Key Takeaways
- Flexport CEO Ryan Peterson says he doesn't "lose sleep over" national security fears concerning Chinese open-source AI, despite its widespread adoption by Silicon Valley startups.
- Peterson believes the source of open-source models is irrelevant, stating, “if they're open source they're open source like who cares where they're from like we can use them.”
- Harry Stebbings brought up concerns from figures like Keith Rabois, who suggest using Chinese models could “feed data back to the CCP,” creating a “window into Silicon Valley.”
- Peterson argues the US and China share a “huge amount of mutual dependence” and that their interests are “much more aligned than people give credit,” making casual talk of war irresponsible.
The Disagreement
Is Silicon Valley inadvertently inviting national security risks by adopting Chinese open-source AI models? That's the core tension Harry Stebbings put to Flexport CEO Ryan Peterson. Stebbings, channeling worries from investors like Keith Rabois, painted a picture where “CCP funded open source models are powering the majority of early stage Silicon Valley companies.” The fear, he explained, is that this could allow data to flow "back to the CCP," essentially giving China a "window into Silicon Valley." It's a vivid warning: the very tools driving innovation could be Trojan horses.
Peterson, however, sounded a starkly different note. He acknowledged that “The majority of open source models are Chinese,” but quickly dismissed the concern. “I don't personally spend a lot I don't lose sleep over it,” he told Stebbings. His logic is simple: if the code is truly open source, its origin becomes less relevant. "If they're open source they're open source like who cares where they're from like we can use them." For Peterson, the utility and accessibility of open source trump geopolitical concerns at the technical level. He took the argument further, suggesting that anxieties about conflict between the US and China are overblown, pointing to a deep "mutual dependence" between the two powers. To him, the "saber rattling" and casual talk of war ignores the reality that "such a thing would be a nuclear war and you'd all be dead." Peterson puts the odds of that outcome as "pretty low."
Who's Right (and When They're Wrong)
Both Peterson and those he pushes back against make valid points, depending on your frame. Peterson is likely right for a certain class of ambitious builder. If you're running a small startup and your product isn't tied to critical infrastructure or highly sensitive government data, the immediate benefits of using the best available open-source tools – regardless of origin – probably outweigh abstract geopolitical fears. His broader point on mutual economic dependence also holds weight; it's a powerful disincentive to outright military conflict, explaining why two powers often posture without directly clashing. For a founder focused on building and shipping, Peterson's view offers permission to move fast and grab the best tech.
However, the concerns raised by Stebbings (via Rabois) aren't simply dismissed. Even if open source is open source, there are nuances. Models can have biases, hidden backdoors, or simply provide intelligence about who is building what by observing usage patterns or contributing communities. A company working with sensitive personal data, national security interests, or critical industrial control systems needs to approach any foreign-sourced technology, open or closed, with extreme caution. Peterson might be too optimistic about the types of conflict. While nuclear war may be unlikely, other forms of state-sponsored intelligence gathering or economic warfare are very real, and technology is often the vector. For a founder building a company that could eventually operate at scale or touch sensitive sectors, ignoring the origins and implications of your core tech stack is a gamble.
What to Do With This
Audit your core technology dependencies. If you're using foreign-sourced open-source AI models, specifically Chinese ones, understand what data flows through them and how that data is handled. Can you contain it? Could the models reveal anything sensitive about your product or users if compromised? Balance the immediate gains from powerful open-source tools with a realistic view of longer-term geopolitical risks to your supply chain.