Chinese Circuit Boards in U.S. AI Accelerators Trigger National Security Review

U.S. national security agencies launched an interagency review after discovering that approximately 60% of printed circuit boards underpinning domestic AI accelerators are manufactured in China, creating hardware attack vectors that existing semiconductor controls failed to address.

The revelation exposes a critical blind spot in U.S. technology policy: while export restrictions have focused on advanced chips since October 2022, the multilayer circuit boards that connect those chips to power and memory systems have escaped regulatory scrutiny. Circuit boards present opportunities for adversaries to insert malicious components during manufacturing — a vulnerability the Defense Department now considers urgent enough to mandate domestic sourcing for military purchases, according to CNBC.

Supply Chain Concentration Risk

Nearly all AI circuit boards for Nvidia, Google, and Apple accelerators are fabricated in Chinese facilities, per CNBC reporting. The concentration reflects broader manufacturing patterns: 90% of printed circuit boards globally originate in Asia, with half from mainland China, according to congressional testimony from Representative Raja Krishnamoorthi.

This dependency creates asymmetric leverage. Circuit board fabrication requires specialised equipment, trained labour, and economies of scale that take years to replicate. A single advanced multilayer board for AI accelerators contains hundreds of copper traces, precisely drilled vias, and embedded components — manufacturing complexity that Chinese facilities mastered while U.S. capacity atrophied over two decades of offshoring.

Cost Inflation Hits AI Infrastructure

The security review arrives as PCB prices surge from supply-side constraints and demand spikes. Prices rose up to 40% between March and April 2026, driven by competition between commercial AI deployments and defense contractors for limited domestic capacity, CNBC reported. TTM Technologies, the largest U.S.-based PCB manufacturer, increased prices by 5% to 25% in May as order backlogs extended.

“We are competing with the AI demand,” Cathie Gridley, executive vice president at TTM Technologies, told CNBC. “The commercial side is willing to pay a much higher price to get access to that capacity, and so what that does is that really drives prices up across the board.”

The PCB squeeze compounds broader input cost inflation. High-bandwidth memory components face 40% to 60% price increases, translating to 15% to 25% higher costs for AI-capable server configurations, per analysis from Trax Technology. Helium shortages from strikes affecting Qatar’s production, combined with copper and bromine supply concentration risks, further compress margins for hyperscale data centre operators.

Legislative and Enforcement Response

Congress responded with the Protecting Circuit Boards and Substrates Act, which would provide a 25% tax credit for domestic PCB purchases by AI companies and allocate $3 billion in financial assistance to U.S. manufacturers like TTM, modeled on the CHIPS for America program. The legislation aims to reverse offshoring trends by making domestic sourcing economically competitive with Chinese suppliers, according to congressional records.

Separately, the Chip Security Act, approved 26 March 2026, mandates embedded tracking technology in advanced semiconductors to prevent diversion to sanctioned entities. The law followed Justice Department indictments of executives accused of diverting $2.5 billion in Nvidia server shipments to China through shell companies, demonstrating enforcement escalation beyond export licensing violations.

The DOJ National Security Division adopted a corporate enforcement policy on 30 March 2026 emphasising criminal prosecution for violations of export administration regulations, International Traffic in Arms Regulations, and CFIUS review requirements, Alvarez & Marsal noted in enforcement guidance to clients.

Domestic Capacity Buildout

TTM Technologies is expanding U.S. production with a new factory in Syracuse, New York, set to begin operations in mid-2026, and a larger facility under construction in Wisconsin. Each factory costs $250 million to $400 million to build, requiring multi-year payback periods that make private investment dependent on sustained government demand or subsidies, CNBC reported.

The capital intensity explains why onshoring has lagged despite bipartisan political support. Circuit board manufacturing involves toxic chemical processes, environmental compliance costs, and labour skill requirements that disappeared from U.S. industrial regions over decades of deindustrialisation. Rebuilding that capacity requires not just factory construction but workforce training programs and regulatory frameworks that balance environmental protection with strategic necessity.

Geopolitical Implications

The PCB vulnerability intersects with broader U.S.-China technological competition. While semiconductor export controls aim to slow Chinese AI development by restricting access to advanced chips, the discovery that Chinese facilities manufacture the boards housing those chips reveals control system limitations. A sophisticated adversary could theoretically insert hardware backdoors, modify power delivery systems to enable side-channel attacks, or simply gain intelligence on U.S. AI architecture through access to board schematics and component layouts.

Former Trump administration Deputy National Security Advisor Matt Pottinger argued that export policy mistakes enabled Chinese access to U.S. AI technology. “Congress needs to step in, reverse the policy, and put durable guardrails in place so the mistake can’t be repeated,” Pottinger stated in analysis from Mayer Brown.

What to Watch

The interagency review will determine whether CFIUS authority extends to foreign investment in PCB manufacturers supplying AI hardware, potentially blocking Chinese capital from acquiring domestic facilities. Watch for Defense Production Act invocations to prioritise military PCB orders over commercial demand, which would further inflate civilian AI Infrastructure costs. TTM Technologies’ production ramp timeline in Syracuse and Wisconsin will signal whether domestic capacity can scale fast enough to meet 2027-2028 data centre buildout plans without forcing delays or design compromises. Finally, monitor whether hyperscale cloud providers announce PCB manufacturing acquisitions or joint ventures, signaling a shift toward vertical integration driven by supply security rather than cost optimisation. The next 18 months will reveal whether U.S. policy can rebuild PCB manufacturing capacity faster than geopolitical tensions escalate, or whether AI infrastructure costs and deployment timelines become casualties of the broader technological decoupling between Washington and Beijing.