Summary
- Cambridge researchers mapped 14 layers of AI production, comprising 137 elements and 351 sub-elements across materials, equipment, computing and software.
- More than 33% of the elements were absent from 117 policy and regulatory documents covering 43 countries, the EU and OECD.
- The UK holds positions in 41 elements but mentions just three in its reviewed AI policy documents, highlighting overlooked industrial capabilities.
Governments are overlooking substantial parts of the industrial infrastructure supporting artificial intelligence, according to University of Cambridge research that identifies a mismatch between national AI strategies and the technologies, materials and manufacturing processes required to produce advanced computing systems.
The research, published on 18 November, maps the global AI supply chain across 14 layers, containing 137 elements and 351 more detailed sub-elements. Examining 117 AI-related policy and regulatory documents from 43 countries, the European Union and OECD, the researchers found that more than one-third of the identified elements received no explicit attention.
The analysis was conducted through Cambridge’s AI and Geopolitics project at the Bennett School of Public Policy. It identifies particular gaps in what researchers call the physical middle of AI production, including semiconductor design tools, specialist materials, advanced packaging, manufacturing equipment and networking technologies.
These capabilities are necessary to produce and deploy advanced computing hardware, but government AI strategies frequently emphasise finished chips, data centres, models and investment. The findings suggest that countries may overlook industrial strengths and dependencies that influence their ability to develop and operate AI infrastructure.
Chips depend on an extensive industrial network
Advanced AI processors attract considerable policy and commercial attention, partly because access to them can constrain model development and computing capacity. However, manufacturing a processor requires far more than a completed chip design.
Specialist software is needed to develop electronic circuits, while fabrication relies on manufacturing equipment, purified materials and tightly controlled processes. Finished semiconductor components then require packaging and testing before they can be integrated into servers and other computing systems.
Each stage depends on suppliers possessing capabilities that may be concentrated in a limited number of locations. An organisation controlling a final product can therefore remain dependent on companies supplying equipment or materials that are difficult to replace.
The Cambridge framework attempts to capture those relationships by identifying individual elements rather than relying on broad categories such as computing and semiconductors. Researchers mapped positions held by nearly 800 companies across 56 countries and territories.
The result demonstrates a supply chain extending far beyond the United States and China, even though those countries dominate much of the discussion about advanced AI models and computing infrastructure.
It also complicates the concept of complete technological sovereignty. A country may build its own data centres or support domestic model development while still relying on imported manufacturing equipment, specialist chemicals or intellectual property held elsewhere.
The UK has capabilities its policy barely acknowledges
Cambridge identifies UK positions in 41 of the 137 supply chain elements, while the country’s AI policy documents included in the study explicitly recognise only three.
That difference does not establish that Britain has no policies affecting the other 38 areas. The researchers examined a defined group of AI-related documents, rather than every industrial strategy, research programme or procurement policy. Their finding concerns the visibility of these capabilities within the selected AI policy material.
Among the strengths identified are semiconductor design intellectual property and specialist industrial materials. Cambridge-based Arm plays an international role in processor technology, while the UK has relevant research and expertise in areas including AI security and healthcare datasets.
Industrial gases provide another example. Semiconductor manufacturing depends on materials meeting demanding purity requirements, with specialist suppliers serving fabrication facilities across international markets.
Such businesses may receive less attention than companies developing AI models, but disruption to their supply can affect the ability to manufacture essential computing components.
A more detailed understanding of national capabilities could influence decisions about research funding, international cooperation and industrial investment. Policymakers may find opportunities to strengthen existing specialist positions rather than attempt to reproduce every stage of AI production domestically.
Japan illustrates the importance of overlooked suppliers
Japan emerges from the analysis as one of the most significant industrial participants in the global AI supply chain. The researchers identify 101 Japanese companies holding positions across 69 elements, giving the country the second-largest breadth of involvement after the United States.
Yet Japan’s reviewed AI policy documents explicitly recognise just three of those elements, revealing a substantial difference between industrial capability and the attention it receives within the selected policy material.
One example is Ajinomoto, whose build-up film is used as an insulating material in advanced semiconductor packaging. The researchers report that the company controls more than 95% of the global market for the material.
Such a position can create significant influence because the component is used in the manufacturing chain supporting advanced processors. However, market concentration also presents dependencies for customers if supply is interrupted or demand exceeds available capacity.
The researchers identify comparable relationships in semiconductor manufacturing equipment. Dutch company ASML holds an important position in extreme ultraviolet lithography, but its systems themselves depend on components and technologies supplied by businesses in Germany and the United States.
Even a company controlling a critical stage of production can therefore remain dependent on other specialist suppliers. These relationships make simplistic accounts of national control difficult to sustain.
European industrial capabilities are widely distributed
Beyond the Netherlands, Germany contributes specialist optical and industrial technologies, while Belgium has capabilities involving semiconductor materials and advanced research, including the presence of imec.
Ireland illustrates a different form of influence. The country does not have the same position in semiconductor manufacturing, but its hosting of international technology companies and data centre infrastructure gives electricity policy and planning decisions relevance to European AI capacity.
These examples show that influence over AI infrastructure can arise from manufacturing, software, research or the conditions required to operate computing facilities.
The Cambridge report proposes a more detailed audit of national positions and dependencies so governments can distinguish between resources they control, capabilities they influence and technologies for which they rely on external suppliers.
Such an assessment could help identify opportunities for diversification and areas where cooperation with other countries offers greater practical resilience than attempting to establish completely independent supply chains.
Policy omissions are not proof of absent industrial strategy
The study’s methodology provides evidence about the contents of the documents examined, but its findings require careful interpretation. The absence of an element from a country’s AI strategy does not necessarily mean it receives no attention from broader industrial, trade or research policy.
Nevertheless, the mismatch indicates that national AI discussions can fail to reflect the industrial networks on which their objectives depend. Commitments to increase computing capacity may encounter constraints associated with equipment or materials that receive little attention within the strategy itself.
Researchers recommend using the framework to identify potential bottlenecks, determine where national capabilities provide leverage and assess whether alternative sources of supply are realistic.
For businesses financing data centres, advanced processors and related infrastructure, these dependencies can influence delivery schedules and costs even when demand for computing services remains strong.
The research provides a broader map of AI production than the categories typically used in public debate. Its policy significance lies in showing how materials, manufacturing processes and specialist suppliers can shape the development of AI infrastructure without necessarily appearing in national AI strategies.












