Industrial Policy Without the Slogans
Subsidy programs are easy to announce and hard to evaluate. Enough time has now passed to ask the uncomfortable question: did they work?
Ground has been broken. Ribbons have been cut. Fabs that existed three years ago only as renderings on a slide deck now employ thousands of people pouring concrete in Ohio, Arizona, and upstate New York. By the most literal measure of industrial policy — did money get spent and did buildings get built — the answer is unambiguously yes. The harder question, the one policymakers would prefer to defer indefinitely, is whether any of it was worth doing on the terms originally promised. Enough time has now elapsed since the first checks cleared to attempt an answer, and the honest version is neither the triumphant one nor the cynical one.
Start with what the money actually bought, because the public accounting of subsidy programs tends to conflate three distinct things: capital committed, capital deployed, and capacity that will still exist in ten years. A headline subsidy figure — call it on the order of tens of billions across the major semiconductor and battery programs — describes commitments, not outlays. Disbursement has lagged announcement by design, since the incentive structure was built around milestones: hit a construction date, hit a hiring threshold, hit a yield target, and the next tranche releases. That structure is defensible policy design. It also means that a company can announce a marquee project, capture the headline and the stock-price bump, and then quietly slip the timeline eighteen months without anyone outside a trade press ever writing the correction. Several of the largest announced projects in the sector have done exactly this, and the slippage rarely makes it back into the summary statistics that get cited a year later.
The measurement problem is not a technicality
The honest difficulty in assessing programs of this kind is that the counterfactual is unobservable and the timeline for judging success extends well past any election or budget cycle. A subsidized fab that reaches volume production in year four looks, at that moment, indistinguishable from a fab that would have been built anyway, slightly later, with slightly less favorable terms, absent the subsidy. Distinguishing “additional” investment from “redirected” or “accelerated” investment requires knowing what a firm’s capital allocation committee would have decided in a world that didn’t exist — which is to say, it requires a model, not a fact. Economists who study place-based subsidies have wrestled with this problem for decades in the context of stadium deals and corporate relocation packages, and the consistent finding is that self-reported “but for the subsidy” claims from the recipient firm are close to worthless as evidence, because the firm has every incentive to claim maximal counterfactual dependence regardless of the truth.
What can be measured with more confidence is capacity: wafer starts, module lines, employment on-site, supplier ecosystems forming in the surrounding metro area. On that dimension the record is genuinely mixed-to-positive. Several regions that had essentially no advanced manufacturing presence a decade ago now have a real one, with second- and third-order effects — a testing-and-packaging supplier here, a specialty gas vendor there — that are the sort of agglomeration economists actually believe in. That is not nothing. But capacity is not the same as competitiveness, and it is the second question — will this capacity be cost-competitive with East Asian production once the subsidy tapers off — that the programs were implicitly supposed to answer and mostly have not yet been tested against, because most of the subsidized capacity has not yet had to compete unsubsidized.
The programs bought capacity with reasonable efficiency. Whether they bought competitiveness is a question the data cannot yet answer, and anyone who tells you otherwise is reading their priors, not the record.
Where the money actually went
A less comfortable line of inquiry concerns distribution. Subsidy programs of this scale inevitably produce a gap between the sectors intended as primary beneficiaries and the sectors that captured the largest dollar amounts. Leading-edge logic fabrication absorbed a disproportionate share of headline subsidy dollars relative to its share of the stated national-security rationale, which nominally centered on a broader base including legacy-node chips, packaging, and materials — the less glamorous parts of the supply chain that are, if anything, more exposed to single-source foreign dependency than cutting-edge logic. The politically legible project — the giant fab, the recognizable brand, the photogenic groundbreaking — crowds out funding for the unglamorous middle of the supply chain that a security-minded observer would actually prioritize. This is not corruption; it is the predictable result of a process where legislators need visible wins and agencies need defensible, low-controversy grant recipients. But it means the portfolio of what got built is a reasonable proxy for what photographed well, and only a rough proxy for what closes the vulnerability the policy was sold to close.
Labor costs and construction costs are the other place where the promised numbers and the delivered numbers have diverged, and diverged in a fairly predictable direction. Building a fab in the US costs more than building the equivalent fab in Taiwan or South Korea — estimates in the trade press have put the premium at somewhere in the range of thirty to fifty percent of total project cost, driven by specialized-labor scarcity, permitting timelines, and a construction sector that simply has less recent experience with cleanroom-grade builds. Subsidy design treated this premium as a fixed gap to be closed with a check. In practice the premium has proven somewhat elastic — it compresses as more projects create a domestic ecosystem of contractors and skilled trades who’ve done this before — but that compression happens on a timescale of years, and the first cohort of subsidized projects paid the full uncompressed premium, which shows up as lower effective subsidy-per-dollar-of-capacity than the original models assumed.
What a second-generation program would need to look like
None of this argues for abandoning the instrument. It argues for redesigning the accountability structure around it, and the fixes are not exotic. First, separate the announcement from the disbursement in public reporting, so that headline commitment figures stop functioning as a proxy for delivered capacity — a simple dashboard distinguishing committed, disbursed, and operational capital would do more for honest evaluation than any amount of retrospective academic work. Second, build in a genuine sunset review — a fixed date, five to seven years out, at which an independent body assesses whether subsidized capacity is on a plausible path to unsubsidized cost-competitiveness, with the finding made public regardless of how uncomfortable it is for the agency that ran the program. Third, weight the unglamorous middle of the supply chain — packaging, materials, legacy nodes — more heavily relative to headline fabs, since that is where the actual single-source vulnerability concentrates and where political incentives are least likely to deliver adequate funding on their own.
The deeper lesson is about time horizons and honesty. Industrial policy of this scale is, functionally, a multi-decade bet dressed in the language of a single legislative session. The programs so far have bought real capacity, at a real and somewhat higher-than-advertised cost, distributed somewhat unevenly relative to the stated rationale. Whether that capacity survives contact with unsubsidized competition is a question the next five years will actually answer, whether or not anyone in office today wants to be around to hear it.