ECONOMICS / POLICY · 30 SEP 2026

Industrial policy is back, and the old arguments decide whether it works

The size of a subsidy matters less than the test it has to pass.

or three decades, the respectable position in economic policy was that governments should not pick winners. Tariffs came down, subsidies were treated as failures waiting to be audited, and the phrase "industrial policy" survived mostly as a warning. That era is over. The United States now spends tens of billions subsidizing semiconductor factories. Europe has its own chips program and a green industrial plan. India pays manufacturers per unit of output across a dozen sectors. China never stopped. The argument is no longer whether the state should shape the industrial structure of the economy, but how, and on what evidence.

The strange thing about this revival is how little of it is new. Every argument being made today, for and against, was worked out decades ago, some of them centuries ago. The revival is less a new idea than a stress test of old ones, and the outcome will be decided by which of those old arguments the new programs manage to respect.

Start with the case for. The strongest version was never "government knows best." It is that some markets systematically under-produce things with large spillovers. New industries are hard to start because the first entrant pays costs that later entrants free-ride on: trained workers, supplier networks, proof that the technology works at scale. This is the infant industry argument, and in its modern form, associated with economists like Ricardo Hausmann and Dani Rodrik, it is an argument about information. A country does not know what it can be good at producing until someone tries, and the someone who tries captures only a fraction of the knowledge their attempt creates. Left alone, markets under-experiment.

Add to that the agglomeration point: once an industry clusters somewhere, the cluster feeds itself. Chip fabrication in Taiwan, shipbuilding in Korea, precision manufacturing in southern Germany. The cluster is a real productive asset, and no single firm will pay to build it. Strategic trade theory in the 1980s, from Paul Krugman among others, showed that in industries with steep learning curves and few global players, a well-placed shove can determine where the cluster ends up, and that where it ends up matters for national income.

Now the case against, which is really two cases. The first is Hayek's: the knowledge required to pick winners is dispersed, tacit, and changing, and ministries do not have it. The semiconductor subsidy that looks obvious in hindsight looked obvious about five other industries too, most of which did not become semiconductors. The second case is political, associated with Anne Krueger's work on rent-seeking: once subsidies exist, they get allocated by lobbying power, not economic logic, and they are much easier to start than to stop. The history of import substitution in Latin America and South Asia is the cautionary tale here. Protected infants grew into protected, uncompetitive adults, and the protection outlived every rationale for it.

The interesting thing is that both sides are right, and the empirical record says so. East Asia's celebrated interventions worked when they came with discipline. Korea's subsidies in the 1960s and 70s were tied to export performance: sell abroad or lose support. Export markets provided the test the ministry could not, a hard external check on whether the bet was paying off. The failures, by contrast, are mostly stories of support without tests: subsidies with no expiry, no benchmark, and no consequence for missing one.

This suggests the real question about the current wave is not the size of the checks but the design of the discipline. Are the semiconductor grants conditioned on anything measurable, and what happens if the conditions are missed? Does the program build the public goods, the trained workforce and supplier base, that the spillover argument points to, or does it reimburse capital expenditure that firms would have made anyway? Is there a sunset, a review, a number that would count as failure? These are unglamorous questions, and they are the whole game. An industrial policy with good answers is a different instrument from one with bad answers, even if the press releases are identical.

Two newer arguments deserve attention because the old frameworks handle them awkwardly. The first is resilience. Economies optimized for efficiency concentrated fragile supply chains, and the pandemic and the chip shortage priced that fragility in a way models had not. Paying for redundancy is rational in the same way insurance is rational, but it should be scored as insurance: a premium paid to reduce variance, not a claim that subsidized production beats the market at production. Confusing the two is how insurance budgets become permanent entitlements.

The second is geopolitics. When the argument for a subsidy is that a rival power must not control a chokepoint, the usual cost-benefit arithmetic does not apply, and economists should be honest that this is a security decision borrowing economic language. That does not make it wrong. It makes it important to keep the accounts separate, because security logic has no natural spending limit.

Where does this leave the citizen reading the next subsidy announcement? With a short list of questions older than the policy itself. What spillover is this paying for, and is the payment tied to evidence the spillover exists? What test will this support face, and who imposes it? When does it end? Governments that can answer all three are running an experiment. Governments that cannot are running a patronage program with better branding.

The free-market era did not end because its economics was refuted. It ended because the world changed under it: supply chains got brittle, a strategic rival industrialized faster than any in history, and the costs of concentration stopped being theoretical. The economics that decides what happens next is the economics that was always there, waiting on the shelf. Spillovers are real. So are rent-seekers. The countries that remember both will get the semiconductor factories and the growth. The ones that remember only the first will get the bill.