that original mission?

Dan (15:40)

I don't know. It just makes me think of the movie Alien, I'm afraid.

Tom (15:47)

Yeah.

Brandon (15:47)

It does kind of, doesn't it? Not exactly a great idea, the thought of an AI –

Tom (15:51)

Someone took the wrong lesson from that movie.

Brandon (15:56)

– the AI grasping onto your face and shoving models down your throat.

Tom (15:59)

Yeah.

Brandon (16:02)

Well, speaking of the response to it or the perception, Dan, you were on the conference call and I know they were taking questions from people on there. What kind of questions were you hearing on that call? Was there a lot of concern? What was the general tone to people speaking?

Dan (16:22)

I think some of the questions were more about why Hugging Face decided to go for this because they had rejected an offer of investment from Nvidia just last year.

Brandon (16:38)

Right. I think it was a $500 million dollar offer.

Dan (16:43)

And the CEO couldn't really give a good reason why they suddenly changed their minds. His explanation that the planets aligned this time is a nice metaphor. Perhaps they did come around to thinking they need some kind of financial stability.

Brandon (17:11)

Yeah, I think like Tobias said, if Hugging Face was to collapse due to running out of funding, it would be even more devastating than a sale to Nvidia.

Dan (17:25)

What would happen to all their data and all the models if that happens?

Brandon (17:28)

Right. Right.

Dan (17:31)

I suppose you could take that angle, but somebody stopped just short of saying that it's the typical startup pathway to being acquired by a big company and that's their profitable exit.

Brandon (17:52)

$12.9 billion is a pretty hefty windfall for a company. I think you mentioned in your story that there was something like a hundred million reserved for staff from Hugging Face who are joining Nvidia, is that right? Or was that in Tobias's article?

Dan (18:11)

That was Tobias.

Brandon (18:14)

Okay.

Dan (18:15)

I don't think he mentioned a figure, but he said there was something like a billion dollars earmarked.

Brandon (18:20)

Yeah. I think he said a billion dollars earmarked for Hugging Face employees joining Nvidia. Like a startup seeking an exit here, this is a fantastic deal for anyone involved at that company.

Dan (18:36)

Right.

Brandon (18:36)

They're getting a major financial windfall, which can start to make all those proclamations about pledges of openness fall by the wayside when you're looking at all those zeros.

Dan (18:47)

Yes, you could say that. Maybe we shouldn't comment.

Brandon (18:53)

No, no. We won't.

Tom (18:54)

I think it's interesting that you have people making these sorts of deals rather than trying to exist as separate companies. Hugging Face is clearly not bringing in a lot of revenue on its own. And I think that there are a lot of companies in the AI space looking around going, "Okay, we're actually not making any money, how are we going to survive?" And Nvidia buying them seems to be the solution for a lot of problems. So we'll see how long that lasts.

Brandon (19:22)

Yeah.

Dan (19:24)

It does seem that Nvidia seems to crop up everywhere in AI. They seem to be the big investor behind an awful lot of the companies in AI. They've invested in both Anthropic and OpenAI. A lot of these neolouds that are popping up to offer GPU-as-a-service to anyone wanting to develop AI. It just seems like they're kind of like the kingpin behind the entire AI industry in a way.

Brandon (20:03)

They're the big player right now. That company is worth so much money and Jensen Huang seems willing to just open up his checkbook to anyone who seems to be moderately of value to Nvidia's bottom line. I think Tom hit it on the head; it goes back to all that circular investing. We're seeing this consolidation of the AI industry right now, and it's all consolidating around Nvidia.

Who would have thought twenty years ago when I was buying graphics cards for my gaming PC that, you know, Nvidia would be what they are now, but here we are.

All right, guys. Well, thanks for coming on, even if it seems like we're just hashing over the inevitability of the way the AI industry is changing. And we will be here week after week to talk about it right here on The Kettle.