Inside the Biden Administration's Gamble to Freeze China’s AI Future

Alan Estevez was sitting at his dining room table wearing a T-shirt when Secretary of Commerce Gina Raimondo called on Zoom to ask if he wanted to be the Biden administration’s top export control official. “You’re going to have to sell me on this,” Estevez recalls telling her.

It was 2021, and the outspoken New Jersey native thought he had finally left public service behind. After more than three decades at the Pentagon, he had left and taken a job in consulting. He wasn’t sure if he was ready to go back.

Could he be tempted by the opportunity to help oversee the tens of billions of dollars in semiconductor funding the administration was seeking from Congress? “I came from DOD,” he recalls saying. “$50 billion is OK money, but it’s not a lot of money.” Then Raimondo appealed to his sense of service. Estevez gave in and took the job.

By the time Estevez was sworn in as under secretary of commerce for industry and security in the spring of 2022, he had his work cut out for him. The role he had accepted would turn out to be at the center of America’s first serious effort to confront the geopolitical risks of artificial intelligence. Within a few years, the project would reshape relations between the world’s two largest powers and alter the course of what may be one of the most consequential technologies in generations.

Shortly after he joined the Commerce Department, Estevez says, he began hearing from officials in the White House. Over breakfast one morning, Tarun Chhabra and Jason Matheny, two key figures working on technology and national security, told Estevez they were planning something big, and they would be needing his help.

Over the next six months, the US government revolutionized its strategy for competing with China, now widely considered the nation’s most important technological rival. For years, US officials had sought to keep China one or two generations behind in semiconductors, the building blocks of modern technology that power everything from smartphones to AI. But now, as National Security Adviser Jake Sullivan declared that September, “we must maintain as large of a lead as possible.”

On October 7, 2022, the Biden administration announced a sweeping set of export controls designed to cut off China from the most advanced chips used for training powerful AI models, as well as specialized tools that China would need to upgrade its own lagging chipmaking industry. Officially, the controls were intended to throttle China’s military modernization and stem human rights abuses fueled by surveillance technology. But as observers digested the dozens of pages of technical specifications and legalese, it became clear that the implications of the new policy were far more vast.

In practice, the United States was delivering a targeted blow that would ripple out across the Chinese economy, impacting research and development in every industry and scientific field that relied on computationally intensive machine learning. Anywhere cutting-edge AI or high-performance computing held promise—futuristic weapons, yes, but also curing disease and modeling climate change—the policy could undermine China. A New York Times writer called it a “declaration of economic war.”

War or not, the Biden administration was placing a high-stakes bet that the United States could use its leverage to hold China back and that the losses in terms of forgone US exports to China and collateral damage to bilateral ties would be worth it. On one hand, it was a gamble that relied on ideas established in Washington decades ago. US policymakers had been using tech restrictions to stymie China’s military modernization and punish the country for human rights abuses since the Cold War. More recent advancements in missiles and surveillance technology reinforced that logic. But several people who served in the Biden administration say that a more novel concern was also behind the big bet.

Key officials believed AI was approaching an inflection point—or several—that could give a nation major military and economic advantages. Some believed a self-improving system or so-called artificial general intelligence could be just over the technical horizon. The risk that China could reach these thresholds first was too great to ignore.

This account of how the Biden administration chose to respond is based on interviews with more than 10 former US officials and policy experts, some of whom spoke on the condition of anonymity to discuss internal government deliberations.

When the Biden administration introduced its transformative policy, it did not start from scratch. During his first term, President Donald Trump had also targeted Chinese tech, including semiconductor firms, as part of a broader effort to curb the country’s technological rise and global influence.

In 2019, the Commerce Department added the Chinese IT giant Huawei to its Entity List, which effectively cut it off from US supply chains, including chips, unless it got a special license. Officials justified the measure with allegations that Huawei had violated US sanctions on Iran. But experts believed they were also trying to undermine the company more generally, fearing that Huawei’s exports of 5G wireless infrastructure around the world could give Chinese spies and saboteurs a leg up.

Then the Trump administration doubled down, this time by turning to an obscure legal provision called the “foreign-produced direct product rule.” The FDPR was originally designed to make sure that goods made through US innovation and technology—like missiles or airplane parts—didn’t go into weapons systems sold to adversaries, even if those systems were built abroad. In 2020, the Trump administration turned this long-arm tool on Huawei, explicitly targeting the company’s “efforts to obtain advanced semiconductors developed or produced from US software and technology,” as Commerce Secretary Wilbur Ross said at the time.

While the FDPR had previously been used to enforce multilateral arms controls, the move against Huawei targeted “items made with US technology that were not sensitive, that were not on the control list, that had nothing to do with any AI,” says Kevin Wolf, a former Obama administration export control official.

“Everybody thought that would be the end of this very novel extraterritorial control,” Wolf added. Instead, the US government found the FDPR irresistible. It would later turn it on Russia after the 2022 invasion of Ukraine and would eventually wield it to constrain high-powered computing in China. “Obviously we started using it like candy,” says Estevez. “Certainly threatening to use it if not actually using it.”

In the first Trump administration, the US added Semiconductor Manufacturing International Corporation (SMIC), a leading Chinese chipmaker, to the Entity List and brought criminal charges against another Chinese semiconductor company, Fujian Jinhua (a judge later found it not guilty). It also worked with the Netherlands, home to the chipmaking tool giant ASML, to restrict exports of especially cutting-edge semiconductor equipment. All these moves were, in a way, part of normal business amid worsening ties between the US and China. They didn’t require visions of advanced AI to make sense in Washington, where traditional concerns about military modernization, rights abuses, and fair markets were more than enough to justify targeted efforts against Chinese tech.

When Biden took office in 2021, some members of his policy team brought with them a novel anxiety: AI systems might become so capable that they would supercharge national power, leaving rival militaries and economies at a great disadvantage.

In early 2019, when you had to be a bit of a geek to have heard of OpenAI’s early GPT models, Matheny—one of Estevez’s early breakfast partners from the White House—was founding director of the Center for Security and Emerging Technology. Essentially a think tank housed inside Georgetown University, CSET was launched with a $55 million grant from Open Philanthropy, a group aligned with the effective altruism movement that funds work addressing issues ranging from “potential risks from advanced artificial intelligence” to global health. “AI and other emerging technologies will deliver profound benefits to society, but they will also introduce new risks,” Matheny said at the center’s launch.

It wasn’t long before CSET scholars identified China’s growing strength in AI development as a US national security risk. Those scholars included Tarun Chhabra—the other White House breakfast partner—as well as future Biden administration officials Saif Khan and Ben Buchanan, all of whom studied technology and national security at CSET.

Chhabra coauthored a report published in February 2020 that recommended the United States and key allies “coordinate on the terms under which they will export leading-edge AI chips to China” and “forge a democratic way of AI.” That April, Khan coauthored a separate paper arguing for putting controls on chipmaking equipment to ensure that China remained “reliant on democracies for state-of-the-art chips.” In August, just months before Biden was elected president, Buchanan identified computing power as a key input that allows countries to develop AI and warned that the US was likely running out of time before it could no longer restrain China’s access to it. The National Security Commission on Artificial Intelligence, a research panel created by Congress that included Matheny, also called for “implementing coordinated export controls on advanced semiconductor manufacturing equipment with allies.”

Chhabra, Matheny, Khan, and Buchanan—four experts with varying backgrounds who crossed paths at CSET—would all go on to work on AI policy in the Biden White House. Once there, they and others, including Chris McGuire, a State Department official who had worked on the final AI Commission report, set to work operationalizing their ideas. One former official recalled that their work to keep advanced chipmaking equipment out of Chinese hands began as early as the transition. The theory was that gatekeeping the tools would ensure there would be no Chinese version of Nvidia and would prevent Chinese companies like SMIC from building cutting-edge factories. This would, as Khan had advocated, keep Chinese AI developers dependent on US suppliers.

As the Biden team assembled in 2021, however, its ambitions were broader. Limiting China’s chipmaking abilities would block it from independently building computing power, but Chinese developers could still simply buy the best chips from US companies. As close observers of AI were debating the potential of large language models, no one knew for sure what might be possible in the future with vast quantities of what were then the most powerful chips. Some officials believed that keeping China dependent on US supplies would no longer be enough. It was time, they judged, to build support across the administration for cutting off the chips themselves.

Former officials recall being motivated by a combination of different risks. Some say the idea of artificial general intelligence was explicitly on their mind. (While there is no universally accepted definition of AGI, the term generally refers to an AI system superior to humans across all intellectual fields.) Others say they were concerned more about specific AI capabilities, whether things that had already been envisioned—supercharged propaganda, automated cyber hacking, synthetic bioweapons—or ones yet to be foreseen. Others say their worries were not even directly related to AI, pointing to reporting that a Chinese company may have used US software to develop chips that in turn powered a supercomputer used to develop hypersonic missiles for the People’s Liberation Army. The stakes were especially high, some believed, because a powerful AI system might be used to develop even more capable AI systems, compounding a country’s lead and leaving competitors not just incrementally behind, but wallowing in a bygone paradigm. If some of these fears sound vague or far out, that’s because they are. Policymakers, however, are forced to take action in the face of uncertainty, and here, the uncertainty itself was cause for alarm.

The prescription that officials came up with to meet this challenge—a broad effort to limit China’s independent access to computing power—would require buy-in from across the federal government, including several cabinet secretaries and the president himself.

Not everyone was on board at first. The White House team had identified risks that demanded action, but there would also be costs. Nvidia, one of the crown jewels of US tech, would be banned from selling some of its most lucrative products to an enormous growing market. Producers of chipmaking equipment, such as Applied Materials and Lam Research, would lose the ability to sell their advanced products to Chinese factories. Experts believed that China might still eventually succeed in developing its own independent semiconductor ecosystem, and cutting the country off could motivate actors around the world to build an America-free supply chain to avoid future disruptions.

Former White House officials say they worked intensively to understand the potential blowback. The Chinese government’s reaction to the move would clearly be negative, but no one could know what form it would take. Would Beijing target iconic US firms like Apple? Cut off supplies of critical minerals? If China retaliated hard, what options would Washington have?

Then there were the specifics. How would the policy distinguish between equipment that really posed a risk, and products companies should still be able to sell? Estevez says he remembers the White House pushed for restrictions on a larger number of items, while the Commerce Department, which is responsible for promoting economic growth, sought a more tailored approach. “Trying to hold China back is a fool’s errand,” Raimondo, the commerce secretary, told The Wall Street Journal toward the end of Biden’s term, describing export controls as mere “speed bumps” for China.

Yet the administration kept plowing forward. Several former officials specifically cited Chhabra’s bureaucratic skill and determination as central to making the chip strategy happen. “American technology should not enable adversaries to build AI capabilities that will be turned against American troops, strategic assets, and critical infrastructure,” says Chhabra, now out of government and leading national security policy at Anthropic. “Strong export controls are essential for America's national security and AI dominance.”

It’s not unusual for a group of scholars with a bold new vision for policy to join the government, but it’s far less common for their ideas to be put swiftly into action. “Look, Tarun and I argued all the time,” says Estevez, but “moving in the same direction was not the issue.” At least among this group of staffers, the core dispute wasn’t over whether they should try to constrain China, but over how—broad restrictions versus targeted measures that preserved more flexibility for industry.

Finding that balance has been a moving target. After the first round of controls in October 2022, the Biden administration decided it needed to further tighten restrictions. Officials had already banned Nvidia from selling its best AI training chip to China, but the company then developed a new, China-specific chip with capabilities that pushed right up to the limits of the existing rules. In October 2023 and December 2024, the Biden administration tightened the controls on both chips and chipmaking equipment to plug what were perceived as unintentional loopholes.

To make any of this stick, however, the Biden administration first needed help from Japan and the Netherlands. Keeping advanced chips out of the Chinese market was a relatively discrete task, targeting just a few products. Undermining Chinese efforts to build cutting-edge chips of their own, on the other hand, was a multinational endeavor. That’s because semiconductor fabrication relies on precision machinery and software from around the world, with particularly crucial inputs coming from the Dutch company ASML and Japanese companies such as Tokyo Electron. If the United States banned its equipment suppliers from selling to China, but Japan and the Netherlands kept selling, US businesses would lose revenue, and China would still be able to upgrade its domestic manufacturing.

The Biden administration had sought Japanese and Dutch cooperation at the outset, but there was no quick agreement. So the White House decided to go it alone and announced the 2022 controls before the allies signed on, knowing full well that the move would hurt US companies. The Biden administration then had to convince Tokyo and Amsterdam that joining the effort was worth losing some exports and risking Chinese retaliation. After decades at the Defense Department, Estevez was well aware that AI represented the future of warfare, he says. Whether or not an AI inflection point was coming, he knew military planners would still prefer to face a Chinese adversary that was lagging behind technologically. This idea seemed to also carry weight with allied officials. “The sales pitch to the Dutch and the Japanese was: Artificial intelligence is the future,” says Estevez. “And they bought that.”

One way or another, the persuasion worked. In January 2023, Japanese and Dutch national security officials joined Sullivan at Blair House, the traditional guest quarters for visiting dignitaries across from the White House, to sign an agreement to institute parallel export controls on chipmaking equipment sold to China.

Today, Chhabra works at Anthropic, one of the most well-funded AI startups in the US. Buchanan is a professor at Johns Hopkins University. Khan is a fellow at the Institute for Progress, a think tank focused on innovation. Matheny left the administration in 2022 to become president and CEO of RAND, a storied research organization that often serves government clients. They are each, in various ways, still working on AI, computing, and national security. And they’re watching as the Trump administration pulls the country in seemingly conflicting directions.

On one hand, most of the semiconductor controls the Biden team enacted have remained in place. Earlier this year, Trump even put restrictions on another Nvidia chip called the H20, which has proven useful for certain kinds of AI development tasks. Several Biden officials said they would have also moved to cut China off from H20s if they had more time in office.

But after Nvidia CEO Jensen Huang reportedly began lobbying Trump, the White House abruptly reversed course. In a legally questionable move, Trump said he would authorize licenses for Nvidia to sell H20s to China in exchange for a share of the revenue.

Buchanan says the move undermined what Republicans and Democrats had both agreed was a key national security priority. “Reversing the effective bipartisan course on China AI chip controls, when those chips matter more than ever, is a unilateral surrender,” he says.

Trump, however, has described the situation very differently. “The H20 is obsolete. You know, it’s one of those things, but it still has a market,” he said at a press conference on Monday. “So we negotiated a little deal.”