AI policy overlooks a third of mapped supply-chain elements, Cambridge study finds
An audit of 117 policy documents finds gaps around the industrial systems behind AI—and overlooked leverage beyond the U.S. and China.
Loading page…
An audit of 117 policy documents finds gaps around the industrial systems behind AI—and overlooked leverage beyond the U.S. and China.
Listen to this story
Cambridge’s AI & Geopolitics Project mapped 137 supply-chain elements and 351 sub-elements, then checked 117 policy documents across 43 countries and territories, plus the EU and OECD. Policies often emphasize models, data centers and governance while giving little attention to chip design, packaging, testing and server manufacturing; the review measures mentions, not what officials know. The team says those gaps obscure practical dependencies and leverage, and offers an audit toolkit to help governments weigh supplier diversification against domestic capacity and cooperation.
Japan has 101 identified firms across 69 supply-chain elements, but its reviewed policy documents mention only three.
Synopsys, Cadence and Siemens EDA dominate electronic design automation, software essential to chip developers but nearly absent from policy documents.
Advanced packaging operations are concentrated in Asian countries and territories including Taiwan, Vietnam and the Philippines; Ajinomoto supplies more than 95% of global build-up film.
Governments’ AI strategies are missing much of the machinery behind the technology. A new Cambridge study finds that just over a third of mapped supply-chain elements receive no mention in the policy documents reviewed. The gaps cluster around chip production and supporting hardware, leaving important dependencies—and sources of national leverage—outside the policy picture.
The invisible AI supply chain, from Cambridge’s AI & Geopolitics Project, compares a detailed production map with 117 AI-related policy documents. The review covers 43 countries and territories, plus the European Union and OECD. Its central finding concerns what those documents mention; it does not by itself establish that governments lack knowledge of every omitted component.
The researchers mapped 14 layers, containing 137 elements and 351 more detailed sub-elements that help determine who can supply them. Nearly 800 firms hold positions across 56 countries and territories. The map stretches from minerals and ultra-pure gases through chips, servers and trained models, while also accounting for shipping routes and governance standards.
Policy attention concentrates on models, data centers, funding and governance, the researchers found. Much less reaches the physical middle of production: chip design, packaging, testing and AI server manufacturing. These are not interchangeable steps. They include the tools used to design a processor and the operations needed to turn separate chips into usable AI hardware.
One example is electronic design automation, or EDA: software used to design AI chips. The study identifies Synopsys, Cadence and Siemens EDA as the three dominant suppliers. Their software is essential to chip developers including Nvidia, AMD and Google, yet the researchers found it receives almost no policy attention.
Another is advanced packaging, which connects multiple semiconductor chips into high-performance processors used for frontier AI models. The report places those operations overwhelmingly in Asian countries and territories, including Taiwan, Vietnam and the Philippines. The team warns that disruption to an upstream component could cascade into AI supply shortages, with consequences for investment and productivity.
Policymakers focus on finished chips rather than the ecosystem behind them. They focus on GPUs but ignore the packaging operations needed to turn those chips into usable AI hardware.
Dr. Siyi Liu, lead author
The country-level findings challenge a view of AI power centered on the United States and China. Even a familiar bottleneck has dependencies of its own: ASML, the Dutch supplier dominating extreme ultraviolet lithography used in chipmaking, relies on essential parts controlled by German and U.S. companies, the report notes.
The researchers argue that complete AI sovereignty—one nation autonomously operating its own supply chain—is impossible. Their proposed response is cooperation built around countries’ specific strengths, rather than an assumption that every dependency can be brought inside national borders.
The team has created a government audit toolkit to turn that argument into supply decisions. It is intended to help policymakers identify their dependencies, decide when to diversify suppliers, and judge when leaving an outside supplier and building domestic capacity makes sense.
Loading discussion...
Join the conversation
Explain where international dependence is acceptable and where domestic capacity matters most.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.