In a July 2025 interview with CNBC, Commerce Secretary Howard Lutnick defended the Trump administration’s abrupt decision to reverse a ban on Chinese purchases of the American technology company Nvidia’s H20 semiconductor, a powerful chip for running advanced AI models. “You want to sell the Chinese enough that their developers get addicted to the American technology stack,” he explained. In Lutnick’s invocation of addiction, Chinese commentators heard echoes of the Opium Wars, a major episode of imperialistic exploitation in China’s historical memory. Officials called for the country to remain vigilant against “insulting” and enfeebling imports.

The aspiration to make China dependent on American semiconductors did not originate with Lutnick. Jensen Huang, Nvidia’s CEO, has avoided such phrasing but estimated that Nvidia could earn several hundred billion dollars “by the end of the decade” in China—a windfall even for a company that recently broke all records with a $5 trillion market capitalization. Sensing opportunity with Trump’s reelection, Huang spearheaded an all-out assault against export controls on chips that Nvidia wanted to sell to China, attending a $1-million-per-head dinner with Trump at his Mar-a-Lago resort, launching a lobbying blitz in Congress, and pledging to manufacture up to $500 billion worth of AI technology in the US. Trump took note and invited Huang to join him on a trip to the Middle East in May 2025, where the president called him a “friend” at a major investment forum in Saudi Arabia.

This campaign culminated in Huang being invited to an Oval Office meeting with Trump in July, where he secured an initial prize: the United States would permit the sale of H20s to China. In exchange the US government would take a 15 percent cut of the resulting revenue, an arrangement that legal experts quickly criticized as a violation of the Constitution’s prohibition on export taxes. Others stood to profit, too: Trump personally owned as much as $1.3 million in Nvidia stock as of the end of 2024, and before becoming commerce secretary Lutnick was chairman of the financial services firm Cantor Fitzgerald, which had a reported stake in Nvidia worth $1.19 billion as of late 2024. His two twentysomething sons now run the firm.

Fresh from the Oval Office, Huang hopped on a plane to Beijing and announced triumphantly, “We will start selling H20s to the Chinese market very soon, and I’m very happy about that.” He praised the cutting-edge AI models built by Chinese firms like DeepSeek and Alibaba and delivered his sales pitch: “Remember, H20, although it isn’t the most advanced [chip] that Nvidia provides, is still extraordinary in terms of capability.”

Trump’s reversal was remarkable because it undercut the strategic logic of years of American policy toward China. Beijing has been working to produce advanced chips domestically for over a decade, including under the Made in China 2025 industrial strategy, which President Xi Jinping introduced in 2015. But China made limited progress in catching up to the cutting edge and thus resorted to buying the most advanced chips from abroad. The growing clusters of advanced semiconductors that enable ever more powerful AI systems remain the most significant American advantage in this competition, even as China has caught up to or surpassed the US in areas like engineering talent and energy generation and management.

The first Trump administration placed restrictions on US sales of semiconductors to China’s leading technology conglomerate, Huawei, in May 2019. In October 2022 the Biden administration (in which I helped lead China policy) introduced broader controls that prevented the most advanced US semiconductors and chipmaking equipment from being sold to China. The global AI boom caused demand for Nvidia’s chips to far outstrip supply, and the company’s stock price soared. In 2024 the Oracle cofounder Larry Ellison recounted a dinner at which he and Elon Musk, the world’s richest person, begged Huang to sell them more chips. “Please take our money,” Ellison said. “We need you to take more of our money, please.” Even so, Nvidia complained bitterly about its lost sales in China until the Trump administration removed export controls at its behest.

Although many Chinese AI firms were eager to purchase the H20, Chinese regulators had no intention of allowing them to become “addicted” to American technology that could be cut off again at Trump’s whim. In the weeks after Lutnick’s comments, China’s top Internet regulator ordered tech companies not to acquire the H20 and alleged that the chips had “serious security issues.” Though Chinese-made chips perform less well than their American-made counterparts and China cannot make nearly as many of them, Beijing soon issued guidance that tech firms should avoid purchasing Nvidia chips and instead double down on domestic alternatives. It subsequently banned foreign chips from new state-funded data centers. Nvidia frantically pushed back, but Chinese state media asserted, “When a type of chip is neither environmentally friendly, nor advanced, nor safe, as consumers, we certainly have the option not to buy it.”

These actions should not have come as a surprise. Earlier in 2025 Xi had instructed China’s Politburo to “concentrate resources to overcome challenges in core technologies such as high-end chips” and to ensure that China’s AI system was “independent” of undue reliance on foreign suppliers, even though the CEO of DeepSeek, Liang Wenfeng, had complained that the “embargo on high-end chips” was limiting his firm’s success. Chinese premier Li Qiang admitted that the “insufficient supply of computing power and chips” was a bottleneck for the country’s AI sector, and DeepSeek experienced delays in its second model release because Chinese authorities pressured it to use Huawei rather than Nvidia chips.

The contrast is striking: in Trump’s America, one company has captured the country’s China policy, but in Xi’s China, the party controls the country’s AI sector. Trump is so focused on profiteering that he is apparently willing to sacrifice long-term American competitiveness, while Xi is so obsessed with control and self-reliance that he may be willing to sacrifice near-term Chinese competitiveness.

In a strange coincidence, Nvidia’s market capitalization crossed the $5 trillion mark the day before Trump and Xi met in South Korea on October 30, 2025, to call a truce after months of economic brinkmanship. Trump indicated ahead of the meeting that he might pitch Xi on buying an even more advanced type of Nvidia semiconductor, the Blackwell, which he called a “super-duper chip.” It is many times more powerful than the H20—or anything that Huawei can currently produce. Rumors swirled that Trump might bring Huang to the meeting with Xi. When asked later about the reports, Huang sounded sorely disappointed. “I missed—I tried to get to Korea as fast as I could,” he said, but “unfortunately he was already finished.” Trump told reporters that he and Xi did not discuss these chips and that at least for now, this super-duper chip would not go to China.

Yet Nvidia’s campaign was far from over. In early December 2025, after another meeting with Huang, Trump announced that he would authorize the sale to China of Nvidia’s second-most-powerful chip, the H200, and take 25 percent of the revenue. As of this writing, Chinese regulators are still assessing whether to permit eager AI firms to have limited access to the chips, which are six times more powerful than the H20. If sales proceed, this will be a revealing acknowledgment that the H200 is too good to turn down and that China is a long way from producing anything comparable, but Beijing remains focused on preventing Chinese firms from relying entirely on Nvidia over domestically produced chips.

Both governments are in curious positions. The US government under Trump is hawking the products of the world’s most valuable company to Beijing after years of trying to block China from gaining access to them. And the Chinese government is trying to force Chinese AI firms to resist the seductions of technology that is far better than what domestic chipmakers can supply at scale.

With a determination underappreciated in Trump’s Washington, Xi decided long ago never to let China’s AI sector become “addicted” to American technology. He sees the risks of dependence as outweighing the benefits; any lead in AI that isn’t grounded in China producing its own physical components is too fragile to be worth having. And he believes that Chinese tech firms can innovate even if they do not have topflight American chips, in keeping with his long-standing view that periods of hardship make China stronger. It may seem contradictory, but because AI is so important to Xi—he has called technology “the main battlefield of international competition”—he wants to force the country’s tech firms to create a self-reliant system, even if it means slowing their near-term progress, in order to gain a more durable long-term advantage. Even as Trump approved sales of the H200, Xi considered providing as much as $70 billion in additional support for China’s chip industry, in part to accelerate his effort to produce advanced semiconductor manufacturing machinery as good as that of the global leader, the Dutch firm ASML. But this equipment, often called “the world’s most complex machine,” may be the single most challenging technology to replicate domestically, far harder than solar panels, electric vehicles, or other Chinese success stories. It is a high-stakes gamble.

To understand why Xi is making these decisions, there is no better guide than the Harvard sociologist Ya-Wen Lei’s The Gilded Cage: Technology, Development, and State Capitalism in China. Her masterful study of what she calls China’s regime of “techno-development”—the model of “[science-and-technology]-oriented socioeconomic development” that has gradually replaced the prior model of “labor-intensive, export-oriented manufacturing”—makes clear that China’s leaders see advanced technology as the country’s lifeblood:

Chinese governments, especially at the upper levels, exemplify technology fundamentalism—an enduring belief in the ability of…scientific methods along with high and new technologies, to solve socioeconomic, developmental, and political problems.

Relying on a hostile US to sustain an area so fundamental to China’s development would make no sense.

Lei focuses on the relationship between the Chinese state and large technology companies like Tencent, Baidu, Alibaba, Huawei, and ByteDance, which work on everything from online search engines and social media to digital payments and advanced large language models. She describes them as “cobuilders” of the tools and systems that create “an arsenal of technological and legal instruments” that increasingly forms the basis of China’s power, prosperity, and security.^*

The Chinese state has unquestionably come to dominate these firms during the past decade. Through extensive fieldwork in China, including interviews with employees at leading tech companies, local government officials, and gig workers, Lei exposes the evolution of this relationship. Initially, the state’s demands of the tech giants focused on censorship; they were mostly left alone as long as they removed any content related to the Chinese Communist Party’s voluminous list of forbidden topics, from the brutal history of the Tiananmen massacre to lighthearted images of Winnie the Pooh reminiscent of the portly Xi. But as these firms became larger and reshaped more of China’s economy and society, Xi grew worried. Lei writes that he began to fixate on “preventing and preparing for worst-case scenarios,” ranging from a potential financial panic exacerbated by rampant and underregulated digital lending to foreign cyberattacks on Chinese data and technology infrastructure. (Xi’s anxieties about the latter are particularly understandable because, during his time in power, China has mounted enormous cyber campaigns against precisely these types of targets overseas, including in the US.) The pressure on Chinese tech firms to address these risks increased sharply as the country’s economy slowed and its international relations became more fraught in the late 2010s.

A surprising act of resistance led to the final co-optation of China’s tech sector. On October 24, 2020, Jack Ma—the founder of the e-commerce giant Alibaba and China’s most famous technology leader—delivered a speech at a conference in Shanghai that took barely veiled swipes at Xi’s approach. “To make risk-free innovation is to stifle innovation,” he declared. But his risk-taking backfired spectacularly. Within two weeks the planned public offering of Alibaba’s financial services platform, Ant Financial, had been suspended, and Ma had been hauled in by the government to be set straight. Beijing soon launched a huge “regulatory blitz,” Lei writes, spanning antitrust, financial risk, data security, algorithmic security, and labor conditions, which was designed to ensure that China’s most important firms were unquestionably subordinated to the needs of the techno-development regime.

Of all Xi’s worst-case scenarios, the one that has come closest to occurring is the intensification of US efforts to limit China’s AI progress. In March 2023, a few months after the Biden administration introduced export controls on the most advanced chips and semiconductor manufacturing equipment, Xi broke with long-standing protocol and publicly railed against the US: “Western countries led by the United States have implemented all-around containment, encirclement, and suppression of China,” he said in a speech, “which has brought unprecedented severe challenges to our country’s development.” To his audience of officials, another subtext was clear: Xi’s focus on planning for risks had been proved right.

As Lei’s book shows, ensuring that China can continue to be a technological superpower despite US controls is what Xi might call “a matter of life and death” for the nation. “With rising US–China tensions,” she writes, “the conditions that aided China’s global ascendence and techno-development”—an integrating global economy and a cooperative United States—“have ceased to exist.” But Xi’s response to these new conditions is not to move away from techno-development. It is to stay the course and continue moving away from relying on the US.

American policymakers have not yet adapted to these changing dynamics in China, and they are struggling to connect China policy to shifting expectations about the evolution of AI technology. The policy that the Trump administration is eroding—blocking high-end US chips from going to China—has two primary justifications. The first is straightforward: China has been using American technology to improve its military capabilities, for example to power the supercomputers used to develop its increasingly impressive arsenal of nuclear-capable hypersonic missiles. Few would dispute that China should not have access to scarce, cutting-edge US chips with which it can design weapons to threaten the US.

The second reason is more contested. Many prominent voices in Silicon Valley predict that AI advancements will occur very rapidly and may reach human-level artificial general intelligence (AGI) in just a few years. Widely read texts like the technology investor and OpenAI alumnus Leopold Aschenbrenner’s Situational Awareness (2024) depict a short-term race to AGI between the United States and China and argue that whichever country gets there first will reap transformational economic and military benefits. For example, Aschenbrenner predicts “superhuman hacking that can cripple much of an adversary’s military force, roboarmies and autonomous drone swarms…and the inventions of new WMDs with thousandfold increases in destructive power,” which will confer a self-improving strategic advantage that a rival may never catch up to. Many “AGI-pilled” experts also warn that China may not responsibly address the safety risks of advanced AI systems, as they hope the US will. If these claims are true, it follows that the US should, with the greatest urgency, use its semiconductor choke points to prevent China from reaching AGI first.

Yet in recent months, resistance to these ideas in the US has gathered force. Some AI experts raise doubts about the imminence of anything approaching a truly human-level AI, particularly after what some have described as disappointing results from OpenAI’s GPT-5. Others point out that even as AI systems grow more capable, there is not a bright line that developers will cross to arrive at AGI; different systems will have different capabilities and areas of strength, and the idea of a single self-improving system that confers enduring economic and military advantages may simply be unrealistic. Still others dismiss a focus on the risks of AI as “doomerism.”

Perhaps the most influential voice denouncing so-called doomer narratives is the Trump administration’s AI and crypto czar, David Sacks, who has also been pushing Washington “to update its assumptions with regard to export controls.” Sacks believes that the AI race is primarily about whether the world adopts US or Chinese technology, so he has argued that Nvidia should be allowed to sell its chips globally—and he often appears to share Lutnick’s outdated view that China could still become addicted to US technology. But the support of Sacks and others for Nvidia’s case also reflects a broader worldview embraced by the Trump administration: a willingness to conflate the interests of prominent American companies with the interests of the US. That Nvidia benefits from increased sales to China is indisputable; whether the US as a whole does is far more contested. Adding to the muddle, the Trump administration openly rejects a focus on “AI safety” in favor of no-holds-barred “AI opportunity” (in Vice President J.D. Vance’s words), undermining the putative contrast between the United States and China in addressing AI risks.

Sacks’s and Vance’s views echo those of Nvidia’s Huang, who has long been suspicious of AI pessimism. Two new books on the company—the journalist Stephen Witt’s The Thinking Machine: Jensen Huang, Nvidia, and the World’s Most Coveted Microchip and the business reporter and former equity analyst Tae Kim’s The Nvidia Way: Jensen Huang and the Making of a Tech Giant—show that Nvidia’s success has come in no small part from simply focusing on making the most competitive products rather than getting distracted by their philosophical or societal implications. Huang told his team in 2013 to “go all in” on deep learning, Kim reports, and decided to “pull every lever at his disposal to navigate Nvidia to the very center of the tech industry—as the company whose hardware could bring about the AI-powered future.” He was wildly successful but still refuses to entertain speculation about the risks of this future.

Witt’s incisive account, which is less hagiographic than Kim’s, describes how Huang “vocally rejected the whole framing [of AI risk], thought the question was stupid, and thought the people who argued about it were holding back humanity.” In an extraordinary closing chapter, Witt recounts his explosive final interview with Huang, who yelled at him for raising questions about potential job losses or other downsides of AI: “This cannot be a ridiculous sci-fi story…. This is not a freaking joke!” Recalling Nvidia executives he interviewed who were also disinclined to discuss “the potential future implications of the technology they were building,” Witt concludes that “they were more afraid of Jensen yelling at them than they were of wiping out the human race.”

Even if Huang thinks Nvidia should keep its head down and focus on engineering, his recent forays into US–China relations show that it was never realistic for the world’s most valuable company to stay out of politics. After all, geopolitical considerations are central to its business model in at least one way: the chips that it designs are primarily fabricated by Taiwan Semiconductor Manufacturing Company (TSMC), the Taiwan-based foundry that dominates high-end semiconductor production.

Huang refuses to consider the impact on his global business if China were to blockade or invade Taiwan, despite the regular threats emanating from Beijing to do just that. “To my great surprise, Nvidia had done no contingency planning for this eventuality,” Witt writes. He quotes Deb Shoquist, Nvidia’s executive vice-president of operations: “If something happens to Taiwan and TSMC, the ramifications are so large it’s almost like asking me what I’d do if California fell into the ocean.” Huang had instructed her “never to think about this question.”

But a refusal to think about the hazards of runaway technological development or catastrophic geopolitical events does not prevent them from occurring. We are midstream in two of the most significant strategic developments of our time: the intensification of US–China rivalry and the explosion of AI. In isolation, either would be world-changing, but their simultaneity and interconnectedness create unprecedented risk. The issue is not simply whether the United States or China reaches some theoretical end point first but what happens along the way. Some risks are technology-dependent—such as an out-of-control AI system that refuses human guidance or enables terrorist groups to design advanced weapons. Others are geopolitical, such as whether an unexpected technological breakthrough by one country provokes the other to launch a cyberattack to stymie or steal it, whether China lags behind cutting-edge US models but applies AI to further its dominance of global manufacturing, or whether one country’s military leverages advances in robotics and autonomy to puncture the other’s defenses. The list goes on. Nvidia and the AI sectors of both the US and China do not have the luxury of ignoring these questions.

As Trump and Xi continue their diplomatic dance in the coming months, the appearance of bonhomie will mask intense unease and persistent competition between the two superpowers. A temporary truce in their economic conflict does not mean a pause in the AI race. Ironically, leaders in both countries believe that time and momentum are on their side: US leaders think they will be able to outpace China in developing highly capable AI systems, and Chinese leaders think they will be able to leapfrog the US despite American choke points. Both sides cannot be correct. The day when one side realizes that its bet was wrong will be a moment of profound danger—and it may come sooner than any of us would like.

—January 28, 2026