Chapter 1 — Industrial Strategy Is a Chain of Decisions
Opening: The Wealth Machine
In a hypothetical city, a water utility needs to replace a pump. The purchase appears ordinary: an engineer specifies the equipment, a supplier quotes a price and a crew installs it. Yet the decision draws on a wider system. The utility needs money in its budget and a reliable source of electricity. The supplier needs workers, materials and a customer willing to pay. A testing laboratory may verify the equipment; a bank may finance its production; a road and a carrier bring it to the site. Years later, someone must still be able to obtain a seal and repair the motor.
Imagine that the pump fails during a dry season. The utility has a spare housing but not the control component. Its supplier has one in another country, and a payment or transport delay makes delivery uncertain. The immediate task is to restore service. The larger question is how the utility came to depend on that particular chain, what alternatives were available and who was responsible for preparing them. The answer will not be found in a single budget line or a single company's decision.
A national economy is often described through totals: output, exports, investment and employment. These measures matter. They can show whether measured production is expanding and how the value of exports compares with the value of imports. Financing, income from abroad and other transactions also matter to the external position. They cannot, on their own, show how a household reaches a job, whether a factory can maintain its machines or whether a public agency can correct an error in a contract. Those details determine how the totals are produced and how gains and losses are distributed.
Economic strategy concerns the relationships among those decisions. An industrial program may help a producer buy equipment, but the producer still needs customers and qualified suppliers. A research grant may generate a promising result, but a utility needs a safe way to test and maintain it. A new transit line may connect workers to jobs, while rising rents make the area around the station harder to afford. A trade agreement may open a market, while a small firm still lacks the finance or documentation to complete its first order.
The people making these choices do not share one objective. A minister may seek secure supply; a firm may seek a return on investment; a worker may need a predictable income; a household may care about the price and reliability of a service. Their interests can align, but alignment has to be made concrete in contracts, budgets, rules and opportunities to challenge a decision. A plan that assumes cooperation without assigning responsibility can fail even when its stated goal is widely supported.
This book follows that work from industrial and technology policy through resources, production and services, then into cities, clusters, trade and geopolitical relationships. Its examples of firms, utilities and regions are hypothetical unless a source is identified. They are constructed to make a sequence of decisions visible, not to claim that a particular country or company followed it. The sources identify what was actually read and the limits of the evidence.
No country can remove uncertainty from economic life. Demand changes, technologies fail, partners disagree and public resources are limited. A sound strategy does not pretend otherwise. It chooses a purpose, tests the assumptions behind an intervention and gives people the authority and information to correct it. The question is whether the system can continue to produce useful work, sustain essential services and revise its commitments when the conditions that justified them change.
Chapter 1 — Industrial Strategy Is a Chain of Decisions
A government can announce that a country will become a center of advanced manufacturing. The announcement does not train a technician, connect a factory to reliable power, or find a buyer for what the factory makes. Those things require different people to make decisions on different schedules. A strategy becomes real only when the promises in the announcement are connected to the organizations that can perform them—and when someone can see which connection has failed.
Industrial strategy is often presented as a choice between trusting markets and directing them. That division hides much of the work. A government that does not subsidize a plant may still set safety standards, operate a port, fund a technical college, approve land use and negotiate trade rules. Each action affects what enterprises can attempt and what they must pay. A government that does subsidize a plant still depends on private suppliers, lenders, workers, customers and competing producers. Neither the state nor the market acts as a single decision maker.
The practical question is narrower than whether industrial policy is good or bad. What problem is a particular measure intended to solve? Why would the people affected fail to solve it without that measure? Who can implement the remedy, and what evidence would show that it has become unnecessary or harmful? The answers may justify targeted support, a general improvement in public services, a change in regulation or no new intervention. An ambitious name on a program does not answer any of them.
From a national goal to a usable instrument
Consider a hypothetical country that wants a more reliable supply of medical equipment. An assembly plant imports a critical component, while hospitals buy completed devices from several countries. A subsidy for the domestic assembler could increase its capacity, but it would not automatically make the imported component available during a disruption. The government must decide whether the objective is more local assembly, access to spare parts, diversified suppliers, emergency production rights or lower prices for hospitals. Those goals overlap but are not identical.
Suppose an official proposes a grant to expand the plant. The company says it would invest only if the grant covers part of the equipment cost. That statement matters, but it is not proof that the project would otherwise fail to happen. The company might have alternatives, and the government might be able to buy a different safeguard. A procurement contract, shared testing laboratory, stockpile or agreement with a second supplier would address different weaknesses. Comparing them requires a description of the failure each one is meant to prevent and the costs it creates when no disruption occurs.
The government also needs to identify what the grant buys. If it pays for a new machine, the agreement might require the machine to remain available for a defined period, while leaving ordinary production decisions to the firm. A condition that equipment be maintained, workers trained and spare parts obtainable could matter more than a ceremonial opening date. Yet every condition requires monitoring. A small agency may be unable to verify a complex promise, and a rigid requirement may prevent a company from changing a process when technology improves. Designing the obligation is as important as choosing the amount.
The plant may nevertheless be a poor choice even if domestic production would be valuable. It may lack a qualified supplier, face a severe water constraint or have no viable customer after the public contract ends. Funding it could absorb money that would have made several existing suppliers more reliable. A strategy needs a credible way to stop supporting a project, including when political leaders have already associated their reputation with it.
The difference between an award and an outcome
The United States Government Accountability Office examined semiconductor incentives awarded from September 2024 through July 2025. Its summary reports that the Department of Commerce had awarded direct funding to 19 companies for 40 projects; some companies also received workforce development funding. The report describes the projects' anticipated timelines and Commerce's selection and funding process. Those facts are about awards, planned work and reported milestones. They are not evidence that every factory ultimately operated as intended or that the program caused a particular national productivity gain. [1-1]
This distinction matters because a public balance sheet can record money committed before a production system exists. The award may require a company to reach milestones, but a milestone is meaningful only if it is connected to a capability the policy needs. A completed building is not the same as a qualified production line; a qualified line is not the same as a reliable supply of materials; and an operating line does not establish that customers can obtain the products at an affordable price. These are separate claims requiring separate evidence.
The GAO summary also describes upside-sharing arrangements negotiated for a portion of the projects. Such an arrangement can change the distribution of unexpectedly high returns between the public and the recipient. It cannot, by itself, settle whether the original grant was necessary or whether an alternative use of funds would have produced greater benefits. The conditions, thresholds and enforceability of each agreement matter; a summary of their existence does not reveal every contract term. [1-1]
A program must fit the institution that operates it
Imagine that two countries identify the same shortage of electrical equipment. One has a procurement office that has purchased complex equipment for years and can compare suppliers against a published specification. The other has only a small purchasing team, while its electricity utility keeps incomplete records and pays suppliers late. The same subsidy can behave differently in the two settings. In the first, an application may be evaluated against operating information. In the second, the state might pay for capacity without being able to determine whether the equipment fits the grid or whether the utility can maintain it.
Administrative capacity is not a fixed national personality. It consists of specific abilities: writing enforceable terms, checking technical claims, retaining records, paying invoices, resolving appeals and learning from results. An agency can perform one of these tasks well and another poorly. A policy assessment should name the task that matters. Saying that a government has weak capacity is too broad to show whether a particular procurement or grant can be improved.
A simpler instrument may reduce the burden on officials, but it also creates tradeoffs. A uniform grant per unit produced is easier to calculate than a negotiated agreement covering a dozen outcomes. Yet the simple payment may reward output that would have been produced anyway or encourage firms to make a product that has little use after the grant expires. A negotiated contract can address specific risks but makes public comparison harder and gives discretion to the officials who negotiate it. Neither design removes the need for oversight; it changes where oversight is needed.
An implementation plan should include the people who can say no. A safety inspector may refuse approval until a plant changes its process. A municipality may deny a water connection. A local community may challenge a land acquisition. Those interventions can delay the project, but a delay is not automatically a defect in the strategy. The affected parties may possess information that the central government and the company do not. The question is whether the process can resolve the disagreement with evidence and a decision, rather than leaving a costly uncertainty indefinitely unresolved.
The public claim on private returns
A subsidy changes the division of risk. If the public pays part of the initial cost and the project succeeds, shareholders and workers may receive benefits while taxpayers receive indirect benefits through taxes, employment and the services the project makes possible. If the project fails, the public may not recover the grant. That allocation can be justified when wider benefits exist, but it should be visible in the decision.
Suppose a hypothetical project costs $100 million. A government grant covers $30 million and the firm supplies the remaining $70 million. If the project generates a return, the grant's cost cannot be evaluated by comparing $30 million with the entire revenue of the plant. Revenue also pays suppliers, wages, energy, finance and maintenance; the project might have generated some of that activity without the grant. The relevant comparison is between plausible alternatives, including what the public could have financed instead and what the firm would have done under different support terms.
An agreement can request a repayment if the recipient fails to meet conditions. It can also provide for a public share of unusually high returns. Each mechanism has limits. A failed company may not possess enough assets to repay; a threshold based on reported profit can be influenced by accounting and by costs shared with related firms. A public share in returns requires someone to understand the records and enforce the agreement years later. The possibility of repayment should not be treated as money already recovered.
A government may reasonably decide that benefits it cannot capture in cash justify a grant. Training a workforce or proving a production process may help other firms. But spillovers cannot simply be assumed from the label strategic. The government can ask how trained workers will be able to move, whether smaller firms can use shared facilities, and what information can be published without revealing legitimate trade secrets. If all the learning is locked inside one facility and the facility closes, the expected wider benefit may be much smaller than the announcement implied.
When a target changes behaviour
A target can make a distant objective actionable. It can also turn attention toward the number that is easiest to report. A workforce program measured by certificates issued may have an incentive to shorten instruction or count people who never use the skill. A manufacturing program measured by announced investment may reward projects before financing has closed. A trade program measured only by exports may overlook imported inputs and profits transferred abroad. These examples do not make measurement pointless. They show why a useful measure needs an account of what it leaves out.
Suppose a hypothetical region announces a target of 2,000 new jobs. A recipient reports that a facility has hired 1,500 people. The figure is useful, but several questions remain: how many positions are full-time, how long did they last, where did the workers previously work, and would another employer have hired them without the grant? A facility can create real opportunities while also drawing scarce workers from local employers. Counting the new jobs and ignoring displaced activity would overstate the region's net gain; counting only displacement would ignore new income and learning. A careful evaluation states which question its evidence can answer.
The choice of comparison is consequential. Comparing a supported plant with a plant in a different region may confuse the grant's effect with differences in transport, labour supply or existing firms. Comparing the same region before and after the award may confuse the policy with an economic recovery or a change in demand. A credible counterfactual is difficult, especially when the policy changes the behaviour of firms outside the supported group. Difficulty is a reason to state uncertainty and combine methods, not to present a target as a causal finding.
The OECD's industrial strategy guidance recommends planning evaluation and identifying indicators and counterfactuals before results are assessed. It also describes the role of oversight beyond the team responsible for implementation and the possibility of revising or discontinuing measures in response to findings. This is guidance about how a government might organize learning; the guidance itself does not prove that any particular policy worked. [1-2]
Who can contest the plan?
A plan that coordinates public money and private investment distributes authority. An agency selects applicants; a company decides whether to apply and what information to provide; a local authority controls permits; workers and residents experience consequences that may not be visible in the application. The plan needs a way for each relevant objection to reach a person who can act on it. A consultation meeting is one possible route, but its value depends on what decisions are still open and whether participants can obtain enough information to challenge the assumptions.
Imagine a hypothetical coastal region where a government wants to support a battery-materials plant. The developer expects a large tax contribution and employment. Fishing households are concerned that water discharged from the site could affect a breeding area. An environmental permit may require monitoring, but the community also needs to know who takes samples, where results are published and what happens if a threshold is exceeded. If the only remedy is a lawsuit after damage has occurred, an apparently clear rule may be difficult to use in time.
The project could be modified: a different discharge system, a different site, or a phased start tied to independent measurements. Each option has a cost and may change whether the investment proceeds. A credible comparison includes the cost of delay as well as the cost of permitting harm. It cannot be resolved by calling one side pro-growth and the other anti-industry. The public decision concerns the terms on which one productive activity can coexist with another.
Participation can also be uneven. A large company can afford engineers and lawyers to interpret a technical report; a small supplier or neighbourhood group may not have the same resources. Providing access to an independent technical explanation can improve a hearing without guaranteeing agreement. Conversely, an unbounded process in which every participant can reopen every decision may make an investment impossible to schedule. A workable procedure identifies the decisions, the evidence available, the period for objections and the authority responsible for responding.
An incentive can outlive its purpose
The case for temporary support often depends on an expectation that a capability will become viable after firms learn, suppliers enter or demand grows. That expectation needs a date or condition at which it can be reconsidered. A grant may be spent once, but tax exemptions, price guarantees and import protection can continue for years. The longer they remain, the more people make plans around them. Employees depend on jobs, suppliers extend credit and local budgets anticipate tax revenue. Withdrawal then becomes a practical problem rather than a simple exercise in ending a line of expenditure.
Suppose a hypothetical government protects a domestic machinery producer while it learns to manufacture a more efficient pump. After five years the firm has lowered defects but still cannot deliver enough units on time. Users continue to pay more than they would for imports. The government faces at least three choices: end protection as scheduled, extend it with revised conditions, or shift support toward testing and supplier qualifications available to several firms. Each choice exposes a different group to risk. An extension preserves time for learning but asks customers to bear another period of higher prices. Immediate withdrawal protects customers but may end a capability before it can mature.
The evidence should distinguish causes. If delays come from a public port that fails to clear imported components, withdrawing support from the producer will not repair the port. If the firm repeatedly misses delivery dates despite reliable inputs, blaming the port would shield a performance problem. A review should establish what the parties promised, what actually occurred, which events were outside their control and whether a remedy is available at reasonable cost. It should also identify who has authority to change the arrangement.
A sunset clause can force a decision onto the calendar, but it is not a guarantee that support ends. Legislatures can extend a measure, agencies can redesign it, and firms may lobby to define continuation as essential. An automatic expiry that ignores an external shock can also destroy useful capacity. The purpose of a review date is to make continuation a decision that must be explained, with evidence available to people outside the original negotiation. The explanation should say what has changed since approval and why the remaining benefit justifies the remaining cost.
Coordination is a sequence, not a meeting
Industrial projects involve dependencies that are often owned by different institutions. A technical college can create a training course, but students need transport and a reason to believe jobs will exist. A factory can offer jobs, but it may delay hiring until equipment arrives. A utility can plan a connection, but its investment may require a permit and a commitment from the factory. If each party waits for certainty from the others, no one acts first.
A coordination agreement can assign a sequence. The utility might begin a limited design study while the company completes financing; the college might design a short curriculum before purchasing specialized equipment; the company might provide a schedule and pay a deposit for the connection. These commitments are smaller than full construction but create information about whether the larger project is feasible. They also identify who bears the cost if the plan stops. A memorandum that says the parties will cooperate is less useful if nobody owns the next decision or the money to carry it out.
Coordination can become a channel for favoritism. If a single large employer writes the training requirements and controls access to a publicly funded facility, other firms may be excluded even though they contributed through taxes. If an agency tries to satisfy every possible employer, the course may teach skills too general for any actual job. The design needs a reasoned boundary: which equipment is shared, who can apply, how fees are set and how disagreements are reviewed. Open access is a matter of operational rules, not a sentence in a mission statement.
The sequence must continue after the initial announcement. Equipment ages, instructors leave and markets change. A program that funds construction but assigns no operating budget can create a building that looks complete while its equipment sits unused. A durable strategy therefore distinguishes the money needed to start an institution from the recurring resources and authority required to keep it useful.
The decision to change course
A strategy can fail in a way that still produces useful information. An equipment project may reveal that local suppliers cannot meet a required tolerance, or that a training course assumes a level of mathematics many applicants have not had an opportunity to acquire. The government can use that information to redesign a test or support a different course. But learning is not automatic. The parties must record what was attempted, allow someone to inspect the result and give the next decision maker the authority to act on it.
A project review should separate three questions. Was the task carried out as agreed? Did carrying it out contribute to the intended capability? Is that capability still worth the resources required to maintain it? A company might meet every contractual milestone while global demand moves elsewhere. Conversely, a delayed project may remain valuable because the bottleneck was a public connection that has since been repaired. Treating every delay as evidence of incompetence would hide the second possibility; treating every milestone as success would hide the first.
Imagine a hypothetical government that funded a common testing facility for small manufacturers. The building is complete, but few firms use it. An initial report calls the project a failure because the number of customers is below target. A closer inquiry finds that testing fees are due before firms receive payment from their buyers, and appointments take three months to obtain. Lowering the fee or changing the schedule might help, but neither action is free. A longer opening week requires staff and maintenance; a credit arrangement exposes the facility to unpaid bills. The agency should compare those costs with the evidence of demand rather than presume that an empty building proves an absence of need.
There is another possibility: firms do not use the facility because the tests do not correspond to the standards their customers require. In that case, lowering the fee may only subsidize an irrelevant service. The government might need a different accreditation, a partnership with a customer, or a decision to close the facility. The people who approved the original specification may be reluctant to acknowledge the mismatch. An independent review is useful because it changes who can ask the question; independence alone does not give the reviewer complete information or guarantee that their advice will be followed.
Stopping also requires a plan for obligations. Equipment bought with public money may have a resale value, a training use or a maintenance cost. Workers need notice and may possess skills useful elsewhere. A site may have environmental responsibilities even after production stops. The choice is not simply between continuing a subsidy and deleting its cost. Ending a project can be the right decision while still requiring money and attention to settle what the project leaves behind.
A strategy that can answer for itself
A strategy must remain answerable after the grant is awarded. The agency needs to show what the recipient delivered, the people bearing costs need a way to challenge the result, and a reviewer must be able to ask whether the public contribution made a difference.
This sequence does not guarantee that every intervention succeeds. Uncertain demand, political change, rival policies and unforeseen technical difficulties can defeat a well-reasoned plan. Nor does a failed intervention prove that no intervention was justified. The comparison is with the alternatives available when a decision was made, revised as new information arrives. A country can spend public money on a factory and still lack the engineers, supplier network or purchasing institutions that make production useful. It can also improve those shared conditions without owning a factory itself.
The next chapter follows one of those conditions: the movement of knowledge from an idea to a process that people can use, maintain and improve. Technology strategy depends on invention, but it also depends on the institutions that allow learning to travel beyond the first successful demonstration.
Notes and Sources
[1-1] U.S. Government Accountability Office, “Semiconductors: Information on Projects Funded to Strengthen U.S. Supply Chain,” GAO-26-107882, https://www.gao.gov/products/gao-26-107882, accessed 2026-10-31. Read the Fast Facts, highlights, full-report scope and methods, and the section on upside-sharing arrangements. The report covers awards from September 2024 through July 2025 and excludes subsequent amendments, including an August 2025 amendment affecting six projects. GAO analyzed Commerce documents and interviewed officials. The remaining report and underlying agreements were not reviewed; later changes and causal outcomes are not established here.
[1-2] OECD, *Industrial Policy Handbook: From Strategy Design to Implementation* (2026), Part I, “Designing Industrial Strategies,” section “Phase 4: Evaluate and adapt,” https://www.oecd.org/en/publications/industrial-policy-handbook_ff099713-en/full-report/component-5.html, accessed 2026-10-31. Read the Phase 4 checklist and action paragraphs on evaluation, indicators, counterfactuals, oversight and adaptation. The source is policy guidance; the remainder of the handbook and its country examples were not reviewed for this passage.