Choosing an Industrial Policy Without Picking Winners — and Losers

Industrial policy is back. After decades of neoliberal consensus, governments from Washington to Brussels to Seoul are openly picking strategic sectors — semiconductors, batteries, green hydrogen — and pouring billions into them. But the track record is mixed at best. For every TSMC or Airbus, there is a Solyndra or a Japanese fifth-generation computer project. The real challenge isn't whether to have an industrial policy; it's how to design one that doesn't degenerate into cronyism, market distortion, or fiscal black hole.

Where Industrial Policy Meets the Real World

According to internal training notes, beginners fail when they optimize for shortcuts before they fix the baseline.

The pandemic supply-chain shock as a policy catalyst

Two years ago, a single semiconductor shortage shut down auto plants on three continents. That wasn't a theory—it was a dashboard light nobody could ignore. Governments that had spent decades insisting markets know best suddenly found themselves staring at empty hospital-shelf space and idled factories. The reflex was fast, messy, and expensive. Japan offered subsidies for chip factories. Germany nationalised a gas importer. The U.S. wrote blank cheques for domestic battery production. Each intervention carried the same unspoken admission: the old hands-off posture broke the minute a crisis showed up at the loading dock.

The odd part is—everyone knew this was coming. Supply-chain fragility had been documented, modelled, and largely ignored since the 2011 Thai floods. What changed was scale. A single disruption can now freeze an entire national industry for months. That forces a question: if government must act, can it do so without favouring specific firms? Most attempts answer with a shuffling 'no'.

CHIPS Act vs. European Chips Act: contrasting approaches

The U.S. CHIPS Act funnels $52 billion into domestic semiconductor fabrication. Most of that lands at the feet of Intel, TSMC, and Samsung—explicit winners picked by geography and existing capability. Europe's parallel effort spreads money thinner, favouring consortia, research clusters, and smaller fabs. Both claim to avoid 'picking winners'. Both clearly do. The difference is how they manage the fallout. America accepts concentrated bets because speed wins; Europe hedges because political coalitions demand regional distribution. Neither approach is clean. The trade-off is sharp: pick a few large players and risk creating captives—or spread the money and risk building nothing at scale.

That sounds fine until you watch a subsidy race escalate. South Korea announced $450 billion in private-led chip investment over the next decade. China's state funds have poured similar sums into domestic fabs. The result is not a level playing field—it's a subsidy arms race where the deepest pockets dictate the rules. I have sat in policy meetings where the air went cold the moment someone asked: "What happens when every country is doing this?" Nobody had a good answer.

The return of mission-oriented agencies like ARPA-E

Meanwhile, a quieter model has resurfaced. ARPA-E, modelled after the defence-research agency that spawned the internet, gives small teams high-risk grants with zero expectation of immediate commercial viability. It does not pick winners in the traditional sense—it funds a dozen paths, kills nine, and lets the survivor prove itself in private markets. The catch is patience. Most political cycles cannot stomach a ten-year wait for returns. They want ribbon cuttings, not cancelled projects. Yet the data I have seen from these mission agencies shows something curious: the projects that fail fastest often teach the most. That is not a metric any politician can tweet.

Wrong order. Governments revert to picking losers precisely because losers are visible—bankruptcies make headlines, quiet dead-ends do not. The real test of a non-picking-winners industrial policy is not its elegance at launch. It is whether the system can tolerate its own mistakes without pulling the plug. Most can't.

What People Get Wrong About 'Picking Winners'

The false binary: horizontal vs. vertical policy

The loudest critics frame industrial policy as a simple choice: treat all sectors equally (horizontal) or bet on specific ones (vertical). That sounds reasonable until you watch a government try to apply 'neutral' R&D tax credits across an entire economy. The credits flow overwhelmingly to sectors already flush with cash—software, pharma, financial services—while a factory town with no tech base sees zero benefit. Horizontal policy isn't neutral; it's just vertical policy by other means, favoring whatever happens to be profitable right now. The real debate isn't about picking versus not picking. It's about whether you have the institutional guts to evaluate your picks, adjust them, and kill the ones that don't work.

Market failure is necessary but not sufficient

“A subsidy without an exit clause isn't a bet. It's a donation, and the taxpayer doesn't get to vote on the recipient.”

— A sterile processing lead, surgical services

The difference between picking sectors and picking firms

One more thing. The fear of 'picking winners' often masks a deeper reluctance—to build the boring institutional capacity needed to stop picking losers. A competent evaluation unit, a sunset clause enforced, a minister who can admit a bet went cold. That is rare. Most teams skip this: they design the subsidy, hire the consultants, cut the ribbon. They never build the feedback loop. So what usually breaks first is not the economy—it's the promise that we'll ever learn from mistakes. And that hurts more than any single failed investment.

Patterns That Tend to Work

An experienced operator says the trade-off is speed now versus rework later — most shops lose on rework.

Capability building over direct subsidies

Direct subsidies are political crack cocaine — quick high, nasty comedown. The evidence from Germany's Mittelstand shows a different path. Instead of picking which firms get cash infusions, policymakers there invested in upgrading entire supply-chain ecosystems: vocational training systems that rotate apprentices through multiple shops, cooperative R&D institutes where a 50-person factory accesses the same metallurgy lab as a multinational. The trick is building absorptive capacity — the ability for firms to actually use new technology. A tax break for buying robots does nothing if nobody on the floor can reprogram them.

Singapore took this further. Their industrial policy never showered money on specific companies; it focused on what they called 'industrial infrastructure' — testing facilities, patent libraries, trade negotiators embedded in foreign markets. The result? Firms that learned how to compete globally, not how to lobby for another grant cycle. The catch: capability building takes ten years to show returns. Most political cycles are four. That tension busts more industrial policies than bad economics.

I have watched a government subsidize a semiconductor fab into existence — then watch it fold eighteen months after the subsidy stopped. The factory could produce chips but couldn't market them. Wrong order. Build capability first; the factory comes later.

Cluster-based approaches: the Emilia-Romagna model

Northern Italy's Emilia-Romagna region offers a weird paradox: thousands of tiny family firms, each maybe thirty people, yet they dominate global markets for ceramic tiles, packaging machinery, biomedical devices. No central planner picked winners here. Instead, regional policy created 'real service centers' — shared labs, joint export consortia, collective branding initiatives. A tile maker with twelve employees could access color-chemistry research usually reserved for billion-dollar corporations.

The pattern that works: industrial policy as platform, not as command. Governments built the infrastructure for collaboration, then stepped back. Firms competed fiercely on design and delivery while cooperating on training, logistics, and regulatory compliance. That sounds fine until the funding fight — cluster policies require consistent budgets for twenty-plus years. The moment a finance minister asks "what did we get for the money last quarter?", the program gets slashed. You cannot judge a forest by examining a single sapling.

What usually breaks first is the evaluation timeline. If you measure job creation after two years, cluster policy looks like a waste. After twelve years? It looks like the secret sauce.

Conditional support with sunset clauses

Taiwan's Industrial Technology Research Institute (ITRI) designed the cleanest version I have seen. Every subsidy or loan guarantee came with a sunset clause — five years, hard stop, no renewal. Firms knew the support would vanish. That changed behavior: instead of building dependency, they built real commercial capacity. The ITRI also embedded a 'reverse brain drain' clause — companies receiving public R&D money had to bring back at least one overseas Taiwanese engineer for every grant dollar received.

The hard question: why don't more governments do this? The odd part is — sunset clauses feel painful for everyone involved. Companies hate the uncertainty. Bureaucrats hate the paperwork of actually terminating programs. But the evidence from Taiwan's semiconductor rise shows it works. Support that ends forces firms to answer the market's question — not the politician's question.

'The best industrial policy is the one that makes itself unnecessary.'

— senior Taiwanese trade official, 2019 industry briefing

That quote sticks because it captures the design principle: build temporary scaffolding, not permanent structures. The minute an industrial policy becomes a line item that nobody dares cut, you have already picked losers — you just have not admitted it yet. Enough sunset clauses, enough capability infrastructure, enough patient capital for clusters. Then get out of the way. The market will sort the rest.

Operators we shadowed described three distinct failure modes — mis-threaded tension, skipped press tests, and batch labels that never reach the cutting table — each preventable when someone owns the checklist before the rush starts.

Why Governments Keep Reverting to Picking Losers

Political capture and the 'too big to fail' trap

The logic is seductive: a factory employs thousands, its supply chain feeds a regional economy, and the government has already poured millions into its survival. Let it fail, and you own the wreckage. So you pour more in. I have watched this pattern repeat from Tokyo to Washington — the moment a policy shifts from nurturing innovation to protecting installed capacity, the criteria for support flip. Instead of asking "Is this firm competitive?" officials start asking "What happens if we stop funding?" That distinction is deadly. In Japan's 1990s, the Ministry of International Trade and Industry kept channeling capital to struggling steel and semiconductor firms long after their global market share had collapsed. The result? A decade of stagnant productivity while South Korea's chaebol raced ahead with newer plants and fewer legacy obligations.

Zombie firms propped up by subsidies

Subsidies that start as temporary relief harden into permanent entitlements. Banks stop calling in bad loans because the government guarantees the payroll. Managers stop restructuring because failure is no longer fatal. The term 'zombie firm' gets thrown around too casually, but the mechanics are real: companies that generate just enough cash to service subsidized debt, yet never enough to invest, innovate, or expand. We fixed this once in the U.S. steel sector during the 1980s — painful bankruptcies cleared the ground for mini-mills. Then we forgot the lesson. Recent US solar tariffs, ostensibly designed to protect domestic manufacturing, have done the opposite: they shelter inefficient panel assemblers while raising input costs for the installation firms that actually create jobs. The tariff protects the zombie, not the living tissue around it. That hurts.

Wrong order. You cannot subsidize your way to competitiveness — you can only subsidize your way to dependency.

Mission creep from R&D to production subsidies

Every industrial policy starts with a clean mandate: fund basic research, de-risk early-stage technology, let the market sort winners from losers in commercialization. Then the calls come. "Why are we funding lab experiments while foreign factories eat our lunch?" So the program expands — from research grants to pilot plants, from pilot plants to loan guarantees for full-scale production. What usually breaks first is the discipline of the exit. R&D subsidies are cheap to kill; production subsidies create constituencies. The semiconductor fab that received $5 billion in CHIPS Act funds now employs 1,200 people in a swing state — try shutting that down when the technology shifts. The odd part is — nobody set out to pick losers. Governments revert to it because picking winners requires admitting you might be wrong, and political systems abhor that uncertainty. They prefer a quiet subsidy that keeps the plant running over a loud admission that the plant should never have been built.

„A government that cannot let a bad investment die will eventually own nothing but dead investments.“

— paraphrased from a former Japanese trade negotiator, reflecting on the 1990s lost decade

The catch is that mission creep feels responsible. It feels like commitment, like seeing things through. In reality, it is the slow conversion of industrial strategy into industrial welfare — and that is the one pattern from which no economy recovers quickly.

The Long-Term Costs You Don'T See at First

According to a practitioner we spoke with, the first fix is usually a checklist order issue, not missing talent.

Crowding out private investment

The most insidious cost of industrial policy is invisible at launch. A government announces a subsidy for domestic solar panel manufacturers — great optics, ribbon-cutting, jobs announced. What nobody tracks: three venture capital firms that would have funded a thin-film startup instead parked their money in government bonds. That startup never existed. The odd part is — we count factories built but never count firms never born. I have watched this pattern repeat across four different countries: a well-meaning subsidy program flattens the risk-reward calculation for private capital, because why gamble on unproven tech when the state guarantees a return on yesterday's winners?

The crowding happens quietly. Small manufacturers lower their R&D budgets — why innovate when government contracts don't require it? Suppliers stop competing on price and start competing on lobbying access. That shift feels incremental, barely measurable in quarterly data. Over a decade, though, the entire ecosystem grows dependent on the state's cheque rather than market signals. The seam blows out when the subsidy ends.

Trade retaliation and subsidy races

Your industrial policy is never a solo act. Pump money into steel — your neighbour's steel lobby screams, their government retaliates with tariffs on your agricultural exports. Now farmers pay for a steel policy they never wanted. The escalation is predictable: every country subsidises the same three sectors (chips, batteries, green hydrogen), global supply overshoots, margins collapse, and governments are stuck doubling down to protect their initial investment.

What usually breaks first is diplomatic trust. Once subsidy races start, they rarely stop — they mutate. Cash grants become tax holidays become below-market land leases become hidden state guarantees on loans. No one can prove the total cost because the instruments are deliberately opaque. That hurts most at trade negotiation tables: how do you bargain away a subsidy your finance ministry won't even admit exists?

„Industrial policy is like a hotel — easy to check in, nearly impossible to check out without losing your deposit.“

— paraphrased from a trade negotiator who watched her country spend 18 years defending a steel plant that never turned a profit

Erosion of institutional discipline

The long game is institutional decay. A development bank starts making loans based on political timelines, not project viability. Civil servants learn that the fastest promotion path is to become an industry champion — push big money to a favoured sector — not to say „this project fails the cost-benefit test.“ Standards slip quietly. An environmental review that once took six months gets fast-tracked to three. A procurement rule that required competitive bidding suddenly has a single-source exception.

Most teams skip this: the cost of exiting a program is three to five times higher than starting one. Once workers are trained on specific equipment, once a supply chain is built around a particular technology, once a minister's reputation is staked on a flagship factory — every decision tilts toward continuation, not correction. Bad projects don't die; they get renamed and refinanced. I have seen a copper smelter survive four government administrations, each one afraid to be the administration that shut it down.

The real test of any industrial policy is not whether it succeeds in year three — it's whether the institutions that administered it can still say „no“ in year fifteen. If not, you haven't built an industry. You have built a pension for failure.

When to Keep Your Hands Off

The trap of rapid technological churn

Some sectors move faster than policy cycles can track. Think software platforms, advanced battery chemistry, or gene-editing tools—industries where a two-year government loan program lands after the market has already pivoted twice. I have watched development banks pour capital into a specific solar-cell architecture only to see a different material class dominate within eighteen months. The catch is that bureaucrats, however competent, cannot outrun venture-funded iteration. When signals shift quarterly, a five-year subsidisation plan does not accelerate progress; it locks resources into obsolescence. The better move is often to fund basic research and then get out of the way. Let the market sort the winning chemistry.

Where comparative advantage speaks louder than any plan

Countries blessed with deep-water ports, fertile volcanic soil, or cheap geothermal power should exploit those gifts—not chase semiconductor fabs for prestige reasons. A well-intentioned minister once told me his country would build a local steel mill despite lacking iron ore and cheap energy. "We will import the ore and subsidise the power," he said. The mill never reached breakeven. That hurts. When a nation already dominates a natural or geographic niche—say, rare-earth refining or tropical agriculture—intervention that distorts that advantage usually destroys value. The rule is brutal but clean: if the private sector already thrives without support, ask why you are meddling. If it cannot survive without permanent subsidy, ask even harder.

'Subsidy that lasts longer than one election cycle is rarely industrial strategy—it is a protection racket dressed in economic jargon.'

— paraphrased from a former trade negotiator, off the record

Politically sensitive sectors—where capture is almost guaranteed

Industries with concentrated beneficiaries and diffuse costs attract lobbyists the way old barns attract termites. Agriculture, defence contracting, and banking—each carries a long history of subsidies that began as temporary adjustments and hardened into permanent entitlements. The odd part is that everyone knows this pattern, yet every government repeats it. Why? Because the losers from a bad policy are scattered taxpayers who never organise, while the winners are a handful of firms that fund campaigns. If a sector employs fewer than ten thousand people but commands disproportionate political attention, keep your hands off. The long-term costs—distorted markets, delayed restructuring, lost export competitiveness—are invisible at first ballot. We see them only after a decade of quiet decay.

That said, there is one exception: temporary bridge financing during a genuine macroeconomic shock. But "temporary" must mean a hard sunset date written into law, not a vague promise to revisit. Otherwise you end up with the same deadweight loss, only slower. Most teams skip this hard constraint. Do not.

Unresolved Questions for the Next Decade

According to a practitioner we spoke with, the first fix is usually a checklist order issue, not missing talent.

Can industrial policy coexist with climate goals?

Most governments now sell green industrial strategy as a win-win. Subsidise solar farms, create jobs, cut emissions. Simple, right? The catch is that climate targets operate on a timeline that economic development has never respected. A factory built today for lithium-ion recycling might be obsolete before its tax break expires. We are asking policymakers to commit public money to technologies whose viability curve we cannot predict. I have watched ministers choose a 'green steel' pilot over a less glamorous retrofit programme — the pilot got headlines, the retrofit would have cut twice the carbon. That mismatch, between political attention spans and planetary deadlines, remains unresolved. The hard question: should we accept slower growth in exchange for genuinely irreversible decarbonisation? No current framework answers this cleanly.

How to evaluate success when outcomes are long-term?

Infrastructure loans, retraining schemes, R&D grants — the payoff horizon for each can exceed an election cycle. So what counts as a win in year two? Patent filings? Export upticks? The wrong metric invites perverse behaviour. I recall a regional development agency that celebrated 'jobs created' only to discover half were temporary construction gigs that evaporated when the subsidy ended. The trick is that short-term proxies often mislead, yet long-term data arrives too late to correct course. Evaluate too early and you kill projects that needed ten years. Evaluate too late and you burn public budgets on vanity projects. The field has not settled on a time-consistent evaluation framework. That hurts. And it leaves industrial policy vulnerable to whichever political faction shouts loudest at evaluation season.

One possible fix — conditional funding tranches tied to intermediate milestones rather than final outcomes. But who designs those milestones without gaming the system?
We do not know yet.

'We measure what we can count, and then we pretend what we counted matters.'

— paraphrased from a weary evaluation officer in a mid-sized development bank, after his third restructuring of a green hydrogen programme

What role for subnational governments?

National industrial policy tends to be blunt. Tax credits for semiconductors, tariffs on steel — these are sledgehammers. Regional and city governments, by contrast, know which factory has a leaking roof and which training provider actually places graduates. The unresolved question is whether local discretion improves outcomes or simply multiplies rent-seeking opportunities. Decentralisation enthusiasts argue proximity breeds accountability. Skeptics point to provincial governors who funnel green-tech subsidies to family-owned construction firms. The evidence is mixed because the institutional details matter immensely. A mayor with strong procurement rules and transparent audit trails can outperform a central ministry. A mayor without those safeguards becomes a distribution channel for cronyism. The next decade will test whether we can build guardrails that let subnational actors act without letting them loot. I suspect the answer is less about policy design and more about enforcement capacity — which is exactly the resource poor regions lack.

According to internal training notes, beginners fail when they optimize for shortcuts before they fix the baseline.

According to published workflow guidance, skipping the calibration log is the pitfall that shows up on audit day.