US targets Chinese PCB dominance as AI chip supply chain risk

Nearly every AI chip running in American data centers rests on a circuit board made in China — and the US government has concluded that dependency is a national security problem it can no longer ignore.

The US has identified the near-total Chinese domination of printed circuit board (PCB) manufacturing for AI accelerators as a critical supply chain vulnerability, according to government assessments as of June 3. PCBs — the laminated substrates that connect and power every semiconductor package — for Nvidia Corp., Google LLC and Apple Inc. AI systems are overwhelmingly produced in Chinese factories, creating a single-point-of-failure risk that Washington is now moving to address.

"The PCB layer of the AI supply chain has been invisible to most policymakers, but it represents the same concentration risk we saw with advanced packaging in Taiwan," said a US official familiar with the matter, speaking on condition of anonymity because the assessment is not yet public. "If that supply were disrupted, every AI chip in the country would become a paperweight."

The vulnerability stems from a decades-long migration of PCB fabrication to China, which now accounts for more than 50 percent of global production capacity, according to industry data from IPC, the electronics manufacturing trade association. For high-layer-count, high-reliability boards used in AI servers — those with 20-plus layers capable of handling the power delivery and signal integrity requirements of chips like Nvidia's H100 and B200 — China's share is estimated at 70 percent or higher. US-based PCB fabrication capacity, by contrast, has shrunk to less than 5 percent of global supply, down from more than 30 percent in 2000.

The risk is compounded by the physical specificity of PCBs. Unlike semiconductors, which can be redesigned for alternate foundries, a PCB is a custom laminate stack-up matched to a specific chip package and system architecture. Retooling a server design for a new board supplier takes 12 to 18 months, according to supply chain consultants at Seraph Consulting. A sudden cutoff would halt AI server production lines at Dell Technologies Inc., Hewlett Packard Enterprise Co. and Super Micro Computer Inc. — the three companies that assemble the majority of Nvidia's HGX and DGX systems.

The PCB Blind Spot in the AI Arms Race

The US semiconductor industry has spent the past four years diversifying chip fabrication away from Taiwan, with the Chips Act directing $52.7 billion in subsidies to build fabs from Arizona to Ohio. Advanced packaging — the process of assembling chiplets into a single device — has also drawn attention, with TSMC building a dedicated facility in Kumamoto, Japan, and another planned for Arizona. But PCBs, the least glamorous component in the AI stack, have received almost no policy attention.

A single Nvidia H100-based server contains roughly 10 to 12 PCBs — the main motherboard, the GPU baseboard, networking interface cards, power distribution boards and storage backplanes. Each board requires 20 to 30 fabrication steps, including lamination, drilling, plating and solder mask application, before being populated with components at an electronics manufacturing services (EMS) facility. Companies like Hon Hai Precision Industry Co. (Foxconn) and Wistron Corp. perform final assembly in China, Mexico and Vietnam, but the bare boards themselves come predominantly from Chinese suppliers such as Shenzhen Fastprint Circuit Tech Co., Unimicron Technology Corp. (Taiwan-headquartered but with major China operations) and Zhen Ding Technology Holding Ltd.

The European Union faces a similar predicament. The European Commission's tech sovereignty package, announced in recent weeks, targets cloud infrastructure and chips but does not address PCB manufacturing. The EU imports most of its tech products from abroad, and its domestic PCB capacity is negligible — a gap that Brussels has not yet quantified, according to the commission's own digital strategy documents.

Who Wins, Who Loses

The supply chain risk creates a clear opportunity for US and allied PCB manufacturers. TTM Technologies Inc., the largest US-based PCB maker with facilities in California and New York, trades at 14 times forward earnings and has seen its stock rise 22 percent year to date on reshoring speculation. Isola Group, a privately held laminate materials supplier in Arizona, has increased production of high-frequency substrates used in AI server boards. In Japan, Ibiden Co. and Shinko Electric Industries Co. have invested in advanced PCB capacity for AI applications, though both remain small relative to Chinese output.

The reshoring economics, however, are daunting. A US-based PCB factory capable of producing high-layer-count AI server boards costs $400 million to $600 million to build and equip, according to estimates from IPC. Operating costs in the US are 30 percent to 40 percent higher than in China, driven by labor, environmental compliance and energy. Without government subsidies or defense procurement guarantees, the business case for large-scale domestic PCB production remains unproven.

For Nvidia, Google and Apple, the immediate risk is less about an outright cutoff — China has no incentive to halt PCB exports that generate billions in annual revenue — than about leverage. The Trump-Xi summit in Beijing on May 14-15 produced purchase commitments and new bilateral boards, but the structural condition of mutual strategic vulnerability that brought both sides to the table remains intact. As one semiconductor supply chain analyst put it, "PCBs are the kind of dependency you don't think about until someone weaponizes it."

Investor Implications

Nvidia shares, trading at 32 times forward earnings, have priced in continued AI infrastructure buildout but not a supply chain disruption to PCB supply. A 12-month production halt scenario could reduce Nvidia's data center revenue by 15 percent to 20 percent, according to supply chain modeling by Bernstein Research, translating to a $12 billion to $16 billion revenue hit. Google's in-house TPU and Apple's AI server deployments face similar exposure, though both companies have larger balance sheets to absorb expedited reshoring costs.

The most immediate beneficiaries are US defense contractors, which already source PCBs through the International Traffic in Arms Regulations (ITAR)-compliant supply chain. Mercury Systems Inc. and Cobham Advanced Electronic Solutions operate PCB fabrication lines certified for military use, giving them a template for scaling to AI-grade commercial production. Whether the commercial AI industry is willing to pay defense-grade prices for its circuit boards — a 3x to 5x premium — remains the open question that will determine how fast the reshoring timeline moves.

This article is for informational purposes only and does not constitute investment advice.