President Donald Trump recently released his plan for the United States to win the global race for AI dominance. The document has some good ideas about expanding domestic infrastructure, encouraging other countries to adopt American AI models, and imposing export controls on advanced semiconductor chips. But over the past few months, Trump has undermined his own goals and ceded much of America’s leverage to foreign powers. By setting the U.S. back in the AI race, he has created a host of strategic vulnerabilities that will bedevil future presidents.
This backsliding is the result of a rapid ideological shift within the administration, which two men in particular have spurred: David Sacks, Trump’s tech-billionaire AI czar, and Jensen Huang, the CEO of Nvidia and one of America’s most powerful executives in the industry. To trace their growing influence on Trump, consider Nvidia’s H20 chip.
In late 2023, Nvidia designed the H20 chip specifically for the Chinese market—a legal workaround to export controls that President Joe Biden had imposed. Nearly a year before the H20 was brought to market, OpenAI released a transformative large language model called o1, which employs the same kind of complex reasoning that the H20 chips were built to power. Practically overnight, the chips handed Beijing a significant competitive advantage. Biden was planning to outlaw their export to China but left office before he could. In April, Trump enacted the ban himself.
Around this time, the balance of power in the Trump administration began to tilt toward Sacks, who saw the H20 ban as counterproductive, both strategically and economically. He gradually gained a bureaucratic advantage: The right-wing provocateur Laura Loomer persuaded Trump to fire David Feith, an ideological opponent of Sacks who ran a directorate at the National Security Council focused on technology. The NSC itself was weakened and hollowed out. And, earlier this summer, the administration gutted the State Department’s “tech envoy” office, which had supported export controls.
Then, last month, Trump met with Huang in the White House. By this point, support within the administration for export controls had considerably softened, thanks in part to Sacks. Trump decided to lift the restrictions on the H20 chips, allowing their sale to China. Some observers assumed that the reversal was part of a trade deal and expected Beijing to offer some concession in return. But China insisted that Trump had made the decision unilaterally. Indeed, one day after Trump’s announcement, the country imposed new export controls on electric-vehicle batteries.
In effect, the U.S. gave away leverage to China and got nothing back. But Sacks and Huang have defended the decision. They have argued that the sale of H20 chips in China would make the country dependent on American chips rather than encourage Chinese companies such as Huawei to develop their own. As Sacks put it, “We can deprive Huawei of having this giant market share in China that they can then use to scale up and compete globally.” He credited Huang for “making the case publicly for competing in China, and there are a lot of merits to the argument.” (Left unmentioned was Huang’s obvious profit motive of selling his company’s chips in one of the world’s biggest markets.)
Their case is predicated on an unproven assumption: that China would otherwise be able to produce enough chips to compete internationally. In June, though, a senior Trump-administration official testified to Congress that Huawei would be able to produce only 200,000 chips this year—not enough to meet domestic demand, let alone keep pace with America. That’s not for lack of trying. Beijing has spent about $150 billion since 2014 to expand its chip-making capacity. But it still can’t make enough to equip a data center capable of training the most advanced AI models. The quality of China’s chips also lagged behind that of Nvidia’s.
Instead of hindering China, Trump’s H20 reversal bailed it out. The country already had a largely superior electrical grid compared with America’s, and is likely to be able to construct data centers more quickly. Its crucial shortcoming was computing power, which requires lots and lots of advanced chips. Now, thanks to the Trump administration, China is getting them.
Democrats rebuked the decision, and so did many Republicans. Late last month, 20 national-security experts—including Feith; Matt Pottinger, Trump’s former deputy national security adviser; and several conservatives sympathetic to Trump—sent a letter to the administration calling the H20 reversal “a strategic misstep that endangers the United States’ economic and military edge in artificial intelligence.” Steve Bannon, Trump’s former strategist, was less restrained in his critique. “American companies spent decades being made fools of, getting duped by the Chinese Communist Party transferring the crown jewels of our technology. For that they got nothing,” Bannon told the Financial Times. “Unbelievably, the government is poised to make the same humiliating mistake, at the behest of companies that want to drive their own profits with zero concerns for the nation’s security.”
The H20 decision was not an isolated case. In May, Trump announced deals with the United Arab Emirates and Saudi Arabia to build some of the world’s most advanced AI data centers on their soil. Some will be owned and run by American companies; others will be owned by local AI firms—Group 42 in the UAE and Humain in Saudi Arabia. Crucially, Trump also rescinded the Biden administration’s “diffusion rule,” which sought to limit the export of advanced AI chips and models. The move cleared the way for the UAE to import hundreds of thousands of Nvidia’s chips. Saudi Arabia is set to deploy a smaller number of Nvidia chips, but it has ambitions to expand its capacity.
Unlike Trump, Biden seemed to understand that compute, the processing power needed to train advanced AI, is a scarce strategic asset that should be concentrated in the United States or its most trusted allies. The Biden administration also recognized that China might use its close ties to countries such as the UAE to access advanced chips. Even worse, China could acquire the “model weights” of advanced AI—the parameters that dictate how a model operates, like neurons in the brain that decide how to respond to different signals. If bad actors get their hands on a model’s weights, they can reconstruct them for their own purposes.
Speaking in Saudi Arabia in May, Sacks acknowledged the importance of preventing chips from reaching “countries of concern.” But he suggested that this would be easy to accomplish. “All one would have to do,” Sacks said, “is send someone to a data center and count the server racks to make sure that the chips are still there.” It was a convenient dismissal. Counting hundreds of thousands of chips is no simple task, and regimes in the Middle East could decide to give China remote access to their chips. Sacks also did not mention the concern about model weights, which may prove to be an even greater vulnerability.
There are some signs that the Trump administration may be doubting its own decision. In June, Reuters reported that the UAE deal is “far from resolved,” according to five sources briefed on the project, because of outstanding questions related to security and enforcement. Four of the sources said “U.S. officials remain cautious about the UAE’s close relationship with China.”
There is an even more significant concern, though. Building what may be the world’s most important complex of data centers in the UAE—and, perhaps later, in Saudi Arabia—means placing some of America’s most important strategic assets in the world’s most geopolitically volatile region, within range of Iranian drones and missiles. For the Gulf states, these risks only sweeten the deal. If the U.S. senses that its vital infrastructure is in danger, it will be more likely to rush to their defense. (Indeed, Saudi Arabia has even proposed giving the centers the protected status of U.S. embassies, and the UAE could follow suit.) In this way, the data centers would offer a silicon shield for the Middle East nations, as well as grant them significant leverage in their relations with both America and China. Moreover, the deal threatens to pull the U.S. further into the region, at a time when successive administrations have tried to focus on the Indo-Pacific.
If AI becomes nearly as powerful as some of its inventors believe it will, the data centers it relies on must be built in America, where the government can better ensure that basic safety and national-security concerns are taken into account. Diverting massive quantities of advanced chips to subsidized data centers in the Middle East could make that functionally impossible. There just aren’t enough chips to go around.
Trump’s H20 reversal and Middle East deals could be just the beginning. The Financial Times recently reported that the Trump administration had “frozen restrictions on technology exports to China to avoid hurting trade talks with Beijing and help President Donald Trump secure a meeting with President Xi Jinping this year.” On a visit to China last month, Huang said, “I hope to get more advanced chips into China than the H20.” Beijing is pushing to ease more restrictions.
The Trump administration’s AI plan is a sophisticated document with some sound aims, but the administration’s recent actions have cut against them and made winning the AI race much harder. Providing China with advanced chips and prioritizing next-generation data centers in the Middle East over ones built in America could have enormous negative consequences, ones that subsequent administrations may not be able to reverse.
