Clem Delangue | The a16z Show Summary

1. Key Themes

The Open Source Power Shift: From U.S. to China

The U.S. pioneered open source AI — the transformer architecture itself was open sourced by Google — but the trend has reversed. Closed APIs now dominate U.S. frontier labs, while China has become the dominant open source contributor. This has practical consequences: most U.S. startups and academics building on open source are now relying on Chinese models.

"If you ask most startups, most academia in the U.S. that are using open source, they're usually using Chinese open source models, right? You've probably heard of DeepSeek, of Qwen, of Kimi. There are kind of like a bunch of companies and organizations in China contributing massively to the field of open source." — Clem Delangue [00:03:13]

The LLM Bubble Thesis

Clem draws a careful distinction: AI broadly is not in a bubble, but the specific sub-sector of large language models distributed behind closed APIs may be. The concern is massive infrastructure investment with uncertain long-term margins and moat sustainability.

"If there's one specific domain of AI where there's so much investment that there's maybe a risk of overinvesting, it's large language models distributed behind APIs, right? Like you see the building of crazy data centers for it. And obviously you see a lot of revenue growth, but with kind of like uncertain margins and certain kind of like long-term sustainability and moat for it." — Clem Delangue [00:03:43]

Robotics as the Next Open Source Frontier — With a China Risk

Clem sees robotics as the next domain unlocked by AI and is actively building in it. But he explicitly warns that China is already dominating robotics hardware, making this a strategic national concern, not just a market opportunity.

"There's an important China-U.S. component there because it's very likely that Chinese are going to dominate robotics. So at least they're already dominating. And so on this topic, too, it's really important that we build more in the U.S. on this topic." — Clem Delangue [00:11:57]

2. Contrarian Perspectives

Open Source AI Is Actually the Safer Option, Not the Riskier One

The conventional wisdom is that open sourcing powerful AI models creates security risks. Clem inverts this: restricted access creates the real danger, because attackers may gain capabilities that defenders lack. Open access levels the playing field and empowers defenders more than attackers.

"If we talk, for example, for cybersecurity, the biggest risk is that a few players have capabilities that other people don't have. Right. And so the attackers could have capabilities that the defenders don't have. Whereas kind of like if you make it more open, actually, it's usually easier for the defenders to react and kind of like make the whole system safer." — Clem Delangue [00:06:33]

AI Safety Concerns Are Chronically and Predictably Overblown

Most would view AI safety debates as a reasonable ongoing concern requiring constant reassessment. Clem argues the pattern is clear and consistent: each generation of models triggers alarming safety warnings that prove overblown, going back to GPT-2 which was just an autocomplete at the time.

"I think six, seven years ago, at the time it was GPT-2. And there was already like a lot of people saying that it was too dangerous to release in open source at the time. Right. It was six, seven years ago when basically it was nothing more than just an auto-complete. I think we've seen progressively that these were quite overblown. And I think they're also overblown today." — Clem Delangue [00:05:12]

Regulating Access to AI Capabilities Makes Everyone Less Safe

The common policy instinct is to restrict access to dangerous capabilities. Clem argues the opposite: restricting access slows progress, creates dangerous capability asymmetries, and actually increases net risk. The right model is universal access plus targeted enforcement against bad actors.

"The idea of restricting a technology like AI based on risks is just like, for example, you would say, okay, some people can punch other people. So let's tie down everybody's hands, right?... The way you want to control it is untie everyone and then regulate or fight the bad actors." — Clem Delangue [00:00:00]

3. Companies Identified

Hugging Face The leading open source AI platform, often described as the "GitHub of AI." Hosts models, datasets, and now robotics hardware. Handling massive data scale — two petabytes added in a single week — with both public and private usage growing rapidly.

"Just last week, we added two petabytes of data to the platform just last week. It's kind of like a matter of comparison. It's the equivalent of 500,000 two-hour movies that have been uploaded to Hugging Face just last week." — Clem Delangue [00:13:09]

DeepSeek / Qwen / Kimi Chinese open source AI model providers cited as the primary resources now being used by U.S. startups and academics. Mentioned as evidence of China's dominance in open source AI contribution.

"You've probably heard of DeepSeek, of Qwen, of Kimi. There are kind of like a bunch of companies and organizations in China contributing massively to the field of open source." — Clem Delangue [00:03:13]

4. People Identified

Jensen Huang (NVIDIA CEO) Cited approvingly by Clem for joining U.S.-China AI diplomatic conversations and for having the right perspectives on fostering collaboration and openness between nations.

"I'm glad that Jensen hopped into the plane and joined these conversations. Because I think he has a lot of the right perspectives on this topic. To kind of like basically create more collaboration between countries." — Clem Delangue [00:09:46]

Sam Altman (OpenAI CEO) Referenced in the context of the robotics and new AI device opportunity, cited as a fellow believer in the importance of bringing new physical AI devices to market.

"That's why OpenAI and Sam Altman, for example, have talked a lot about their excitement about bringing new devices to market." — Clem Delangue [00:11:27]

5. Operating Insights

Build the Infrastructure Moat That Platform Competitors Won't

Hugging Face succeeded where GitHub failed not by being smarter about AI strategy, but by building the unglamorous infrastructure required: petabyte-scale file storage and data handling specific to AI artifacts. GitHub never prioritized this because it wasn't their core problem. The lesson for operators is that winning a platform adjacency often comes down to building the specific technical substrate the incumbent chose not to invest in.

"Hosting and sharing AI artifacts is quite different than hosting code... For example, for us, the volume of files, of data that we're dealing with is much, much larger than what GitHub is doing." — Clem Delangue [00:12:43]

Use Hardware as a Distribution and Ecosystem Lock-In Strategy

Hugging Face shipped nearly 10,000 Reechy Mini robots globally, and the product includes an app store with over 300 apps already created. This is a deliberate ecosystem play — physical hardware creates a new distribution surface for AI capabilities that a purely software product cannot replicate, and the app store creates compounding network effects.

"We've shipped almost 10,000 of them all over the world. So it's probably one of the most widely distributed robots of the year at this point... There's been over 300 apps that have been created for it already." — Clem Delangue [00:10:28]

6. Overlooked Insights

The "Private Usage" Signal at Hugging Face Is the Real Business Story

Clem briefly and almost casually mentions that Hugging Face now has significant private usage of its platform — enterprises hosting proprietary models and datasets on Hugging Face infrastructure. This was mentioned in a single sentence but is enormously significant: it means Hugging Face is quietly becoming enterprise AI infrastructure, not just a public open source community. This is the GitHub → GitHub Enterprise transition, and it likely represents the bulk of the monetizable, defensible business.

"We have a lot of private usage now. So that's kind of like some of the reasons why we managed to do it, whereas GitHub focused on other things." — Clem Delangue [00:14:02]

U.S. Startups Are Already Operationally Dependent on Chinese AI Models

This was stated plainly and then moved past quickly, but the strategic implication is significant: a large portion of the U.S. AI startup ecosystem is already building on Chinese open source models. This creates a hidden geopolitical dependency that has received almost no attention. If DeepSeek, Qwen, or Kimi were sanctioned or restricted, a meaningful swath of U.S. AI products would be disrupted. This is an underappreciated systemic risk — and an investment signal for whoever can credibly offer U.S.-origin open source alternatives at comparable quality.

"If you ask most startups, most academia in the U.S. that are using open source, they're usually using Chinese open source models." — Clem Delangue [00:03:13]