Hugging Face's CEO on Open Source AI, Model Routing, and the Future of Competition

As governments weigh new restrictions on frontier AI models, one question is becoming increasingly important: what role should open source play in the future of artificial intelligence? Theo Jaffee and Sofia Puccini speak with Hugging Face CEO Clément Delangue about AI regulation, open source safety, model routing, and why he believes competition—not consolidation—is essential for the industry's future. They discuss GPT-5, government oversight of frontier models, Hugging Face surpassing $100 million in annual recurring revenue, local AI, China's open-source ecosystem, Europe's AI ambitions, and why routing workloads across specialized models could fundamentally reshape where value is created in AI.

• 0:00Speaker 00

  I think distillation is a very common, you know, practice that everyone is using. It's something that everyone uses, but that is not, you know, the main reason for success. Like if you suck, you suck with or without distillation. It's hard for me to say like, oh, poor Entropic, poor OpenAI, you're getting unfairly competed with when you're like the fastest growing company in the world. If anything, I think they need more competition than less competition.

• 0:27Speaker 01

  Yeah.

• 0:27Speaker 00

  Because we're heading towards a world where a few companies are completely dominating, concentrating all power, all capabilities, all wealth.

• 0:39Speaker 00

  And that's much more dangerous than them maybe losing a couple of billion dollars of revenue.

• 0:47Speaker 02

  As AI on models become more powerful, governments are beginning to ask new questions about safety, regulation, and who should control access to frontier technology. In this episode, Theo Jaffe and Sophia Puccini sit down with Hugging Face co-founder and CEO Clement DeLonge to discuss why he believes open source AI is inherently safer, what Hugging Face reaching $100 million in annual recurring revenue says about the business of open source, and why the next phase of AI may be defined by model routing rather than a handful of dominant frontier labs. They also discuss GPT-5, AI regulation, local models, Europe's AI ecosystem, and the growing importance of competition across the AI stack.

• 1:35Speaker 03

  And we are back with Clement DeLonge from Hugging Face, his second time on MTS. Hugging Face is basically the open source AI platform. So Clem, welcome back to MTS. Absolutely.

• 1:50Speaker 01

  Yeah, so, well, go on. I think we're about to say the same thing.

• 1:53Speaker 03

  Yeah, we were, we were talking about this interesting recent piece of news where basically the government is going to restrict GPT-5.6's release, sort of unilaterally, basically without precedent. I don't think the government has ever asked a frontier lab not to release a model before. Certainly, a government has not asked a frontier lab to be able to oversee which customers the model is released to. This seems very unnatural. What are your takes on this?

• 2:26Speaker 00

  Yeah, so it was funny. I was in DC last week, so I was kind of like had some sort of front row seat to what was happening. Interesting anecdote is that we bumped into Tom Brown there, like the co-founder of Entropic. And we were like, oh, it seems to be like a change, change of staff or something like that. Before, before it was made public that they changed a little bit the people talking to the White House there. Listen, I mean, what I've seen, what I'm hearing is that there's a lot of, you know, interest from a lot of people in the USG to really understand the risks of AI

• 3:07Speaker 00

  and take kind of like a safe approach to the deployment of frontier models. To be honest, I can't really blame them because of the fact that, you know, the frontier labs were basically doing marketing for the past few years, right? Like if you remember, GPT-2 was too dangerous to release, right? It was like, what, like four, four, five years ago?

• 3:32Speaker 00

  And so, you know, I don't really blame them for doing things like that. I just hope that progressively we'll get a little bit more transparency about, you know, what is safe, not safe, so more focused on, you know, transparent evaluation of models and things like that. I also hope that it's going to stay contained to, you know, a few frontier generalist models, because frankly, I think they're the most dangerous ones, right? And also, you know, these companies are trillion dollar companies with armies of, you know, DC people. So they can, they can kind of like deal with it.

• 4:12Speaker 00

  I hope it's going to stay contained to that and not, you know, permeate to a lot of different players. For example, startups, small companies, you know, academia, or like people who don't necessarily have, you know, the money, the size, the ability to really deal with these things. That, that would be kind of like my, my main concern.

• 4:36Speaker 01

  Totally, yeah. What we were just talking about before was just like, is this, could this be applied to open source companies and models in any way? Like, is there a way that the US government could come in and restrict open source companies, either practically or legally?

• 4:53Speaker 00

  I don't think so, because, you know, I think, I think open source models are inherently less dangerous than kind of like the models that are getting restricted now. Because, you know, these models are a little ahead in terms of like being closer to the frontier. Also, open source models are less generalist, they're more specialized. And it's just like, nobody or very few people focusing on building dangerous like cybersecurity capabilities. So it's like a little bit different than, than kind of like the proprietary labs.

• 5:30Speaker 00

  So I, yeah, so I, I don't think, I don't think it would make sense. I don't think it's going to get to open source just because of some of these differences. At more high, high level, you know, I think in general, open source AI is much, much safer than kind of like proprietary AI for a bunch of different reasons that I've talked about in different outlets. But, yeah, so, so I think the approach there is going to be, hopefully is going to be a little bit different between kind of like closed source, proprietary, frontier models, and then the rest of the industry and the ecosystem that has, in my opinion, posed much less

• 6:10Speaker 00

  enter much less threats, where we want to keep kind of like, you know, supporting and building up to focus on, you know, competition, to enable kind of like little tech, small companies, everyone to be able to basically participate in AI.

• 6:28Speaker 03

  Totally. I agree that open source models are less dangerous now, but in, you know, six months, if trends continue, you're going to have an open source model from somewhere with like mythos level capabilities that freaks out the government. And you could imagine them at least wanting to restrict open source models at that time.

• 6:47Speaker 00

  I'm not, I'm not that sure, because there's this weird thing where basically the most dangerous things usually are not so much developed in, in open source. You know, maybe it's this thing that people say like sunlight is the best disinfectant. You know, like funny, funny thing is that sometimes safety people are talking about the nuclear bomb. A nuclear bomb has never been built in open source, and I think it would have never been built in open source. It's been built in a closed source, you know, proprietary team with billions of dollars of, you know, of resources. So it's much less alike, you know, some players than some others.

• 7:32Speaker 00

  So I think, I think there's also a path where open source keeps building kind of like more specialized models for different domains and not necessarily for the domains that are presenting the biggest risks. You know, I don't think it's automatic that you build a more powerful model that it's more dangerous for cybersecurity. For example, you could build a more powerful model that is not more dangerous for cybersecurity. For example, if you don't train it on cyber security, which weirdly people are not really talking about. Why are we not talking about that rather than talking about, you know, removing the ability to release or putting kind of like safeguards after the fact that we know are not really working because everyone can jailbreak them.

• 8:16Speaker 00

  So, so I think I wouldn't be totally surprised if the open source

• 8:22Speaker 00

  you know, community, just because it's structurally very differently set up than the big labs, are actually taking different directions that keeps it kind of like safer and never really requires the kind of regulation that you need in kind of like a closed source AI labs.

• 8:38Speaker 03

  Well, is this not also kind of an argument against the capability of open source AI research?

• 8:44Speaker 00

  Think so, because you know, it solves different problems. Like I sometimes take the example of, you know, local models, right? Local intelligence, you know, being able to be on a flight in airplane mode without network and still be able to get intelligence. You can only get that with open source, right? Like you can't get that with API. There's just like literally no way, no way, no way to do that. So I think it's just different, different layers of the stack, right? And actually, open source is on top of closed source. A lot of the closed source is using open source models and is using open source infrastructure. And it solves different things. The analogy is like, you know, open source maybe is the engine, and the API is the car, right?

• 9:27Speaker 00

  And obviously, the engine is never, never gonna be like a Ferrari. But you know, that's what kind of like powers, powers, powers the Ferrari.

• 9:37Speaker 01

  Right.

• 9:37Speaker 00

  So I think that's more like the way, the way we approach it. And also, you know, being less good at bad things doesn't mean that you can't be better at good things. Maybe, you know, open source is better at, you know, solving people's problems. Maybe it's better at, you know, kind of like helping, you know, do stuff really, really important, but it's worse at, you know, creating cybersecurity attacks. That would be, that would be kind of like the ideal, ideal case. People right now are talking about frontier as kind of like this general thing. The reality is that the frontier is kind of like jagged between kind of like different tasks, different domains, right?

• 10:22Speaker 00

  This one model is gonna be better at some things and it's gonna be worse at other things. I think that's, that's more kind of like how we should approach it.

• 10:29Speaker 01

  Yeah.

• 10:30Speaker 00

  Does it make sense?

• 10:31Speaker 01

  No, yeah, this totally makes sense to me. So you guys just crossed a hundred million dollars in ARR, right? So I'm curious as like, what do you think this means for like, like the business model of open source?

• 10:44Speaker 00

  Obviously, a lot of companies are doing much more revenue than, than we do in AI these days, you know?

• 10:52Speaker 00

  It hasn't really been our priority to optimize for monetization and for, for revenue, given what we're building, which is more kind of like a usage-based platform to reach kind of like how many and empower as many AI builders as, as possible. But at, you know, at this, at this small scale, I think it shows that, you know, there's a business model for open source, there's a business model for open source platform. We kind of knew it, right? Because there's, there's been GitHub before and there's been kind of like a bunch of open source successful companies. But I, I guess, I guess it's a, it's a validation of, of that. And we've seen, I mean, for the past, past, past few weeks, we've seen quite, quite a lot of growth in terms of interest and adoption of not only open source models, but also

• 11:37Speaker 00

  local models.

• 11:40Speaker 00

  And so, you know, that, that also speaks, speaks a little bit to that.

• 11:45Speaker 01

  Yeah, what are some of the specific use cases that like people are using local models for?

• 11:52Speaker 00

  So local models are kind of like, kind of free, right? Because they're running on your, on your phone or on your laptop. And so you don't even, you don't really have to, to, to pay for them. So they're, they're much cheaper, they're much more privacy, kind of like preserving by design, because you don't have to send your data to an API, right? Your data stays on your, on your phone.

• 12:18Speaker 00

  And so we, we see people using it a lot for the things when it matters the most. So like, for example, if you wanna talk about your private health and you don't wanna share that with, with someone else. If you wanna share some of your private company data and you don't wanna share it externally to an API provider.

• 12:37Speaker 00

  Or if you wanna run really, really heavy workloads, for example, agentic workloads and really have something that runs 24/7, then doing it on, on your laptop or like on the Mac Mini or, or kind of like your local hardware, makes it much, much more sustainable. So that, that would be kind of like the, the use cases. We have this library called Llama CPP, which is the most used runtime for, for local AI workloads that people are using a lot and they're using GPT OSS, they're using Gwen,

• 13:13Speaker 00

  Gemma four, like all, all these models locally on their, on their laptop.

• 13:18Speaker 01

  Yeah, for sure, for sure.

• 13:19Speaker 03

  So going back to open models, you can imagine that the government will want to restrict open models because a lot of them come from China. Maybe not restrict them in a strict legal way, but maybe in like an export-related way. So do you think that might happen? And if so, what would that mean for Hugging Face?

• 13:38Speaker 00

  Yeah, I mean, like it's, open, open weights are fundamentally different than an API, right? The way, the way you can restrict it is very different. So for example, you know, if you remove an open weight from Hugging Face, then it's still gonna be on ModelScope, for example, which is like the Chinese equivalent, or it's gonna be like on Torrent platforms. So like restriction like looks very, very different, I think, for, for open source than, than for APIs. You can't really block because it's open, so almost by, by definition, there's gonna be some ways, some ways to, to access, access them.

• 14:18Speaker 00

  And also, the provenance for open, open weights doesn't really matter as much because, because, because it's open, the, the people who are sharing it kind of like give up the control on it and give up kind of like their ability to influence you. So to me, it doesn't matter so much where open weights are coming from. It's a different game, for example, for, for APIs to run the difference, because if you're using an API provider or a cloud from China, obviously you're sending your data and also they could cut your access or bias your access, right? That's a much, much bigger problem. But for open weights and open source, it doesn't really matter where it comes from

• 15:02Speaker 00

  because it's, it's kind of like, you, you get all the control, you get all the transparency, there's no way to kind of like trick you, bias you, manipulate you, remove your access. So I, I think for open source, like the provenance doesn't, doesn't matter as much.

• 15:19Speaker 01

  Yeah, I'm curious, your thoughts on just like the US government sort of like restricting the release of like GPT 4.6? Do you think that comes from like them knowing the stakes or not knowing the stakes? Like, do you think they're well-informed on this matter? Or they just like know that they don't know and that's why they're taking these measures?

• 15:38Speaker 00

  It's a good question. I can't, I can't talk, talk for them. I, I do think there's a lot of interesting learning and progress to be made everywhere, not just at the, the USG, but really everywhere on evaluating models, evaluating risks or for models. I don't think we have really good benchmark for, for this, and that's a big problem. In general, ultimately, I hope, yeah, we'll have more transparency, right? There, there's this agency called Casey.

• 16:08Speaker 01

  Yeah.

• 16:09Speaker 00

  That is, is amazing. I think they're doing an amazing job and they're building up this capability to really evaluate and work on benchmark and things like that. And I'm really excited for them to take a little bit more of the workload there and, and, and kind of like take a, a very scientific approach to, to evaluating these, these models. And I think when, when they, when they will, it's, it's gonna be really good for the shields.

• 16:37Speaker 01

  Yeah, we definitely wanna talk to people from Casey soon.

• 16:39Speaker 00

  Yeah.

• 16:40Speaker 03

  It's gonna be really tough, I can't lie.

• 16:42Speaker 01

  We'll try, we'll try.

• 16:44Speaker 00

  Yeah.

• 16:45Speaker 01

  Oh, also, yesterday I was talking to Andrew Trask from DeepMind and he had like a very interesting viewpoint that he, like, there's a model on OpenRouter called OpenFusion, I think, and it's this like fusion model of basically a bunch of different models. And that like had a lot of advanced capabilities and like surpassed, like inefficiency in a lot of ways. So do you think we're gonna see more of that?

• 17:09Speaker 00

  I think so, yeah, yeah. I think what we're seeing right now is that a lot of people, companies are realizing that it's, it's too dangerous, it's too risky, it doesn't make any sense to rely exclusively on one model. Why? Because this model can be taken away, this model can be biased, this model can refuse or tell you the wrong things, right? That's, that's also what we've seen before, right? With, with Fable 5 before, before it was taken out, right? Is that there was some domain where it was intentionally by design, kind of like telling you the wrong things, right? To confuse you. And so I think people are realizing that, that we need to rely on a multitude of, of models.

• 17:53Speaker 01

  Yeah.

• 17:54Speaker 00

  Right?

• 17:57Speaker 00

  And so I think that's, that's driving to, to, to this, to this outcome of doing more routing.

• 18:05Speaker 00

  But there was an interesting study from, from Stanford, published last year, end of last year, that was showing that 70% of the queries that people ask to chat GPT could be accurately answered locally on your laptop.

• 18:20Speaker 01

  Okay.

• 18:20Speaker 00

  So for free, like, you know, questions, there are, most of the questions you ask or, or most of the, the AI workloads that people do today with frontier models could be done by models that are cheaper, faster, more customizable, more controllable, right? And they don't do it because frankly, it's a pain to take the picker, model picker and be like, okay, this one, I'm gonna go for like a cheaper one because, you know, also you're subsidized, so you don't have to care because you have your subscription, so you root, you, you direct everything. It's like directing everything to Einstein, right? It's like, hey Einstein, what's, what's, hey Einstein, what's the weather today? You know?

• 19:01Speaker 03

  Yeah.

• 19:02Speaker 00

  In normal life, you would be like, fuck you, I'm not answering your silly question. But because it's AI and subsidized by the, by the AI labs, all the questions are getting routed to Einstein versus in an ideal world, you can have different people, different models that are more specialized and, and better at answering your, your questions in different, different domains. So that's kind of like, yeah, what we're seeing and the, the way to route instead of giving you the model picker, I think is a very, very smart option and a better one. Lovable is starting to do that too, right? Like doing the routing under your hood. And I think it will, it's possible that it's gonna redistribute a lot of the value capture from frontier models, which have been the case now, right?

• 19:47Speaker 00

  Like majority of the revenue capture was on frontier models to a more like long tail of of models, uh, which in my opinion makes much more sense. It's like, it's like AI, AI maturing, right? Like we were in the first phase of AI where it's very simple, very simplistic, everyone using just one gigantic model behind proprietary APIs. Now we're moving to the second phase of the AI field more maturing and using several models, using open source, having control, building themselves. I'm, I'm quite excited about it.

• 20:20Speaker 01

  Hmm, yeah.

• 20:21Speaker 03

  So another big news story of yesterday was Anthropic accused Alibaba of doing distillation attacks, which is, you know, there's two perspectives on this. One is like, this is like, these are the, these evil people who are like stealing the capabilities of our models, violating our terms of service, like fraudulently accessing our product. And then the other is like, what do you mean? They're creating accounts and they're paying for tokens. And you know, you can't accuse people of stealing when you stole the entire internet. So which one of these two positions are you closer to?

• 20:52Speaker 00

  Well, I mean, I think distillation is a very common, you know, practice that everyone is using. Uh, you know, I wouldn't be surprised if Anthropic used distillation in the past for some of their models, for some of their specialized models using kind of like someone else who's better. For example, you know, when, when OpenAI was better at coding, like you use this model to kind of like help you a little bit in the training of your, your, your coding model. It's something that everyone uses, but that is not, you know, the main reason for success. Like if you suck, you suck with or without distillation. It's just kind of like a little bit accelerating thing, but it's, it's not, it's not what makes you good or bad at training models.

• 21:35Speaker 00

  So if you start distillation tomorrow, the Chinese labs won't like go down and disappear because it's still gonna be good. Uh, it's not, it's not really gonna change the game. Uh, and you know, I mean, the, the, the only point that I'm, I'm a little bit, uh, uh, you know, biased towards that, uh, like if, if there was big competition problems, right? Where it's like, oh, it's really unfair, it's biasing competition. But it's, it's hard for me to accept this one because frankly, you know, I mean, Anthropic, OpenAI, they've been the fastest growing companies in the world. They've, they've come overnight trillion dollar companies. And so I don't think they have competition problems, you know?

• 22:15Speaker 00

  It's hard, it's hard for me to be, to say like, oh, poor Anthropic, poor OpenAI, you're getting unfairly competed with when, when you're like the fastest growing company in the world. Competition has been okay for them. If anything, I think they need more competition than less competition.

• 22:33Speaker 01

  Yeah.

• 22:34Speaker 00

  Because we're heading toward a world where like a few companies are completely dominating, concentrating all power, all capabilities, all wealth. Uh, that, that's much more dangerous than, than maybe, you know, losing a couple of billion dollars of revenue, you know?

• 22:54Speaker 01

  Yeah, totally.

• 22:56Speaker 00

  I think. So it's, it's just hard, yeah, it's just hard to, um, you know, empathize and kind of like the, uh, think that this is, this is an important problem. I think there are many, many more, much more important problems than that in the world of AI right now.

• 23:13Speaker 03

  Sure. So since the last time we talked, there have been two big pieces written about Europe. There's this essay Europe 2031, which is basically AI 2027, but for Europe. Like, uh, basically Europe will slide into irrelevance if they don't lock in on AI right now. And then the other was, uh, Anton Leish, who's a policy writer, wrote this piece on Substack called The Moonshot, which basically explained how if Europe wanted to do so, they could build a frontier lab. Uh, do you think that it's possible for Europe to do this at this point? Could they build a frontier lab if they really tried?

• 23:47Speaker 00

  Um, I think so, yeah. Yeah, yeah. Um, they, they have a lot of really great resources. They, they have great people. Why you already have some great labs, right? Black Forest Lab, Mistral, uh, all these people are, are doing, doing amazingly and, and arguably they're, they're the frontier. Um, they have, they have amazing energy, you know? Obviously France, for example, uh, you know, nuclear energy, very abundant. So they, they, they could really kind of like, uh, uh, use a lot of clean energy for, for AI. Uh, so yeah, I think, I think they, they could. It's just a matter of kind of like, uh, focusing the energies, uh, towards that, uh, building an ecosystem.

• 24:29Speaker 00

  It never, um, sometimes, sometimes we, uh, we build the stories of like companies, uh, emerging out of the blue by their sheer power. But the reality is if it's, it's more an ecosystem, right? And then you see that from, uh, OpenAI, right? The tea of transformers, obviously it's coming from, from Google, that, that open source transformers. Um, and so it's, it's more of a matter of, uh, in my opinion, fostering an ecosystem of like open research, open site, open source AI, which is what happened in the US, right? And, and kind of like fostering that progressively to, uh, um, bring more and more, uh, companies, uh, organizations closer, closer to the frontier.

• 25:13Speaker 01

  Yeah, totally. Um, I'm curious, like, since you run such a large platform, like, what are younger people doing with AI? Are people, are younger people actually becoming very, very proficient in like AI native? Um, or like, how do you see this pattern of behavior?

• 25:29Speaker 00

  Yeah, yeah. Um, we see them a lot, right? Trying, uh, it's almost kind of like, I, I feel like young people went through the first phase of, of being users of AI really quick. And now a lot of them, I think, wanna be builders.

• 25:42Speaker 01

  Yeah.

• 25:43Speaker 00

  So we, we see a lot of, uh, yeah, very, very young people going on a new phase, getting, getting models and, and you know, building, building products themselves or optimizing training, training models themselves, uh, you know, building data sets themselves. Uh, so I, I see a lot of like, yeah, building, uh, appetites, uh, in AI for, for, for young people in a lot of different domains, not necessarily, uh, I, uh, yeah, not necessarily in kind of like the most talked about domains, but also in a lot of like, uh, very, uh, you know, topics that are really not talked about, like, uh, um, climate, climate change, uh,

• 26:27Speaker 00

  you know, biology, chemistry, uh, you know, really kind of like, uh, even, uh, so, it's social media, kind of like, like there's a lot of, a lot of topics that we don't really talk, talk about, that I feel like are closer to, uh, to, to everyone's interest and, and I, I see a lot of young people working on these things.

• 26:48Speaker 01

  It's a good white pill to end on.

• 26:50Speaker 03

  That is, yeah.

• 26:52Speaker 01

  Well, thank you so much, Clem. This was so great.

• 26:55Speaker 03

  Thank you, Clem.

• 26:56Speaker 00

  Thanks, my big man. Thank you.

• 26:58Speaker 01

  Very big fans.

• 26:59Speaker 02

  Thanks for listening to this episode of the A16Z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z and subscribe to our Substack at A16Z.substack.com. Thanks again for listening, and I'll see you in the next episode. This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z.

• 27:42Speaker 02

  Such advertisements, companies, and individuals are not endorsed by AH Capital Management LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.