The biggest fear in the AI race is not sending chips to China. It’s where they go next.

The export ban on advanced AI chips has cut Nvidia off from a market worth billions of dollars, but analysts like Allen argue that Nvidia has benefited in other ways from export controls. For instance, the first Trump administration blocked TSMC from selling advanced seven-nanometer chips to Huawei, stunting its ability to create powerful AI accelerator chips that compete with Nvidia’s.

“I am confident in saying that Nvidia is going to sell a lot of chips to China. That’s because Huawei chips are lousy. Why are they lousy? Because of export controls,” Allen said.

An Nvidia spokesman said in a statement to Semafor that “the suggestion that Biden’s export controls ‘helped’ the U.S. semiconductor industry reflects a fundamental lack of understanding ecosystems and competition — export controls only fueled America’s foreign competitors, deprived the U.S. taxpayer of tens of billions of dollars, and undercut America’s influence around the world.”

In a sense, Nvidia will be getting the benefits of export controls while also benefiting from their premature end. That prospect makes some critics livid.

“You can’t argue that you need zero oversight because we’ve had a ‘Sputnik Moment’ and we are now in a march-or-die race for AI supremacy while simultaneously arming that very enemy,” said Steve Bannon, former chief strategist for President Trump, in a message to Semafor.

“The Oligarchs are men without a country — they will sell anyone or anything to the highest bidder — Jensen Huang is nothing more than an arms dealer,” he added.

The Nvidia spokesman said its chips are “not weapons, arms, or atomic bombs. As the President recognizes, American national and economic security benefit whenever approved commercial businesses can buy American.”

With the ability to purchase H200 chips, China will still lag behind US hyperscalers in the ability to build massive clusters of compute power used for training the most powerful frontier AI models.

But they could still offer “inference” on those models with less powerful chips, selling AI services to consumers in the form of chatbots and “agents” useful to big companies. Chinese companies like DeepSeek have shown how quickly they are able to build models with advanced capabilities, often using American models as inspiration, or even as a way to generate quality training data in a process called distillation.

Companies like OpenAI are losing billions of dollars a year on inference, banking on costs coming down, making the products it offers more profitable. But if Chinese firms push prices down further, it could create challenges for US AI firms.

US firms have courted countries like the United Arab Emirates and Saudi Arabia, offering to build compute clusters to serve populations there. But so far, those companies have had little serious competition from outside the US.

History shows China is willing to subsidize key industries in order to control the market. One example is solar panels, which were pioneered in the United States and are now dominated by Chinese manufacturers after US firms couldn’t compete with Chinese prices that were artificially low due to subsidies.

Solar panels, though, don’t compare to AI compute clusters in complexity. There’s no software “lock in” that protects incumbents and first movers in solar. In AI, Nvidia has been able to hold onto its lead by building an ecosystem that includes its CUDA software, the de facto standard way the vast majority of people build AI models.

China is right on the heels of the US in terms of AI capability, but US companies are gobbling up market share all over the world, building “AI factories” to enable a new technology wave.

President Trump appears motivated, in part, by an impulse that US chip companies — key to the country’s economy and national security — are being held back from a lucrative market in China.

But few believe American tech would ever become the standard tech stack in China.

Indeed, things are going pretty well for American AI, and export controls have undoubtedly helped, even if you believe that they pushed China to speed up its inevitable technological independence.

What people on both sides painted as a clear picture has now grown muddled. The worst-case scenario is that the new exports don’t actually do much to keep Chinese companies on the Nvidia stack, and even help China in some way. But the worst-case scenario is not catastrophic. The technology race with China is a long one, the US will succeed more by moving faster than by trying to hobble its rival.

Ben Thompson noted in Stratechery that the import of H200 chips is not the radical move that some say it is. It actually has historical precedent. “It’s actually a return to the way the U.S. managed technological exports for decades, under what was known as the ‘Sliding Scale,’” he wrote. “The overarching goal has traditionally been to preserve U.S. dominance on the leading edge without creating the conditions for that dominance to be superceded by alternatives created elsewhere, which is not only a national security threat, but also reduces the competitiveness of U.S. firms by denying them the revenue that fuels R&D.”

Thompson wrote that the status quo just wasn’t working: “The reality is that right now we are in the messy middle: we’re not actually stopping Huawei from building a system that is capable of doing large language model training (albeit inefficiently), but we are hurting the fortunes of a U.S. AI champion and limiting their long-term competitiveness. And, of course, we are incentivizing everyone in China, from the government to private enterprise, to ultimately remove the point of leverage that we can’t even wield properly.”