The United States and China are racing to build ever-more-advanced artificial intelligence capabilities that will power the economies and, increasingly, the militaries of the future. To date, the U.S. has sought to restrict China’s access to America’s best AI technology, but some are now pushing to reverse that trend and expand U.S. exports of advanced AI chips to China. That would be a historic mistake.
As it stands, America is firmly in the lead of the AI race. U.S. software companies are responsible for class-leading AI models such as ChatGPT, while hardware industry leader Nvidia, the world’s first $4 trillion company, is producing the semiconductors required to power AI at a quantity and quality that Chinese companies can still only dream of.
Chinese AI companies are desperate to buy as many advanced Nvidia chips as they can to train and power their software models. The very best U.S. chips are barred from being sold to China by export controls. However, there has been a debate about what to do with chips such as Nvidia’s H20 that are designed to avoid export controls: less capable than America’s best chips but still superior in performance, reliability, or availability over China’s homegrown alternatives.
In April, the Trump administration banned the sale of H20s to China, prompting Nvidia CEO Jensen Huang to travel to Mar-a-Lago to try to convince President Donald Trump to overturn the ban. In July, Huang continued his lobbying efforts at the White House, convincing the president to reverse course and lift the H20 export ban. When the Commerce Department was slow to issue Nvidia export licenses, Huang again returned to the White House. In August, Trump agreed to expedite the licenses if the U.S. government got a 15% cut of H20 sales to China. Almost immediately after, reports revealed that Nvidia had already begun working on a considerably more capable B30A Blackwell chip for export to China, if Huang can persuade the Trump administration to grant the necessary licenses.
Some national security leaders have expressed alarm at these arrangements, arguing that America should not be helping Beijing close the AI gap by giving Chinese companies desperately needed computing capacity. When you’re winning the race, why give your top adversary the exact thing it needs to catch up?
Those wanting to expand chip sales to China have offered a series of counterarguments. First, they contend that selling China products such as H20 and B30A won’t undermine American AI dominance because Nvidia makes even more advanced chips for the U.S. Second, they argue that export controls have already failed because Chinese alternatives are already on par with U.S. AI chips. Third, they argue that the U.S. must sell China vast quantities of AI chips to keep China “addicted” to U.S. tech and prevent it from developing competing domestic alternatives.
Finally, Huang recently argued that those trying to prevent China from obtaining America’s best technology were somehow being “unpatriotic.” In a September interview, Huang claimed that experts advocating tougher policies against China, or “China hawks,” should view that label as a “badge of shame.” He also insisted the Chinese Communist Party wants an “open market” because “their leaders say it and I take it at face value.”
In our view, each part of this sales pitch is either misguided or demonstrably false.
First, while AI chips designed for China are less capable than the best available in the U.S. market, they are superior for many AI tasks compared to what Chinese competitors can mass produce. Nvidia’s B30A would be far more powerful than anything widely available in the Chinese market, accelerating China’s frontier AI advancements.
Second, no Chinese company — no company on Earth, in fact — has yet challenged Nvidia’s dominance in chip design and Taiwanese partner TSMC’s ability to manufacture the most advanced chips.
A top Commerce Department official recently declassified that China’s national champion, Huawei, can only make 200,000 advanced AI chips in 2025. Other estimates place the figure closer to 1 million chips, but still a fraction of the roughly 10 million AI-capable chips Nvidia will produce this year.
For now, banning sales of advanced AI chips to China won’t lead Chinese companies to turn to a domestic competitor for a simple reason: there aren’t any. They don’t yet have the quality, scale, and reliability to satisfy Chinese domestic demand, much less offer a competitive “AI stack” globally. Unlocking access to more American AI chips would only aid Chinese AI companies’ efforts to develop and commercialize competitive AI models.
Huawei and the Chinese government are throwing tremendous resources at developing a homegrown AI industry and are beginning to make progress. In fact, Huawei recently released its three-year vision to challenge Nvidia’s dominance. Chinese software models such as DeepSeek, largely trained on U.S. chips, are already increasingly competitive with U.S. models. In hardware, recent reports suggest China may triple its advanced chip output in 2026 — still far short of Nvidia’s output, but a gap that may well narrow in the future.
Nonetheless, export controls were never designed to be 100% effective indefinitely, but rather to protect and prolong a U.S. advantage during a critical period of AI development. And we have definitive proof that China’s progress is being slowed by a lack of computing resources.
“Money has never been the problem for us; bans on shipments of advanced chips are the problem,” DeepSeek founder Liang Wenfeng has admitted. More recently, DeepSeek delayed the release of its second model because it faced “persistent technical issues” trying to train its model on Chinese chips, which “suffer from stability issues, slower inter-chip connectivity and inferior software.”
Third, the notion that we can keep China “hooked” on U.S. technology by selling it more chips is fanciful. China has been sprinting toward an autonomous domestic semiconductor supply chain for well over a decade.
Since 2014, China has spent over $150 billion to indigenize its semiconductor industry. The “Made in China 2025” initiative that Communist Party General Secretary Xi Jinping unveiled in 2015 was a road map to technology decoupling from the West. The argument that Beijing would meaningfully decelerate Xi’s legacy multibillion-dollar quest for self-sufficiency if we sell China more advanced chips strains credulity.
The Chinese Communist Party will not allow long-term addiction to critical products from the U.S. In 2020, Beijing tried to force Chinese companies to adopt domestic chips. The same year, Xi personally instructed the Chinese government and industry to pursue “import substitution” to end reliance on foreign technology and “build a domestic supply system that is independently controllable.”
Earlier this year, Xi instructed top CCP officials to “persist in being self-reliant” in AI technology and focus on “mastering core technologies such as high-end microchips … and construct an autonomously controllable” AI ecosystem. Huawei and other Chinese semiconductor companies have leaped to obey these directives, announcing plans to replace Nvidia products. And the CCP has begun taking overt steps to eject American AI hardware from its market, banning its own companies from purchasing Nvidia’s H20 and RTX Pro 6000D, which the CCP wants Chinese companies to replace with less-proven or less-capable homegrown alternatives, while keeping the door open to import chips China can’t yet replicate, such as the more advanced B30A.
For now, Chinese AI companies are seeking access to as many American chips as possible until Huawei can catch up. At that point, Nvidia will undoubtedly be discarded like the countless Western companies chewed up and spit out by the CCP after gambling that technology transfer would be rewarded with permanent market access.
There is a final, particularly compelling reason to limit advanced chip exports to China: There simply aren’t enough chips to go around. In the words of Nvidia’s CEO, “everything is sold out.” Nvidia has since sought to walk back this claim, but anecdotal evidence suggests major U.S. tech firms are buying all the advanced chips they can get their hands on, and they want more. Ultimately, every manufacturing run of China-bound B30A chips would require resources that can’t be used to meet surging demand from the U.S. and allied countries.
HOW CHINA CORRODES AMERICAN FREE SPEECH
Xi is driving the U.S. and Chinese tech ecosystems toward a divorce, no matter what policy decisions are made in Washington. It therefore makes little strategic sense to invigorate China’s AI industry, which is starved of computing resources, with scarce chips at a time when U.S. AI firms are hungry for more. Doing so would only help China build superior AI models, globalize Chinese AI products to compete with the U.S. “AI stack,” and aid China’s military modernization.
Ultimately, the national interest would be far better served by restricting China’s access to American computing resources, rather than expanding it.
Jeff M. Smith is the director of the Asian Studies Center at the Heritage Foundation.
Bryan Burack is Heritage’s senior policy adviser on China and the Indo-Pacific.