How Industrial Policy Shapes AI Supply

A regional government wants smaller businesses and a technical college to use AI for practical trials. Officials propose a publicly supported computing center. Its funding is approved and servers have arrived. Can the government tell those users that the promised capability is ready? This is Lesson eight, How Industrial Policy Shapes AI Supply. We will follow this fictional proposal from the users' problem through the investment and operating choices, then decide what the government can reasonably claim

and what it should investigate next. Start with the people the program is meant to help. Suppose the college wants students to test a model on an approved training task, while small businesses want help evaluating applications. Interviews reveal that some teams struggle to obtain computing time and others lack people who can prepare and supervise a trial. Those are different constraints within the same proposed service. The government needs to specify which users it intends to serve,

what useful task they should complete and which obstacle public support would change. Follow one college trial into the proposed center. Servers need electricity, networking, storage, software and people who can maintain the service. Their computing chips came through a wider supply chain. A design specifies how a chip should work; manufacturing turns designs into physical devices using specialized equipment and materials.

Packaging connects and protects components within a working system. Memory and software support the task. The center's ability to serve a class therefore depends on several connected suppliers and capabilities. The college's completed trial gives us a reason to ask which link currently limits the service. These relationships help explain the financing problem. Some investments require substantial spending on facilities,

equipment or training before useful output brings revenue. That is capital intensity: large capital commitments relative to the activity being supported. Firms with access to finance may be better placed to make those commitments, which can affect entry and concentration. An application team renting computing services faces a different combination of upfront and continuing expenses. The government should investigate the actual financing

and operating constraint before choosing the scale of its intervention. For our college and smaller businesses, one approach is shared computing access with training and technical support. Teams could try a task without purchasing an entire facility or already employing every specialist. Eligibility rules, available hours, fees and help with preparation would determine who can actually use it. The mechanism is specific: public support changes access to equipment and expertise.

The program still needs evidence that the intended users can complete useful work through that service. A dated example of this policy design comes from EuroHPC's March 2025 AI Factories announcement. It described computing resources, training and technical expertise for startups, smaller firms and researchers. That announcement gives us an intended support model to examine. To describe its later operation or benefits,

we would need the corresponding service and outcome evidence. In our fictional region, the immediate question is whether the proposed combination addresses the barriers the college and businesses identified. A different proposal might support a producer's investment in manufacturing equipment. A subsidy reduces part of the producer's cost and can change the attractiveness of a project. Its purpose might be additional production capacity or a more resilient supply relationship.

That starts a different chain from the shared service. The government would ask what investment the producer would make with and without support, which customers the capacity would serve and whether equipment, skills and demand can sustain production. More manufacturing capacity may still leave our college without affordable computing access or suitable training. Return to the center's approved funding and delivered servers. The college is ready to book a class, but the power connection is unconfirmed,

acceptance testing is unfinished and staff training is incomplete. A delivery record establishes that equipment arrived. The next steps require installation, testing and preparation for operation. The government can report the supported milestones and the unresolved conditions. Students completing their trial reliably would supply evidence about a later stage: useful service for an intended user. Real project reporting also separates these stages. In its 15 October 2026 earnings call,

TSMC said construction of its second Arizona fabrication facility was complete and tool move-in and installation were under way. Production remained a future target. This is an operator's dated account of project stages. It helps illustrate why a completed structure and installed production capability require different evidence. Establishing a public policy's additional effects would also require a suitable comparison and evidence about the benefits and costs being claimed.

Suppose the regional center later passes its tests and opens. The government now needs to examine use. Can the college obtain suitable time and support? Can smaller firms afford to participate? Someone must fund electricity, maintenance, replacement equipment and qualified staff. A center with little relevant demand could consume resources while serving few useful trials. Records of bookings, completed tasks,

reliability and support costs would help the government judge whether the original public objective is being met. Those records also make the alternative meaningful. If equipment is available but qualified support is scarce, expanding training and access to an existing service could deserve comparison with buying another building. If available services cannot meet the required workload or reliability, additional capacity may address a different constraint. Compare the options against the same objective and include their continuing costs.

Spending on one option uses resources that could support another. The college, businesses, workers and local community have interests in which benefits are delivered and which burdens they carry. The proposal may also claim to build domestic capability. Specify what that means. Equipment can be located in the region while ownership, replacement parts, software or operational decisions depend on organizations elsewhere. A manufacturing investment can add one stage

while retaining important supply relationships across borders. Policy may seek resilience or strategic capacity, with costs from duplication or more expensive production. The decision needs to state which dependence matters, what the intervention changes and which relationships remain necessary. Try the Learning Check. Why is a subsidy announcement not capability evidence? Explain the stages and conditions connecting funding to useful operation,

and name one relevant record you would inspect. Pause and write your answer. Funding must become delivered, usable and operated assets and skills serving a defined purpose. Installation, acceptance tests, trained staff, reliable support and actual service each need appropriate evidence. For our center, you might inspect commissioning records and whether the college can complete its intended trial.

Match the record to the claimed stage and purpose. The funding approval is one part of that account. Return to the regional government's promise. It can now explain the users' task, the constraint public support would address, how the chosen instrument changes access or investment, and the steps still needed for operation. Before describing the program as successful, it should examine useful service, continuing resources, beneficiaries and a relevant alternative.

The proposal becomes a decision that people can inspect and revise as those records develop. For Practice, use the different fictional advanced-packaging training workshop in your worksheet. Its funding letter is signed and tools have arrived, while installation, operator training and service trials remain incomplete. Identify the supported milestones, missing evidence and intended capability. Then explain an operating condition, a relevant alternative,

a public trade-off and the domestic attribute being claimed. Complete your own record before using the separate key. You can now follow an industrial-policy proposal from a public objective through a supply constraint, an intervention and the capability people can use. The next lesson develops the cross-border question: how ownership, location, supply, data and decision control shape sovereign AI and dependence on other countries.