On July 23, 2025, the Trump administration released its strategy for outcompeting China and winning the artificial intelligence race. Central to the plan was strengthening export controls to keep America’s most powerful AI chips out of the reach of the Chinese Communist Party (CCP) and the People’s Liberation Army (PLA). It was a sensible recommendation, not only because of Beijing’s dystopian vision for AI, but also because of the technology’s dangerous potential to be used in military applications.
But, less than a month later, the administration sabotaged its own strategy and endangered America’s ability to win the AI race. On August 11, President Trump confirmed that he had allowed Nvidia, the world’s most valuable company, to sell H20 chips to China in exchange for a 15 percent cut of the profits—or, as Washington Post columnist Megan McArdle put it on The Dispatch Podcast, “Fifteen percent of the rope we’re selling to hang ourselves.” (Advanced Micro Devices, another U.S. technology company, inked a similar arrangement with Washington.)
Putting aside the constitutional questions surrounding this unusual arrangement, the U.S. government now has a stake in selling critical defense technology to its foremost geopolitical adversary. A few billion dollars in the federal coffers pales in comparison to the computing power Beijing will accrue from these chips. And contrary to Nvidia’s arguments, the H20 is a significant upgrade for China’s AI models and risks materially aiding the very technology the PLA is planning to use against the United States.
A baseline best practice in grand strategy is not unilaterally ceding a competitive advantage to the enemy. For the United States, AI computing is such an advantage. Beijing holds many advantages in talent and energy, but U.S. innovation outclasses Chinese companies like Huawei in chip design and processing speed.
Earlier this year, Nvidia revealed that the U.S. government had slapped export controls on its H20 chips “for the indefinite future” to protect this innovative edge. Policymakers also had a specific concern that “the covered products may be used in, or diverted to, a supercomputer in China.” The U.S. government knows the value of militarized supercomputing; in June, Pentagon officials publicly admitted to wielding the technology to develop the GBU-57 bombs, or “bunker busters,” used to target Iran’s nuclear sites of Fordow and Natanz. Arming the PLA with the requisite computing power to optimize strike packages on the U.S. military and allied forces borders on suicidal.
Commerce Secretary Howard Lutnick seemed to understand this in a June interview with CNBC, when he boasted about America’s AI edge and assured Americans that the Trump administration wouldn’t surrender the nation’s tech superiority.
Americans are the greatest inventors in the world. We are the greatest thinkers in the world, the greatest entrepreneurs in the world. All the greatest chips in the world are American, right? So, of course, [the Chinese] want them and of course they want to use them. And of course, we said absolutely not. Did they ask? Did they ask 20 times, 30 times? Of course they did. They do every time. And we always say no.
Until, of course, the administration said yes.
Nvidia CEO Jensen Huang played an outsize role in flipping the Trump administration’s position. Thanks to his regular access to the White House, Huang made the case to President Donald Trump that America’s AI export controls had failed, and that selling chips to China could keep Beijing dependent on outdated U.S. technology. It’s a good story, but one divorced from reality. There are no fact-based national security arguments for selling AI chips to China.
First, the export controls. In 2023, the CEO of the Chinese AI company DeepSeek lamented the effectiveness of America’s export controls: “Money has never been the problem for us; bans on shipments of advanced chips are the problem.” A senior executive at Chinese tech giant Tencent echoed this frustration in April 2025: “The most severe problem is the [limited] resources of [graphics] cards and computing resources.”
Next, the chips themselves. The H20 is not, as Lutnick puts it, Nvidia’s “fourth best” chip. Nor is it, as Huang told Trump, obsolete. It is the company’s most powerful semiconductor when it comes to inference, the ability of an AI model to make conclusions or recommendations based on training. It outclasses Nvidia’s H100 chip, which the Department of Commerce has blocked from China. And—as Republican Rep. John Moolenaar, chairman of the House Select Committee on the CCP, warned Lutnick in July—the H20 chip outperforms domestic alternatives in China.
Finally, the military dimension. During a recent conversation with CNN’s Fareed Zakaria, Huang insisted that “we don’t have to worry about” China’s military using American-made chips. “They don’t need Nvidia’s chips, certainly, or American tech stacks in order to build their military,” he said. Shortly thereafter, a group of former Trump administration officials and national security professionals (including this author) wrote a letter to Lutnick warning him that Huang was wrong: “While the biggest buyers of Nvidia’s H20 chip are nominally civilian companies in China, we fully expect the H20 and the AI models it supports to be deployed by China’s People’s Liberation Army.” Days later, news broke that the PLA had submitted multiple procurement requests for the H20 chip and other Nvidia hardware over the past year.
Chinese leader Xi Jinping has long emphasized the importance of the military’s “intelligentization”: incorporating AI, big data, quantum computing, and emerging technologies into warfighting. During remarks to the CCP’s 20th National Congress in October 2022, Xi specifically urged the PLA to “speed up the development of unmanned, intelligent combat capabilities.” But China’s military had already begun to answer the call. A 2021 study from the Center for Security and Emerging Technology examined 343 PLA contracts that incorporated AI in concerning ways, including through its development of armed drones, intelligence-gathering capabilities, and information operations.
Today, the PLA’s investments in AI are stunning. According to a report by Strider Intel and the Special Competitive Studies Project, China brought eight AI data centers online in 2021. By the end of 2022, Beijing had added an additional 18. In 2024, China added 44 more data centers. These centers support large language models (LLMs), several of which serve military clients in China, including the PLA and the Central Military Commission. One such LLM is Tianji Military Model Platform, which supports applications like “counter strikes, ambush missile defense, [and] time-sensitive target strikes.” Consider also XSimVerse, which is fully integrated with China’s DeepSeek open-source AI model and holds more than 600 patents and software copyrights for “military intelligence products” and applications like drone training and low-altitude threat detection and response.
China’s military is also leveraging LLMs in cognitive warfare with models like Sichuan Daily Network Media Development, an AI platform for “international communication” to enhance “accuracy of content delivery” with overseas social media platforms—with accuracy, of course, being defined by the CCP. Earlier this month, researchers at Vanderbilt University warned that Chinese tech company GoLaxy was using “AI-generated narratives [to] shift the political landscape without drawing attention,” with a particular focus on members of Congress and U.S. thought leaders.
Beijing’s efforts to weaponize AI are material, not merely digital. The China-Cambodia joint military exercise “Golden Dragon 2024” featured a robo-dog with an automatic rifle affixed to its back. Beijing incorporated similar systems in drills with Laos, Malaysia, Thailand, and Vietnam in 2023. It has also armed aerial drones with automatic weapons. A scholar at the PLA’s top research institute recently characterized such robots as a replacement for human soldiers in urban reconnaissance and strike missions. While troubling, none of this is surprising. The CCP views robotics the same way it views all technology: as an extension of political power and policing. Mao Zedong infamously observed that “political power grows out of the barrel of a gun.” Robots are merely the party’s newest bullets.
Shareholder capitalism does not exempt U.S. companies (or the elite lobbyists representing them in Washington) from a sense of patriotism. American technology should serve Americans, not the Chinese Communist Party. It should protect the American people, not equip its enemies. In fairness, however, markets are not optimized for national security. That responsibility falls to the federal government and the country’s elected officials.
In this episode with Nvidia, policymakers are failing in their institutional duty to protect and defend the United States. The one entity tasked with guarding the national interest is selling it away for a negligible kickback. President Trump made this concession as a favor to an American company, and now Beijing smells blood in the water. China is suddenly expressing security concerns about the H20 chip in a potential bid to push for more advanced chips. Trump has already signaled an openness to approving the sale of modified Blackwell chips, Nvidia’s newest and most advanced AI semiconductor. Xi may not have to push hard if Huang is opening the door for him.During his Senate confirmation hearing, Lutnick pledged to be “very strong” on export controls: “When we say no, the answer’s got to be no.” If tech companies can turn Washington’s “no” into “yes” for the right price, then America’s China policy is for sale—and with it, the nation’s security.
