In January 2025, Chinese startup DeepSeek released an AI model that appeared to match America’s best —built for a fraction of the cost using chips crippled by U.S. export controls. Whether DeepSeek truly matches OpenAI’s capabilities remains widely disputed, but the market’s verdict was immediate. The announcement wiped $600 billion from Nvidia’s market value in a single day. It should have been a wake-up call. Instead, the Trump administration approved the sale of even more powerful chips to China. 

Allowing Nvidia’s H200 chips, processors that outperform China’s domestic alternatives, to flow directly to Chinese firms is a stunning reversal. In exchange for a 25% cut of sales, we are handing our adversary the hardware it needs to close the AI gap. This is a garage sale of our technological advantage. 

It’s hard to keep up with the policy timeline. In April 2025, the administration banned Nvidia’s H20 chip. By July, shipments resumed. And by December, it graduated to approving the more powerful H200. The administration argues these chips are 18 months behind Nvidia’s cutting edge. But that misses the point entirely. 

China’s problem isn’t a lack of brilliant engineers or algorithms. DeepSeek proved that. It’s accessing computing power at scale. And we’re making it very easy. In late January, Beijing approved H200 imports for three of its largest tech companies: ByteDance, Alibaba, and Tencent. They plan to purchase over 400,000 chips in the first batch alone, with millions more on order. 

As Former National Security Advisor Jake Sullivan put it, approving H200 sales “is nuts,” and “we are literally handing away our advantage.” 

Republican hawks introduced the AI Overwatch Act, which passed the House Foreign Affairs Committee 42-2 in January—a rare show of bipartisan alarm. The bill would give Congress veto power over AI chip exports to adversaries—China, Russia, Iran, North Korea, Cuba, and Venezuela—treating semiconductors like arms sales. Yet the White House has fought it at every turn. 

But the administration’s incoherence extends beyond AI hardware. A separate and equally urgent threat looms in the older, less glamorous cousins of the chips training ChatGPT. Known as mature-node semiconductor chips, these are essential components in almost every part of our critical infrastructure. This includes telecommunications equipment, computing, automotive, memory and storage, and cloud equipment. 

China now controls 31.5% of the global mature-node chip market, up from 18% in 2020, and is on track to reach 38% by 2030. A Commerce Department survey found Chinese chips in the majority of American products. If Beijing decided to restrict exports or a supply shock hit, American factories, hospitals, and military systems would feel it immediately. This means China doesn’t need to beat us in AI if it can hold our supply chain hostage. 

Yet 44% of companies couldn’t even identify whether their supply chains depended on Beijing. As one Commerce official put it, American firms remain “shockingly unaware” of the risk. 

The Biden administration launched an investigation into China’s subsidization of these legacy chips. The Trump administration concluded it, then announced tariffs set at 0%, with the actual rate delayed until June 2027. 

Simple tariffs on the chips themselves would barely register in product costs. What’s needed are tariffs on finished goods containing Chinese semiconductors, making it economically painful for manufacturers to rely on Beijing’s supply chain. Instead, China gets another sixteen months to cement its dominance. 

On both fronts—advanced AI chips and legacy semiconductors—the pattern is the same: bold rhetoric, inconsequential execution. The consensus that powered America’s semiconductor strategy is fracturing under transactional dealmaking with unclear strategic objectives. 

The stakes could not be higher. The country that leads in artificial intelligence will shape military power, economic competitiveness, and global influence for generations. China understands this. Xi Jinping declared 2025 a year of AI breakthroughs. Alibaba plans to invest $53 billion in AI infrastructure. Huawei is doubling production of its Ascend chips in 2026. DeepSeek proved that when denied our technology, Chinese firms innovate anyway. And while we debate whether to sell them advanced AI chips, China is quietly locking up the market for the legacy chips that power our cars, our hospitals, and our weapons systems. They’ll compete with or without us, so why are we making it easier? 

Congress must act where the executive will not. It can take three concrete steps that will make a meaningful difference. First, it should pass the AI Overwatch Act to require congressional review of advanced chip exports. Second, it should require the Executive Branch to implement tariffs on finished products containing Chinese legacy semiconductors rather than waiting until 2027. Third, it should use its toolkit to restore policy consistency so that American companies and our allies can plan long-term investments without wondering whether today’s rules will survive tomorrow’s trade talks. 

Winning the AI race requires choosing a lane. We cannot simultaneously claim to be in technological competition with China while selling the means to compete. That sales cut isn’t worth it. 

Isabelle Kong is a junior from California, pursuing a major in Political Science and a minor in Survey Research & Data Analytics. She is a Domestic Policy Programs intern at Penn Washington

Join Our Newsletter

Subscribe to our newsletter for the latest updates from Penn Washington.