China's AI Models Are Splitting Washington in Two

A new Chinese AI model drops. Silicon Valley notices. Washington panics. Factions form. Rinse and repeat.

Kimi is the latest entry in what has become a reliable genre of tech news, but MIT Technology Review has the most useful framing I've seen on it: the real story isn't that China is building competitive AI models, it's that doing so has cracked open a genuine strategic rift inside the Trump administration. Every time a capable, free Chinese model appears, according to MIT Technology Review, U.S. companies have less commercial incentive to pay for American alternatives—and Trump's AI advisors have less consensus on what to do about it.

The fault line runs roughly here: one camp wants to win through open competition and faster American innovation. The other wants tighter government control over frontier AI development because the national security implications are too serious to leave to market dynamics. Both camps agree Kimi and its peers are a real challenge. They just disagree on what "winning" looks like, and that disagreement has proved surprisingly durable.

The Open-Source Problem

The thing that makes Chinese models strategically disruptive isn't benchmark scores. It's distribution.

The Atlantic notes that Chinese AI models are gaining international traction largely because they're mostly open-source—meaning anyone anywhere can adopt and adapt them without paying. That's not a technical detail, it's a business model attack. OpenAI and Anthropic built expensive proprietary systems and priced them accordingly. Open-source Chinese models undercut that logic for every customer who doesn't have a specific reason to pay for the American alternative.

The irony, noted by The Atlantic, is sharp: Chinese regulators are apparently so spooked by foreign competitors gaining access to Chinese AI expertise that they're discussing whether to restrict overseas access to their own top models. Beijing would be considering pulling up the drawbridge on the same open-source strategy that's been doing geopolitical damage to Washington. Whether those discussions result in actual policy is unclear—The Atlantic attributes this to reporting on Chinese regulatory deliberations rather than any announced decision—but the instinct itself tells you something. Open access is a weapon that cuts in multiple directions.

If both governments end up restricting each other's AI, the global concentration of AI development gets weirder, not simpler. The rest of the world would be choosing between two walled gardens instead of one.

The Distillation Question

Here's where I want to be direct about something that's been getting soft-pedaled.

The Los Angeles Times reports that U.S. politicians, OpenAI, and Anthropic have accused Chinese AI models of illicit "distillation"—essentially using American models' outputs to train Chinese ones, extracting capability without paying for it. Beijing calls the claim "groundless."

Maybe Beijing is right. Maybe it isn't. But consider who's making the allegation loudest: the two American companies most directly threatened by capable, free Chinese competitors. OpenAI and Anthropic have a product to protect, investors to reassure, and an ongoing lobbying project in Washington around AI regulation. Accusing your most dangerous competitor of theft is not merely a legal claim—it's also excellent competitive positioning. That doesn't make the accusation false. It does mean you should weigh it with that context in mind, and notice how conveniently the call for stronger government action against Chinese models aligns with what those same companies have been pushing for on other grounds. The companies most damaged by competition are also the most credible-sounding witnesses to misconduct. Washington has seen that movie before—remember when incumbent telecoms testified about the dangers of internet telephony?

Two Theories of the Race

Forbes calls this a threat to "American AI hegemony." The word "hegemony" is doing some dishonest work there—it frames market leadership as a natural right being violated, rather than a competitive position that has to be maintained. The U.S. didn't win the internet era because hegemony was its birthright; it won because it had the right incentive structures, talent pipelines, and capital markets at the right moment. Those advantages aren't permanent and they're not owed.

ProPakistani captures the internal fracture cleanly: Trump's AI allies agree Chinese open-source models are a serious challenge, but they don't agree on the response. Some want more open competition. Others want stronger government oversight because frontier AI could affect national security. The Trump administration has discussed banning Chinese open-source models outright, while Beijing has reportedly discussed banning overseas access to its top models—a mutual assured restriction scenario that would have seemed like Cold War parody five years ago.

The ban-it camp and the compete-harder camp are actually operating from incompatible assumptions about what AI is. If AI is a product—a service you sell—then competition is the right frame, and better American products win. If AI is infrastructure—the substrate on which economies and militaries run—then you regulate it the way you regulate power grids, not the way you regulate smartphones. The open-source pricing pressure from Chinese models is forcing that question into the open, and Washington doesn't have a settled answer.

This is almost exactly the argument that consumed Washington during the 1990s telecom deregulation fights. Was the internet a product or infrastructure? Was broadband a competitive market or a public utility? Those debates took a decade to half-resolve and produced regulatory frameworks that satisfied almost nobody. The Microsoft antitrust case ran on a parallel track for years while the underlying market restructured itself around the very companies the government wasn't watching. The lesson wasn't that government intervention was wrong—it was that the intervention always lagged the actual dynamics by roughly one product cycle. AI is moving considerably faster than Windows 98.

What Competition Actually Produces

One outcome the current anxiety is generating—regardless of how the policy fight resolves—is cheaper, better AI for users. That's not a minor footnote. AI competition is already pushing prices down in ways that benefit anyone who isn't in the business of selling expensive AI access. Chinese models being free and capable creates real pressure on American providers to justify their pricing, which is a normal and generally healthy market dynamic.

The geopolitical noise around that dynamic is real, though. Who controls what AI gets released, under what conditions, to whom—those aren't questions any single company can answer, and Washington is still working out whether it wants to answer them at all. Meanwhile the Anthropic situation demonstrated that the Trump administration is willing to apply serious pressure on domestic AI companies that don't align with its strategic preferences. The companies being asked to compete harder against China are simultaneously being squeezed harder at home.

That's the actual shape of the problem. Not China's models versus America's models. A set of actors—governments, companies, researchers—each making locally rational decisions that are producing globally incoherent outcomes, while the underlying technology keeps moving and the window to set durable policy keeps closing.

The factions in Washington will keep arguing. Kimi will not wait for them to finish.

---

Mike Sullivan covers technology for BuzzRAG.