Y Combinator’s Garry Tan wants US open-weight AI labs to ‘distill’ frontier models, too

Garry Tan's proposal is quietly radical in a way that cuts against the prevailing narrative in Silicon Valley. For the past two years, the open-weight AI conversation has been dominated by a kind of defensive scramble—Western labs releasing "open" models mostly to avoid being outflanked by DeepSeek and other Chinese entrants. Tan's framing flips the script: instead of treating open-weight as a concession to be managed, he treats it as a strategic asset to be cultivated, and he explicitly wants American labs to use the same distillation playbook that made their Chinese counterparts dangerous in the first place. That's not just a technical suggestion; it's an admission that the frontier labs' moat—proprietary training compute—is less relevant than they'd like to believe when smaller players can distill competitive models from their outputs. The "how" here is important: distillation is cheap, fast, and increasingly effective, which means the barrier to entry for open-weight labs has collapsed. The "why" is geopolitical: Tan is essentially arguing that if open-weight models are inevitable, the US should ensure they're built on American intellectual foundations rather than Chinese ones, so that the global default open model isn't inherently aligned with Beijing's interests. There's also a subtle tension in Tan's proposal that's worth unpacking. He's asking US open-weight labs to distill from American frontier labs—but those frontier labs (OpenAI, Anthropic, Google) have spent the last year trying to restrict exactly this kind of use of their outputs in their terms of service. Tan is essentially proposing a policy carve-out for domestic labs, which would require either regulatory action or a voluntary détente among US AI companies. That's not impossible, but it raises questions about whether the frontier labs would cooperate, given that they view open-weight models as competitive threats to their commercial API businesses. The strategic logic is sound—a robust American open-weight ecosystem would counter Chinese influence—but the incentive alignment is murky. Why would OpenAI want a bunch of domestic distillers eating into its market, even in the name of national competitiveness? The deeper implication is that Tan is normalizing distillation as a legitimate development pathway, which is a notable ideological shift. For years, distillation has been treated in some circles as a form of intellectual property theft—the thing that "bad actors" do to clone models. Tan's framing recasts it as standard industrial practice, which could have ripple effects on the copyright and AI governance debates. If the president of Y Combinator is publicly endorsing distillation as a patriotic strategy, it becomes much harder to argue that the technique is inherently illegitimate. That's a meaningful shift in the Overton window for AI policy. Ultimately, Tan's argument is less about the technology and more about the geography of AI influence. He's betting that whoever controls the default open-weight models controls the global AI ecosystem's norms, and he doesn't want that default to be Chinese. Whether the frontier labs play ball is the open question—but the fact that a major Silicon Valley figure is even floating this idea signals that the open-weight landscape is being reimagined as a strategic frontier, not just a technical niche. Source: https://techcrunch.com/2026/09/11/y-combinators-garry-tan-wants-u-s-open-weight-ai-labs-to-distill-frontier-models-too/