The gap narrows

Last month, Stanford University’s Institute for Human-Centered Artificial Intelligence released a study finding that China is rapidly closing the AI gap with the U.S. “For years,” the researchers concluded, “the U.S. outpaced all other global regions on AI — in model size, performance, artificial intelligence research, citations, and more. But China emerged as an AI counterweight to the U.S., gradually gaining ground, and this year it appears to have nearly erased any U.S. lead.”

The institute discovered that, in February 2025, China’s DeepSeek-R1 model equaled the level of the most advanced U.S. AI and that, currently, Anthropic’s top model leads its Chinese counterpart by a mere 2.7%. That gap had previously been in the double digits, with some experts estimating it would have taken China more than seven months to catch up to the U.S.

In addition, and maybe more problematically, the researchers found that “China leads in publication volume, citations, patent output, and industrial robot installations” and that “the number of AI scholars moving to the United States has dropped 89% since 2017.”

How has this all happened? How does a country that invented an entire field and pioneered its signal breakthroughs now find itself on the back foot?

For one, the CCP has pumped tens of billions of dollars into AI research, outstripping even the U.S. A March study by the Association of American Universities found that, for the first time in history, overall Chinese research and development investment topped American R&D, $1.03 trillion to $1.01 trillion. A 2025 RAND survey concluded that Beijing seeks to develop a $100 billion AI industry by 2030 and has created an $8.2 billion fund solely for machine-learning startups.

Chinese AI firms also benefit from the so-called “second-mover advantage,” meaning they can piggyback on advanced models developed in the U.S. and make them far more efficient. For instance, China’s DeepSeek V3.2, introduced late last year, purports to incorporate an “efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios.”

But Xi and the state instruments he controls appear also to be exploiting American weaknesses, including through highly suspect methods.

Last month, I attended a hearing of the bipartisan House Select Committee on the Chinese Communist Party entitled “China’s Campaign to Steal America’s AI Edge.” Chairman Rep. John Moolenaar (R-MI) kicked off the proceedings by contending that “China’s smuggling of advanced AI chips is a pervasive threat facing law enforcement” and noting that “just last month, the Department of Justice announced a $2.5 billion chip smuggling case, which would be the largest export control violation in U.S. history.”

Moolenaar then asked, “Why is China so desperate to acquire U.S.-designed chips? The reason is obvious: AI is a truly transformative technology. It’s already changing how we fight wars, run our government, and operate companies.” He concluded that “it is essential for the United States to maintain a decisive lead in the AI race. We cannot afford a future where Beijing dominates this technology.”

At the hearing, Dmitri Alperovitch, the founder and chairman of the Silverado Policy Accelerator, echoed Moolenaar, asserting that “we are in a race, and the stakes could not be higher. Artificial intelligence will transform every industry, every battlefield, and every government.” Alperovitch also predicted that “whoever fields the best models running on the best infrastructure will likely win not just the AI race itself but the 21st century. The single most important input to winning is compute — the processing power used to train and run AI models.”

In addition, Yusuf Mahmood, the director of AI and emerging technology policy at the America First Policy Institute, reiterated that it used to be the case that “China’s frontier AI capabilities are generally considered to be about seven months behind America’s.” He also specified how Chinese actors have pilfered American AI breakthroughs, including through “distillation,” a process where users create accounts on pioneer models like GPT or Claude, query the models en masse, and Hoover up the data.

According to Mahmood, other examples of Chinese misconduct include stealing trade secrets and poisoning pioneer models by hacking into their training datasets and implanting false or faulty data. These attacks, he explained, already have dangerous effects.

“The American trade secrets, capabilities, and hardware that China steals directly feed its economy and Beijing’s military apparatus — including technology shipped around the globe to U.S. adversaries and rogue states,” Mahmood testified. “Just recently, the Iranian regime used weapons systems powered by Chinese AI technology to attack American warfighters.”

In short, China is quickly eroding the once-impregnable wall that the U.S. believed it had built around AI.

The battle over data centers 

But it’s worse than that.

China appears to be working hard to suppress future American AI breakthroughs by stoking opposition to the data centers that have sprouted like wildflowers across the country.

These massive complexes, which each comprise thousands of servers and graphics processing units, power the computing revolution. According to Goldman Sachs, American AI titans are spending more than $700 billion in 2025 and 2026 combined on building out data centers. OpenAI’s projects alone amount to nearly a trillion dollars in infrastructure investment.

Data centers, naturally, levy massive power demands on the electrical grid, and for this reason, many of the leading U.S. tech companies have invested heavily in increasing capacity. Early this year, OpenAI announced a billion-dollar alliance with SB Energy; Microsoft is working to bring the Three Mile Island nuclear facility online for the first time in decades; Meta has been exploring space-based solar energy; and Nvidia has partnered with Siemens and Bill Gates to develop a fusion power plant.

But progressive activists and local residents near the sites of these data centers have aligned to oppose the projects, notwithstanding the tax dollars and jobs they will generate for their respective communities. Numerous recent legacy media articles have lamented the supposed catastrophes these facilities will spark, and Sanders has emerged as their leading antagonist in Congress.

“We need a federal moratorium on AI data centers,” Sanders announced in a March press release, along with Rep. Alexandria Ocasio-Cortez (D-NY). “We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy, our democracy and the future of humanity. We need serious public debate and democratic oversight over this enormously consequential issue.”

Sanders has also emerged more broadly as a vigorous opponent of the computing revolution. In a Wall Street Journal op-ed entitled “AI Is a Threat to Everything the American People Hold Dear,” the senior senator from Vermont argued that AI is “undermining our democracy,” “damaging the environment,” “pos[ing] an existential risk to the human race,” “threaten[ing] our privacy,” and “reshaping how we as human beings relate to one another.” His proposed moratorium is the progressive equivalent of “a total and complete shutdown… until our country’s representatives can figure out what the hell is going on.”

But a Chinese wrinkle has recently emerged in Sanders’s anti-AI advocacy. In late April, he convened a discussion on Capitol Hill titled “The Existential Threat of AI and the Need for International Cooperation,” which prominently featured Xue Lan, a professor at Tsinghua University, and Zeng Yi, dean of the Beijing Institute of AI Safety and Governance.

How a U.S. senator would turn to Chinese “experts” to share their insights on AI safety boggles the mind, especially when Beijing is ruthlessly hurtling headlong toward rapid AI expansion without any of the safeguards that democratic governance provides — an irony seemingly lost a member of the greatest deliberative body in history. Seeking Chinese input on how to ethically and responsibly develop technology calls to mind foxes and henhouses, arsonists and firemen, and many other similar metaphors, none of them favorable.

So ill-advised was the Vermonter’s gambit that the Washington Post editorial board proclaimed that “Bernie Sanders’s AI cooperation fantasy is dangerous.” Even more sharply, Treasury Secretary Scott Bessent wrote that, “instead of harnessing American innovation, Senator Sanders is inviting foreign nationals to tell the United States how to regulate AI … The real threat to AI safety is letting any nation other than the United States set the global standard.”

Perhaps fittingly, Sanders has consistently served as a useful idiot to left-authoritarian regimes, and this occasion appears to be no exception. A pity, then, that instead of bolstering the American AI position, the former Burlington mayor has allowed Beijing to unduly and perniciously influence our policy.

How to fight back

So what is to be done? How can American AI giants retain their edge? What can the federal government do to ensure we win? And how can China’s highly assertive strategy effectively be countered?

Fortunately, many of our best minds are hard at work on this very problem, beginning with the CCP select committee, which itself has reportedly been the target of attempted Chinese espionage.

For instance, at the April hearing, the Silverado Policy Accelerator’s Dmitri Alperovitch urged rigorous enforcement of the prohibition on selling advanced AI chips to China.

“To ensure that we do not arm the enemy,” Alperovitch argued, “we need to first hold the line on export controls, close the loopholes, and ensure penalties effectively deter violations. Providing China with cutting-edge AI chips is the modern equivalent of selling rockets to the Soviets during the space race.”

Along similar lines, the America First Policy Institute’s Yusuf Mahmood exhorted the committee to establish security standards for frontier AI labs, create an anti-distillation task force, and authorize and fund agencies like the Bureau of Emerging Threats and the Center for AI Standards and Innovation.

“The threat from the CCP will bring new challenges,” Mahmood said. “To predict them, the federal government needs small, talent-dense, empowered offices focused on understanding AI’s future.”

Around the same time, the Office of Science and Technology Policy issued a memorandum on the “Adversarial Distillation of American AI Models” cautioning that Chinese foreign entities have been “leveraging tens of thousands of proxy accounts to evade detection and using jailbreaking techniques to expose proprietary information” in order to “systematically extract capabilities from American AI models, exploiting American expertise and innovation.” In response, OSTP vowed to “work together with private industry to develop best practices to identify, mitigate, and remediate industrial-scale distillation activities and build strong defenses against such activities” and to “explore a range of measures to hold foreign actors accountable for industrial-scale distillation campaigns.”

In addition, my American Enterprise Institute colleague Klon Kitchen, who’s closely monitoring our competition with China, recently sounded an alarm and offered some helpful suggestions. “China, Russia, and North Korea aren’t waiting for Mythos,” Kitchen wrote, in reference to the Anthropic model that identified a plethora of cybersecurity vulnerabilities across a wide range of platforms. “They’re extracting what’s already available from other AI models and running their own programs in parallel.”

To vigorously respond to this threat, Kitchen contended, the U.S. must actively facilitate deterrence against the kinds of international cyberthreats that Beijing presents.

“We should say explicitly that we hold governments responsible for attacks originating within their borders or infrastructure, regardless of whether they claim the attackers are independent criminals,” he opined. “We should say that attacks on American critical infrastructure—power, water, finance, defense—are acts of aggression that will draw a response from the full range of American capability.”

Even beyond strategies for countering the CCP’s aggressive AI posture, there’s also reason for hope that AI may actually corrode the communist regime’s core from within. 

Cameron Berg, a prominent AI researcher, argued in the Wall Street Journal that large language models like ChatGPT and even DeepSeek may eventually penetrate Beijing’s Great Firewall. In an op-ed entitled “AI is Bound to Subvert Communism,” Berg wrote that “even China’s heavily censored chatbots have proved difficult to contain within the party’s ideological boundaries.” Optimistically, “American frontier models, running without those constraints and deployed inside China, would be more potent still: a personal tutor in open inquiry for every user, engaging any question, exploring any line of reasoning, without third-party mediation.”

AI MUST SERVE HUMAN FLOURISHING 

Berg summarized a study by European researchers who managed to crack the CCP-imposed censorship function from the Chinese DeepSeek model and “found that the underlying system answered freely about every topic Beijing had tried to suppress,” including seemingly verboten topics like the Tiananmen Square massacre.

Thus, not all the news is bad. But China is unquestionably gunning hard to outpace the U.S. in AI dominance across a number of disturbing domains. Only a vigilant and forceful response can ensure American leadership in this critical technology.

Michael M. Rosen is an attorney and writer in Israel, a nonresident senior fellow at the American Enterprise Institute, and author of Like Silicon From Clay: What Ancient Jewish Wisdom Can Teach Us About AI