The artificial intelligence race is built on silicon, but the real vulnerability might be what lies underneath. Printed circuit boards manufactured in China are sitting beneath Nvidia AI chips and inside systems from Google and Apple, creating a supply chain chokepoint that Pentagon officials and lawmakers now say poses a direct national security threat. A detailed report published June 3 brought the issue into sharp focus, revealing that nearly all AI-specific PCBs are produced in China even as the US pushes billions into domestic semiconductor fabrication.
This is not a theoretical risk. US Assistant Secretary of Defense Mike Cadenazzi has warned that compromised circuit boards could sabotage munitions or disrupt sensitive military systems. Al Shaffer, a former senior defense official, put it bluntly: PCBs represent the easiest place to disrupt an entire electronics chain. The concern is not simply about manufacturing capacity. Adversaries with access to the production line can insert malicious components capable of siphoning data or triggering system failures deep inside hardware that American firms assumed was secure.
The scale of dependency is staggering. China now manufactures roughly 60% of the world’s printed circuit boards, a seismic shift from decades past when the US held about 30% of global production. Today, the American share has collapsed to roughly 4%, and the Printed Circuit Board Association of America states that nearly all AI PCBs originate in China. While Washington has spent years obsessing over TSMC and advanced node wafers, the substrate carrying those chips became an afterthought.
That neglect is now colliding with geopolitical reality. At a moment when Taipei is showcasing the next wave of AI hardware, the boards powering those innovations remain anchored overseas. The rush to deploy AI, underscored by Anthropic’s recent IPO filing, only magnifies the urgency. Every new data center and training cluster depends on infrastructure the US does not control.
Congress is moving, though perhaps too late. In May, Senators Ruben Gallego and Jim Justice introduced the bipartisan Protecting Circuit Boards and Substrates Act. The Senate bill offers a 25% tax credit for domestically produced PCBs, while a companion House measure proposes $3 billion in grants aimed at reshoring manufacturing to firms like TTM Technologies. TTM, already the largest US PCB maker, and rival Sanmina are expanding domestic capacity with new facilities in New York and Wisconsin scheduled to come online in the near term. PCBAA details on the legislation make clear the intent is not merely industrial policy but a security imperative.
The Defense Department has already set a hard deadline. Starting in 2027, defense electronics must source PCBs from US factories under existing procurement rules. That timeline gives the Pentagon two years to decouple from foreign suppliers for its most sensitive applications, but commercial AI infrastructure enjoys no such mandate. Apple, which is reportedly upgrading iPhone security features, still relies on the same global supply chains for the hardware running its AI services.
Social media reaction to the June 3 coverage reflected a mixture of alarm and frustration that the gap had taken this long to surface. Industry observers noted that tax credits and grant programs, however welcome, cannot rebuild an ecosystem overnight. It takes years to develop the chemical supply chains, specialized labor, and capital equipment needed to produce high-layer-count boards for AI accelerators at scale.
Washington’s semiconductor strategy has prioritized the most advanced transistors while ignoring the humble PCB. That oversight is now exposing the entire stack. A chip is only as trustworthy as the board it sits on, and right now that trust extends to factories thousands of miles away in a strategic competitor’s territory. The $3 billion House proposal and TTM’s stateside expansion are necessary first steps, yet they amount to catching up on a twenty-year head start.
If the US intends to lead in AI, it cannot afford to treat the foundation as someone else’s problem. The boards carrying America’s most critical computations need to be made in places where Washington has visibility, control, and recourse. Until then, the country’s AI infrastructure rests on a substrate it does not own, cannot easily inspect, and may not be able to secure.
