Introduction: The Space Act versus CHIPS Act model

The Space Act versus CHIPS Act model comparison asks a deceptively simple question. When the United States wants a technology to exist, does it build the capability to make it happen, or does it pay someone else to build it? The two statutes at the center of this article answer that question in opposite ways. National Aeronautics and Space Act of 1958, Public Law 85-568, created a civilian agency, hired the scientists and engineers, took title to the inventions, and bought spaceflight as a government mission. The CHIPS Act of 2022, Division A of Public Law 117-167, appropriated billions of dollars to persuade private semiconductor firms to build fabrication plants in the United States, with conditions attached, while leaving ownership, operation, and commercial risk in private hands. The full statutory anatomy appears in the CHIPS and Science Act guide.

One model builds state capability and procures a mission. The other subsidizes private capital with conditions attached. Both models can be called industrial policy, and industrial policy is contested ground. That contest deserves an even hand. A Republican administration designed and built the agency model, and presidents and legislators of both parties later supported the subsidy model, so neither model belongs to one party. This article presents the state-capability argument and the market-efficiency argument with equal care, and it makes no claim about industrial policy in general.

The deciding factor in the verdict, which the rest of this article defends, is whether the government is the customer for the technology in question. When the government is the customer, meaning it defines the requirement, receives the deliverable, and stands responsible for the mission, the agency model fits. When the government is not the customer, meaning it wants private actors to produce something it will never take delivery of, the subsidy model fits. The Space Act assumed the first condition. The CHIPS Act assumed the second. Once that distinction is clear, the choice between the models stops being a debate about ideology and becomes a question about the nature of the transaction.

The reader should be able to hold both models in mind at once, name the statute that exemplifies each, and state the deciding factor without hesitation. The test of this article is whether that reader, asked to explain the comparison, can say that the Space Act created an agency that held engineers and laboratories and bought a mission through procurement contracts, that the CHIPS Act funded a Commerce Department incentive program that attached conditions to private capital, and that the difference turns on who takes delivery of the technology. If that summary is clear, the article has done its work.

Consider what each statute asks the reader to imagine. The Space Act asks the reader to imagine a government that does the thing itself, because no one else will. The transaction is direct: the public pays, the public’s agency performs, and the public receives the mission. Accountability runs through the chain of command and the appropriations process. The CHIPS Act asks the reader to imagine a government that changes someone else’s arithmetic, because someone else already does the thing and the government wants more of it done domestically. The transaction is indirect: the public pays, a private firm decides, and the public receives a national outcome, domestic fabrication capacity, rather than a deliverable. Accountability runs through the award agreement and the audit. Neither transaction is mysterious, but they are different transactions, and the conditions each can enforce follow from the difference.

The sections that follow trace that difference through the statute that built the agency. The origins section explains why the 1958 design assumed the government was the customer. The capability section describes what state ownership of talent and laboratories made possible and what it cost. The instrument section compares mission procurement with outcome buying and the two kinds of leverage each creates. The invention ownership section shows the agency model taken to its full logic, with the government holding title to the inventions its contracts produced, and explains why that logic had to be corrected by two technology transfer statutes in 1980. The CHIPS Act appears throughout as the counterpoint, and the deciding factor, whether the government is the customer, accumulates its force with each section.

A final note on posture. This article does not ask which model is better in the abstract, because the question as posed has no answer. It asks which model fits which transaction. The neutrality rule is not a courtesy. It is the method. A Republican administration built the agency model, both parties supported the subsidy model, and the verdict will be defended on the deciding factor alone, not on the politics of either design.

Origins of the 1958 Model

The 1958 model was born on October 4, 1957, when the Soviet Union launched Sputnik 1, an 83.6-kilogram polished sphere that became the first artificial satellite to orbit the Earth. The satellite itself was modest. Its beep, broadcast over American radio frequencies, carried the implication that Soviet rockets could deliver more than a sphere. The launch detonated the American assumption of unchallenged technological leadership and forced a strategic choice about how the nation would organize its answer.

The existing institutional landscape offered no obvious home for a national space program. The National Advisory Committee for Aeronautics, founded in 1915, conducted respected aeronautical research at laboratories in Langley, Ames, and Lewis, but it was a research committee, not a mission agency. The military services ran their own rocket and satellite efforts, and the fragmentation was public. The Navy’s Vanguard program, the officially designated American satellite effort, failed when Vanguard TV-3 exploded on the launch pad on December 6, 1957, before a national television audience. The Army’s team under Wernher von Braun succeeded with Explorer 1 on January 31, 1958, discovering the Van Allen radiation belts, but its achievement only underscored the disarray: three services competing, one Navy program in flames, and no coherent national plan. The country had satellites in orbit and rockets on drawing boards, but it had no institution whose job was spaceflight.

President Eisenhower moved within weeks of Sputnik. On November 13, 1957, he called publicly for the establishment of a civilian space agency. The President’s Science Advisory Committee studied the options and recommended expanding the NACA into a broader agency charged with the mission to plan, direct, and conduct aeronautical and space activities. The reasoning was institutional, not partisan. The NACA’s civilian character, recognized technical excellence, and research-focused reputation made it an acceptable vehicle that would not escalate Cold War tensions the way a military program would, and it already employed the aerodynamicists and owned the wind tunnels that any space program would need. The NACA itself had reached the same conclusion from the inside: at the start of 1958 it created a Special Committee on Space Technology, acknowledging that space had become central to its research mission and that the committee’s civilian laboratories would need a larger mandate. In April 1958, Eisenhower sent Congress a message asking for legislation to create that civilian agency. NASA’s own historical account records that bipartisan agreement between a Republican president and a Democratic-controlled Congress produced the drafting of the bill, a fact worth stating plainly because it demonstrates that the agency model was a national-security response rather than a partisan project.

The military-versus-civilian question was the central design decision, and Congress considered the alternative seriously. The Army and the Air Force each had plausible claims to the space mission: the Army had von Braun and the only American team with a satellite in orbit, and the Air Force had the missile programs that would supply the heavy launch vehicles. A military space agency would have been faster to stand up and easier to fund through defense appropriations. The civilian choice won because the administration and the committee judged that a military program would feed the arms race dynamic the nation was trying to manage, while a civilian program could conduct science openly, publish its results, and cooperate internationally. The statute encoded that judgment in its declaration of policy, which provided that the nation’s aeronautical and space activities should be devoted to peaceful purposes for the benefit of all humankind, and in its requirement that the agency disseminate information about its activities as widely as practicable. Those were not decorative phrases. They were the civilian model’s operating theory: openness as strategy, and science as the instrument of national prestige.

Congress began hearings in November to develop a long-term space plan, with the Senate’s preparedness subcommittee under Lyndon Johnson examining the satellite and missile programs in hearings that ran into early 1958. The hearings served two purposes at once: they investigated why the American satellite effort had lagged, and they built the legislative record for the civilian agency the president would propose in April. By the time Eisenhower’s message arrived, the committees had already heard months of testimony about fragmented programs, duplicated effort, and the absence of a single responsible institution. The speed of the legislative process that followed, from the April message to the July signature, reflected a Congress that had already done its deliberating in the hearing room. The agency model was not improvised in a panic. It was the product of nine months of public inquiry into what had gone wrong, followed by a presidential proposal that answered the inquiry’s central finding. Eisenhower signed it on July 29, 1958, as Public Law 85-568. The new agency absorbed the NACA in its entirety: the laboratories, roughly 8,000 employees, and an annual budget of about 100 million dollars. It also absorbed the Vanguard project and other non-military space assets scattered across the government. The Jet Propulsion Laboratory, the Army’s rocket research operation at the California Institute of Technology, came under NASA management before the end of 1958, and von Braun’s Army team at Huntsville followed in 1960 to become the nucleus of the Marshall Space Flight Center. NASA opened for business on October 1, 1958, less than a year after Sputnik, with T. Keith Glennan sworn in as its first administrator in August. Glennan, a university president rather than a rocket engineer, embodied the civilian character of the enterprise. The deeper fact about the opening was personnel continuity: the engineers who had been NACA civil servants on September 30 were NASA civil servants on October 1, working in the same laboratories on an expanded mission. The agency model did not hire a workforce. It inherited one. That inheritance is the reason the model could move as fast as it did, and it is also the reason the model is hard to replicate. A government that already employs the talent can aim it at a new mission in weeks. A government that does not must either build the workforce over a generation or buy the mission from someone who has. The 1958 designers were spared that choice because the NACA gave them the seed crystal. The CHIPS Act’s designers faced the opposite starting point: no federal workforce makes semiconductors, so the statute could only pay private workforces to make them domestically.

The design logic of the 1958 model followed from the nature of the problem as Congress and the president understood it. Spaceflight in 1958 was not a market that needed incentives. It was a national undertaking for which no market existed, no customer besides the government existed, and no private firm could plausibly finance the capability. No American company was going to build a lunar program on speculation, and no procurement contract written in 1958 could have summoned one into existence, because the knowledge, the facilities, and the engineers did not yet exist in private hands at the required scale. The statute therefore did not offer rewards to private actors for doing spaceflight. It created an organization that would do spaceflight. The mandate was deliberately broad, covering the conduct of aeronautical and space activities, the involvement of the scientific community, and the wide dissemination of information about those activities. Congress paired this breadth with a civilian identity, a choice designed to separate peaceful scientific and exploratory work from military programs and to keep the nation’s research reputation intact abroad.

Three features defined the model from the start. First, the government would hold the talent, employing scientists and engineers directly rather than relying on contractors to maintain the capability. Second, the government would own the laboratories and test facilities, the physical plant of research. Third, the government would define the missions and remain responsible for them, contracting with industry for hardware but retaining mission authority. The civilian character was a fourth, quieter feature, signaling that the enterprise belonged to the nation rather than to the armed services. No part of this design assigned the work to a political party or an economic philosophy. It assigned the work to the state because, in 1958, the state was the only customer for human spaceflight, and the architects of the statute built accordingly.

The model proved itself operational almost immediately. Within a week of opening for business, NASA announced Project Mercury, the nation’s first human spaceflight program, giving the inherited NACA engineers and the new agency’s contractors a concrete mission before the paint was dry. When President Kennedy committed the nation in May 1961 to landing a man on the Moon before the decade was out, he was not inventing the agency model. He was giving it the largest mission it would ever carry out, and the appropriations that followed, rising to the mid-1960s peak, funded the model at the scale its architects had assumed it would need. The sequence matters: the institution came first, the mission second, and the funding third. That is the agency model’s characteristic order of operations, and it is the reverse of the subsidy model’s, which identifies a private actor’s decision first and applies the funding to change it.

Capability

The agency model expresses itself most visibly in what it owns. Under the Space Act, the United States did not simply pay for spaceflight. It acquired the capacity to do spaceflight. The NACA’s Langley, Ames, and Lewis laboratories became NASA field centers on day one, and the agency then built the new facilities its mission required: the launch ranges, the Manned Spacecraft Center at Houston in 1961, the Marshall center at Huntsville in 1960, and the deep-space communications network. The laboratories were not leased capability or contracted access. They were government property, staffed by government employees, maintained with public funds, and directed by public mission requirements. When a new mission arose, the state did not need to assemble a supply chain from scratch or coax a contractor to invest. It already held the workforce and the plant, and it could aim them at the new objective.

Holding capability directly shaped how the nation learned. Because the same civil-service scientists and engineers carried institutional knowledge from one program to the next, lessons learned in aeronautics informed rocketry, and lessons learned in uncrewed probes informed crewed missions. The agency’s employees could move between centers, between research and operations, and between successes and failures, carrying judgment that no contract could easily transfer. The centers themselves specialized the way organs specialize: Langley in aerodynamics and structures, Lewis in propulsion, the Goddard center in space science, Houston in human spaceflight operations, Huntsville in launch vehicles, and the Jet Propulsion Laboratory in deep-space exploration. Each center accumulated a depth of practice in its domain that no single procurement could have purchased, because the depth came from doing the work repeatedly under the government’s own roof.

NASA also performed the systems-integration function that holds a complex program together, and this is the capability least visible to outsiders and hardest to contract for. Someone must define the interfaces between the launch vehicle, the spacecraft, and the ground systems; someone must adjudicate the trade-offs between contractors whose hardware has to fit together in a configuration no contractor designed alone; and someone must retain the authority to halt a mission on safety grounds when a contractor’s schedule says go. NASA kept that role because it employed the engineers capable of exercising it. Mission Control in Houston, the flight dynamics teams, the test ranges, and the failure-review boards were all expressions of the same choice: the state would hold the judgment, not merely the funding. Industry contractors participated at enormous scale. North American Aviation built the Apollo command and service modules, Grumman built the lunar module, and hundreds of other firms built engines, instruments, and ground equipment. At the program’s peak in the mid-1960s, the civil-service workforce numbered roughly 36,000 while contractor employment ran to roughly 376,000. The ratio tells the story of the model: a relatively small government core, holding the specifications, the interfaces, and the mission authority, directing an industrial base an order of magnitude larger. The critical distinction held throughout: the contractors built hardware for a customer that could evaluate the hardware, because the customer itself employed engineers who understood the physics, could write the specifications, and could say no.

This arrangement produces a distinctive set of advantages and costs, and the two sides of the ledger deserve equal weight. Proponents of state capability argue that some technologies are too important, too slow to pay off, or too entangled with national security to be left to the investment calendars of private firms. A mission as open-ended as space exploration, where the deliverable is knowledge and national standing rather than a product, fits that description. The returns on fundamental research and on missions that must not fail arrive over decades and cannot be captured by any single firm’s balance sheet, so the state that needs the capability must hold it. Building capability inside the government also preserves the option value of readiness: the engineers are present when the crisis arrives, not on the other side of a procurement cycle, and the institutional memory survives the electoral cycle that created the program.

Critics of state capability answer that government agencies ossify, that civil-service workforces reward longevity over excellence, and that public ownership dulls the cost discipline that competition imposes. They point to the Apollo-era cost-plus contracting culture, in which contractors recovered allowable costs plus a fixed fee, as evidence that when the customer is guaranteed and the mission is grand, spending expands toward the available budget. The Apollo program succeeded on schedule and before the decade ended, but at its mid-1960s peak it consumed over four percent of the federal budget, a share no successor program could sustain, and the agency’s subsequent programs inherited habits of requirements creep and schedule optimism that the fixed customer base did little to correct. A cost-plus environment teaches contractors to manage the customer relationship rather than the cost baseline, and it teaches the agency to manage the schedule rather than the budget, because the budget is the one variable the appropriations process keeps supplying. A standing agency also creates its own political constituency: centers, contractors, and congressional delegations develop stakes in the agency’s budget that can outlive the missions the budget was meant to fund, turning capability into an end in itself. Programs persist because the workforce exists, and the workforce exists because the programs persist. Both critiques have force. The first says the market underinvests in missions without commercial customers. The second says the state overspends once the customer is guaranteed. The agency model accepts the second risk as the price of solving the first problem.

Capability also persists between missions, and that persistence is both the model’s strength and its fixed cost. When Apollo ended, the workforce did not disperse. The same engineers, the same centers, and much of the same hardware were re-aimed at Skylab, the Apollo-Soyuz test project, and then the Space Shuttle, because the state held the capability and could redirect it. A subsidy program ends when the money is spent; the fabs remain, but the government’s relationship with them ends at the award’s closeout. An agency endures between appropriations, carrying institutional memory through the quiet years so that the next mission does not start from zero. The price of that endurance is the fixed cost of the standing workforce and the political economy that grows around it. Centers must be fed with programs, programs must be justified to appropriators, and justification drifts over time from the mission to the institution. The capability model solves the problem of starting from zero by making it expensive to ever stop.

The capability model also changes how an enterprise handles failure, and the difference is instructive. When the Apollo 1 fire killed three astronauts in January 1967, NASA investigated the disaster with its own engineers, redesigned the command module’s systems, and returned to flight, because the technical judgment needed for both the investigation and the redesign lived inside the agency. A funder without capability cannot do that. It can commission a report, withhold the next tranche of money, or terminate the award, but it cannot reach into the design and fix it. The agency model’s answer to catastrophe is institutional: the same people who built the system diagnose it. The subsidy model’s answer is contractual: the funder’s leverage ends at the award agreement’s remedies. Each answer fits its transaction. A customer that owns the mission needs the engineering judgment to recover from failure. A funder that does not own the factory can only price the risk of failure into the award and hope the conditions hold.

The contrast with the CHIPS Act model clarifies what capability means here. Under the CHIPS Act, the Department of Commerce administers financial incentives for semiconductor manufacturing, but the department does not employ chip engineers, does not operate fabrication plants, and does not take delivery of the chips. The private firms that receive the funds build and run the fabs, hire the workforce, manage the supply chains, and bear the commercial risk of filling the order books. The state holds no capability to make semiconductors. It holds a checkbook and a set of conditions. This is not a criticism of the CHIPS design; it follows from a different premise. If the premise is that the technology will have commercial customers and the government wants more of the production to happen domestically, then owning the capability would be wasteful and possibly harmful, since a government-run fab would compete with the private fabs the policy is trying to encourage, and the state’s engineers would be managing a commercial enterprise rather than a mission. The agency model assumes the opposite premise: that the government is the customer, that no commercial market can sustain the capability, and that the nation therefore needs the state to hold the talent, the laboratories, and the responsibility in its own hands. The premise selects the model, not the other way around.

Instrument

The two models also differ in the legal instrument through which the government acts, and that difference determines how conditions are enforced. Under the Space Act model, the government procures a mission. NASA issues contracts for hardware, services, and research under federal procurement law, and the conditions the government imposes arrive as contract terms: technical specifications, delivery schedules, inspection rights, testing requirements, data rights, and termination clauses. A contractor that fails to perform faces the machinery of procurement law, including termination for default, disputes before the boards of contract appeals, and remedies measured against the contract’s terms. The relationship is that of buyer and seller, and the government’s leverage comes from being the buyer. It can withhold payment, reject nonconforming deliverables, demand corrective action, and choose a different seller next time. The conditions are enforceable because the government is on the receiving end of the transaction, inspecting what arrives and paying only for what conforms.

The procurement instrument evolved as the programs matured. In the Apollo years, NASA relied heavily on cost-plus contracts, under which the government bore the financial risk of development because the technology was so novel that no contractor could price it firmly. Over time, as launch vehicles and spacecraft systems became better understood, the agency shifted toward fixed-price and performance-based arrangements for more routine work, reserving cost-type contracts for genuine development. The federal procurement system itself was consolidated in the Federal Acquisition Regulation in 1984, which standardized the contract types, competition requirements, and dispute procedures across agencies. The evolution did not change the transaction’s shape. Whether the contract was cost-plus or fixed-price, whether the dispute went to a board of contract appeals under the Contract Disputes Act of 1978 or to the Court of Federal Claims, the government remained the buyer, the specifications remained the measure, and the deliverable remained something the government would take into its own hands. The Space Act also gave the agency flexible contracting authorities beyond standard procurement, including the cooperative Space Act Agreements under section 203(b)(5), which allowed NASA to partner with outside parties on shared-interest activities without the full machinery of a procurement contract. Article 171 of this series examined those agreements in detail. The point for this comparison is that every instrument in the agency model’s kit, from the strictest cost-type development contract to the most flexible agreement, keeps the government in the customer’s seat. The government defines the requirement, evaluates the result, and stands accountable for the mission.

Under the CHIPS Act model, the government buys an outcome rather than a mission. The statute’s semiconductor division, Division A of Public Law 117-167, signed on August 9, 2022, establishes and funds the CHIPS for America Fund with appropriations of roughly 52.7 billion dollars, including 39 billion dollars for incentives to build, expand, and equip domestic fabrication facilities and 11 billion dollars for semiconductor research and workforce development through the National Institute of Standards and Technology, a National Semiconductor Technology Center, a National Advanced Packaging Manufacturing Program, and related institutes. A separate 25 percent investment tax credit for semiconductor manufacturing equipment costs extends the subsidy through the tax code. The Department of Commerce awards the incentives as grants, loans, and loan guarantees to private firms that apply and meet the program’s criteria, working through a dedicated program office that issued funding notices, evaluated applications, and negotiated award terms with each recipient. The government takes no delivery of semiconductors. The firms sell their chips to commercial and defense customers as they always did, with American fabrication capacity that the subsidies helped pay for. The statute’s semiconductor provisions were first authorized in the National Defense Authorization Act for Fiscal Year 2021, Public Law 116-283, which created the incentive program that Division A of the later law funded.

The conditions attached to the subsidy are real, and they are enforceable, but they operate through administrative grant law rather than procurement law. Recipients face guardrails limiting expansion of advanced semiconductor manufacturing in foreign countries of concern. They face restrictions on stock buybacks and dividends during the award period, conditions designed to keep taxpayer money flowing into fabrication rather than shareholder returns. They accept prevailing-wage requirements for construction, workforce and community commitments, and clawback provisions that allow the government to recover funds if recipients fail to meet their obligations. The Department of Commerce ran the program as a negotiated process: it issued notices of funding opportunity, invited applications, evaluated them against published criteria, and negotiated individual award agreements with each recipient, converting statutory conditions into binding contractual commitments with compliance schedules, reporting requirements, and audit rights. When the law was signed, the administration promised the grants would not be blank checks and that guardrails would protect taxpayer dollars, a promise that translates legally into those negotiated agreements. A recipient that violates the terms risks revocation, repayment, and administrative sanctions. But the relationship remains that of funder and grantee, not buyer and seller. The government cannot reject a nonconforming fab and buy a different one, because it is not buying fabs. It is paying to shift the recipient’s investment decision, and its enforcement tools run through the award agreement rather than through inspection of a deliverable.

The distinction matters because the two instruments create different kinds of leverage and different failure modes. Mission procurement concentrates leverage in the buyer’s hands: the government specifies what it wants, inspects what it receives, and pays for performance. Its failure mode is the well-documented pathology of captive contracting, in which the buyer and the seller become so entangled that competition atrophies and the government ends up funding the contractor’s business rather than its missions. Cost-plus development contracts, sole-source awards, and requirements written around a single supplier’s capabilities are the classic symptoms, and the agency’s own engineers can become the captive’s advocates, defending the incumbent contractor’s approach because it is the approach they helped design. Outcome buying reverses the leverage: the government wants something the private actor might do anyway, so it must make the offer attractive enough to change the actor’s calculus without becoming so generous that the actor takes the money and behaves as it would have behaved regardless. Its failure mode is windfall, subsidy capture, and compliance theater, in which conditions are technically satisfied while the underlying behavior changes little. A firm might build the fab it was always going to build, accept the conditions as a cost of doing business, and pocket the difference. Money is fungible, and attributing any particular investment decision to the subsidy requires a counterfactual no audit can fully establish. Equal care requires naming both pathologies. The agency model risks a contractor ecosystem that cannot survive without the agency. The subsidy model risks a set of conditions that cannot reach the boardroom decisions that matter. The instrument follows the transaction: buy the mission when you are the customer, subsidize the outcome when you are not. The conditions themselves illustrate the point. Under mission procurement, conditions are priced into the transaction before it begins. The contractor reads the specifications, the inspection regime, and the termination clauses, and bids a price that reflects them. If the conditions are onerous, the price rises or the bidders walk away, and the government learns what its requirements cost. Under outcome buying, the conditions are negotiated around a decision the firm has largely made. The firm knows its true investment calculus and the government does not, which gives the firm the information advantage in every negotiation over how generous the award must be and how binding the conditions can be before the firm walks away. The government can protect itself with guardrails and clawbacks, and the CHIPS program did, but it cannot eliminate the asymmetry. The buyer in a procurement knows what it is buying. The funder in a subsidy program knows only what the recipient agrees to accept.

Enforcement follows the same divide. In procurement, the government’s ultimate remedy is to stop being the customer: terminate the contract, for default if the contractor failed or for convenience if the mission changed, pay the settlement the law requires, and take the requirement elsewhere. The remedy works because the government is the source of the demand. In a subsidy program, the government’s ultimate remedy is to take the money back, which works only to the extent the money has not already been spent on concrete and steel, and which leaves the underlying outcome, the fab the nation wanted, unbuilt. Revoking an award punishes the recipient but does not produce the fabrication capacity. The two instruments therefore fail differently by design. Procurement fails expensively and visibly, with the customer paying for its own changing requirements. Subsidy fails quietly, with the conditions satisfied on paper and the investment decision unchanged in substance. A clear-eyed comparison names the failure mode each instrument owns.

Invention Ownership

The Space Act carried the agency model to its logical conclusion in the treatment of inventions. Section 305 of Public Law 85-568, the title provision, provided that any invention made in the performance of work under a contract of the Administration would be the exclusive property of the United States, with patentable inventions to be patented in the name of the United States on application by the Administrator, unless the Administrator waived all or any part of the government’s rights in conformity with subsection (f) of the section. The logic was direct: the taxpayer had paid for the research, so the taxpayer should own the result. The statute further provided that no patent would issue to an applicant other than the Administrator for an invention with significant utility in the conduct of aeronautical and space activities unless the applicant filed a statement of the circumstances of the invention, a gatekeeping provision that extended the title policy beyond NASA’s own contracts. Article 171 of this series found that the title rule was waivable by the Administrator, and the statute’s waiver standard required a determination that the interests of the United States would be served by the waiver, a standard that NASA applied through its waiver regulations and its Invention and Contributions Board.

The government-title default created a serious problem, and it was recognized almost immediately. Industry contractors, whose engineers made many of the inventions under NASA contracts, had little incentive to commercialize technology the government owned. The NASA Administrator told Congress that the patent provisions could be detrimental to the kind of industry cooperation the joint effort required, and in July 1959 the House committee overseeing space announced a subcommittee to review the whole subject of the agency’s patent requirements. The concern was not abstract. A title policy that assigned every contract invention to the government left the inventions legally available but commercially orphaned, since no private firm would invest in developing a product built on a patent it did not own and could not license exclusively. NASA did grant waivers, and the regulations described the waiver process as liberal and expeditious, with waivers available in advance of an agreement for all inventions made under it or on a case-by-case basis for individual inventions. The Invention and Contributions Board reviewed the requests, and the agency’s practice over the 1960s and 1970s was to grant them routinely where the contractor could show a plan to commercialize. But the default rule still shaped behavior. Contractors negotiated around it, universities steered their researchers away from it, and the pipeline from government-funded research to commercial product stayed narrow. The waiver was an escape hatch, not a design feature, and it could not overcome the signal the default sent: invent under a NASA contract and the invention belongs to the government. The statistics of the orphan portfolio bore that out. A waiver process that grants most requests still requires the request, the review, and the delay, and each of those steps taxes the commercialization decision at the margin. Defaults govern behavior more than exceptions do, which is why Congress eventually changed the default rather than merely liberalizing the waiver.

By the late 1970s the scale of the orphan portfolio was notorious. Contemporary accounts placed the number of federally held patents at roughly 28,000 to 30,000, with only a small fraction ever licensed to anyone who might bring the inventions to market. Thousands of taxpayer-funded inventions sat in government files, legally the exclusive property of the United States and practically the property of no one. The government had financed the easy part, the research, while denying the private sector the title it needed to finance the expensive part, the development and commercialization that turns an invention into a product. Congress concluded that the title rule, however principled, was defeating the purpose of the research it was meant to protect, and it acted twice in the same year to fix the transfer problem.

The first statute, the Stevenson-Wydler Technology Innovation Act of 1980, Public Law 96-480, enacted October 21, 1980, was the first major technology transfer law. It declared that technology transfer was an economic function of government, required federal laboratories to budget for and actively participate in transfer activities, established offices of research and technology applications at the laboratories, and gave laboratory inventors a share of royalties, with the first two thousand dollars each year plus fifteen percent of additional royalties going to the inventor. Critics found the act’s machinery weaker than its declaration: the law required the transfer offices and the budgets but provided little enforcement, Congress withheld much of the funding the law contemplated, and a mid-1980s review found that of dozens of government-funded technical facilities, fewer than half had even a full-time person assigned to technology transfer. The mandate was real and the compliance was thin, which is why the 1986 amendments had to supply the cooperative research and development agreements the original act lacked. The second statute, examined in the Bayh-Dole guide, was formally the Patent and Trademark Law Amendments Act of 1980, Public Law 96-517, enacted December 12, 1980, and it changed the ownership rule itself. It allowed universities and small businesses to elect to retain title to inventions made with federal funding, subject to a worldwide royalty-free government license and the government’s march-in rights, which permit the government to require licensing to others if the title holder fails to commercialize the invention. It established a uniform federal patent policy to replace the patchwork of agency-specific rules that had made the treatment of an invention depend on which agency funded the research. The balance was deliberate: the inventor’s institution would own the patent and could license it exclusively, creating the incentive to invest in development, while the government retained the license it needed for its own purposes and the march-in authority it needed against nonuse. Congress later extended the election-of-title policy beyond universities and small businesses, including through 1984 amendments that broadened the act’s reach and removed restrictions on exclusive licensing. A further statute, the Federal Technology Transfer Act of 1986, built on Stevenson-Wydler by authorizing federal laboratories to enter cooperative research and development agreements with private firms, giving the laboratories a contractual instrument for joint work that the 1980 acts had not provided. Bayh-Dole’s reach extended directly into the Space Act: section 7(b) of the act amended the very provision, 42 U.S.C. 2457, that had codified Section 305’s government-title rule. The title provision that had made every NASA contract invention the exclusive property of the United States was itself rewritten by the statute that reversed the default.

The significance of this history for the comparison is structural, not merely historical. The agency model, taken to its full logic, assigns inventions to the state because the state paid for them. That assignment proved to be a commercial dead end so severe that Congress rewrote federal patent policy in 1980 to reverse the default. The correction preserved what the title rule got right: the government kept its license to practice the inventions it funded and its march-in authority against nonuse, so the public’s investment was not simply given away. What changed was the default about who commercializes. Under the corrected rule, the institution closest to the market holds the title and the incentive to develop the invention, while the government holds the rights it needs as the customer. The subsidy model of the CHIPS Act starts from the opposite default and never encounters the problem. Private firms that receive CHIPS incentives retain title to their inventions as a matter of course; the government’s conditions attach to the award agreement, not to the patent. The state gets the outcome it paid for, domestic fabrication capacity, without claiming ownership of the intellectual property that makes the fabs valuable. The agency model had to learn, through two statutes and a generation of stalled commercialization, that state ownership of inventions can defeat the purpose of state-funded research. The subsidy model assumes private ownership from the beginning and attaches its conditions to the money instead. The difference is the same one that runs through the entire comparison: the Space Act assumed the government was the customer, so it built the agency, bought the mission, and took the patents. The CHIPS Act assumed the government was the funder, so it left the capability, the factories, and the inventions in private hands, and asked only that the money be spent on American soil under American conditions.

The 1958 story therefore ends with a correction. The agency model assumed that state ownership of everything the mission touched, the talent, the laboratories, the contracts, and the inventions, was the price of national capability, and Congress spent the next generation deciding where that assumption held and where it broke. The 1980 statutes redrew the invention default without touching the agency, preserving the government as customer while returning commercialization to the firms best placed to do it. What follows is the other model, built for the opposite market structure. Where the Space Act assumed that no private actor could or would do spaceflight, the semiconductor division of the 2022 law assumed that private actors already made chips and would keep making them, and that the government needed only to change where. The instrument changes with the assumption: not procurement of a mission but subsidy of an input, not an agency that performs but a department that funds, not title in the state but conditions on the money. The origins of the subsidy model explain why it took that shape, beginning not with a satellite’s beep but with a shortage of chips, and not with a single design decision but with the wreckage of a bill that was too broad to pass.

The asymmetry between the two halves of this article is worth stating plainly. The agency half is a story of design and correction: a statute that built an institution, a title rule that overreached, and two later statutes that drew the default back toward private ownership while keeping the government as customer. The subsidy half is a story of selection and conditions: a collapsed bill that narrowed to the one industry where the case for action seemed undeniable, and a set of strings, guardrails, workforce terms, and upside sharing, meant to prove the money was not a giveaway. The first story ends with the state learning what not to own. The second story begins with the state deciding what to pay for and what to demand in return. The customer test is what makes these two stories one argument rather than two histories. It names the variable that selects the instrument, and it is the only claim in this article that generalizes beyond the two statutes on the table.

Origins of the 2022 Model

The subsidy model did not arrive in Washington as a finished idea. It arrived as the surviving half of a bill that failed. The story begins in the summer of 2020, when the Senate began work on what would become the Endless Frontier Act, a proposal to authorize a massive expansion of federal science spending and to create a new technology directorate inside the National Science Foundation. The idea had bipartisan parents. Senator Schumer, a Democrat, and Senator Young, a Republican, carried it together, and the framing was explicitly competitive: the United States was losing ground in advanced technology to China, and the federal government needed to respond with money, institutions, and planning. Over the next three years the bill kept growing. The Senate passed its version in June 2021 as the U.S. Innovation and Competition Act, a two thousand page package that combined the science authorizations with trade, diplomacy, and regional technology hub programs. The House passed its own version in January 2022 as the America COMPETES Act. The two chambers then had to reconcile their drafts in conference, and there the effort stalled. Conference negotiations dragged through the spring of 2022 with hundreds of provisions in dispute and the midterm elections approaching.

The Endless Frontier Act, introduced in May 2020, proposed a new Technology Directorate inside the National Science Foundation backed by $100 billion over five years. The scale was deliberate: its sponsors argued that the federal research enterprise built for the Cold War was not designed for a competition defined by commercial technology. The Senate’s version of the package passed on June 8, 2021, as the U.S. Innovation and Competition Act, by a vote of 68 to 32, a margin that reflected how broadly the diagnosis was shared. The House answered on February 4, 2022, with the America COMPETES Act, which differed from the Senate bill on trade enforcement, immigration, and climate provisions. Reconciling the two drafts required a formal conference, and the conference became a graveyard. More than a hundred provisions were in dispute, the negotiators faced a shrinking legislative calendar, and each chamber’s leadership had priorities the other chamber would not accept. By early summer 2022 the broad bill was widely described, including by its own supporters, as stuck.

While the broad competition bill was collapsing, a narrower idea was gaining independent momentum. A semiconductor shortage that began in 2020 had idled auto plants, raised consumer electronics prices, and exposed how little leading edge chip manufacturing the United States still hosted. The industry consortiums and the Commerce Department were both warning that the country made none of the most advanced logic chips domestically. In the spring of 2022, a bipartisan group began arguing that the semiconductor incentives could move on their own, stripped of the wider competition bill. The vehicle was the CHIPS Act provisions that had already been authorized in the fiscal year 2021 defense authorization bill, which created the legal authority for incentive payments but appropriated no money. The argument for separating the semiconductor half was practical: fabs take years to build, and every quarter of delay was a quarter in which investment decisions were being made in Taiwan, South Korea, and Arizona by other governments writing checks. In late July 2022, Senate leadership announced that the conference process was being abandoned and that a slimmed bill would move instead: the semiconductor incentives, funded with direct appropriations, plus a package of science authorizations. The pivot was abrupt and it worked. The Senate passed the slimmed bill on July 27, 2022, by a vote of 64 to 33. The House followed on July 28 by 243 to 187. The President signed Public Law 117-167 on August 9, 2022, less than three weeks after the conference was abandoned. The speed of the final passage is part of the story: the semiconductor half had been negotiated, in substance, years earlier, and what the summer of 2022 supplied was the decision to let it travel alone.

The name on the enacted law is the CHIPS and Science Act, and the name describes its structure. Division A is the CHIPS Act of 2022, the semiconductor division. The semiconductor provisions had a legislative head start that the rest of the competition bill lacked. The CHIPS Act itself was first enacted as part of the fiscal year 2021 defense authorization bill, which created the legal authority for the Commerce Department to make incentive payments but appropriated no money to pay them. For eighteen months the authority sat empty, a program on paper. The summer 2022 bill filled it with $52.7 billion in appropriations: $39 billion for manufacturing incentives, $13.2 billion for semiconductor research and development and workforce programs, plus a 25 percent investment tax credit for semiconductor manufacturing facilities and equipment enacted through the tax code. The defense authorization origin matters because it explains the national security framing that carried the bill. The argument that moved votes was not industrial policy in the abstract. It was the concrete fact, repeated in committee hearings and floor speeches, that the United States manufactured none of the most advanced logic chips used in its own weapons systems, and that a disruption in the Taiwan Strait would idle American industry within months. That framing is what let a subsidy program pass with bipartisan majorities in a Congress that could agree on little else. Division B is the Research and Development, Competition, and Innovation Act, which authorizes (but does not fully fund) expanded science agency budgets. The wider pattern that division repeats is traced in the US science policy legislation sequence. The two divisions embody two different theories, and the contrast matters for everything that follows. The science division is the descendant of the old model: fund research, build knowledge, let the economy absorb it. The semiconductor division is something else. It is industrial policy with cash and conditions, a direct federal payment to private firms to build physical plants on American soil. That half is the subject of this comparison, and it is worth being precise about what makes it new. The federal government has subsidized industry before, through tax credits, loan guarantees, and procurement preferences. What the semiconductor division created was a grant program of unusual size, $39 billion in manufacturing incentives and $13.2 billion for research and workforce development, paid to a concentrated set of private companies, with statutory strings attached that reach a decade into the future. The legislative history explains why the strings exist. The bill’s authors were spending public money to build private capacity, and they knew the political vulnerability of that choice. Every condition in the statute, every guardrail and clawback and reporting requirement, is an answer to the same objection: that the government was writing checks to profitable companies and getting too little in return. The origins of the 2022 model are therefore inseparable from its defensiveness. It was born from a collapsed bill, narrowed to the one industry where the case for action seemed undeniable, and loaded with conditions meant to prove the money was not a giveaway.

There is a further point about the origins that the comparison requires. The semiconductor division was not the product of one party’s ideology. The Endless Frontier Act began with bipartisan sponsors. The CHIPS Act provisions were authorized in a defense bill under a Republican president. The final vote in 2022 drew substantial Republican support in both chambers, including seventeen Republican senators. The subsidy model that resulted cannot be described as the project of a single faction. It was built by a bipartisan coalition that agreed on the diagnosis, that leading edge semiconductor manufacturing had become a strategic vulnerability, while disagreeing on almost everything else. That bipartisan parentage is what makes the neutrality rule of this article more than a courtesy. Industrial policy is genuinely contested, and the contest runs inside both parties. The question this article poses is not whether industrial policy is good or bad in general. The question is whether the subsidy model or the agency model fits the particular problem of semiconductors, and the answer turns on something more concrete than ideology.

Instrument: Capital Subsidy

The defining feature of the semiconductor division is its instrument. It buys an input, not an outcome. The statute does not ask the government to design a chip, or to operate a fab, or to hit a production target. It offers money to private firms that build, expand, or modernize semiconductor facilities in the United States, and it lets the firms decide what to make, for whom, at what price. The federal role is to reduce the cost of capital for domestic construction. The market’s role is everything else. This is the core of the subsidy model, and it deserves to be stated plainly because it is so different from the agency model it is being compared with. Under the Space Act model, the government defines the mission, funds the research and development, and in many cases operates the resulting capability itself. The technology is built to serve a government purpose, and the government is the customer. Under the semiconductor division, the government defines no mission and buys no product. It subsidizes an input, fabrication capacity on American soil, and hopes the market produces the outcome: a more resilient supply chain, a larger domestic workforce, a technology base that does not depend on a single island in the Pacific.

The mechanism for distributing the money is a competitive award process run by the Commerce Department. Firms apply. The Department evaluates applications against statutory criteria that include the economic and national security value of the project, the applicant’s financial strength, the workforce development plan, and the broader impact on the domestic semiconductor ecosystem. Awards can take several forms: direct grants, cooperative agreements, loans, and loan guarantees. The statute also created a 25 percent investment tax credit for semiconductor manufacturing equipment and facilities, a separate instrument that works through the tax code rather than through Commerce. The competitive design is deliberate. The authors wanted the Department to be able to bargain, to extract commitments from applicants in exchange for public money, and to favor projects with the largest spillover effects. Competition among applicants was supposed to discipline the process: firms that wanted the money would offer more, build more, train more, commit more.

Commerce organized the competition through a series of funding notices that divided the semiconductor world into categories. Leading edge logic and memory fabs competed in one track. Current generation and mature node facilities, the older process technologies that go into cars, appliances, and defense systems, competed in another. Materials and equipment suppliers had their own notice. The categories reflected a judgment about where the strategic gap was largest, and they let the Department weigh a leading edge fab in Arizona against a mature node expansion in Texas without pretending the two were the same kind of project. Applicants moved through a staged process: statements of interest, pre-applications, full applications, then due diligence and negotiation of award terms. Each stage gave Commerce more information and more leverage, because the terms of the award, including the conditions described in the next section, were negotiated project by project rather than set by formula.

The tax credit deserves separate attention because it works on a different theory. The 25 percent advanced manufacturing investment credit is available to any qualifying firm that builds or equips a semiconductor facility in the United States, whether or not it receives a Commerce grant. There is no competition, no application to Commerce, no negotiated conditions beyond the tax code’s own rules. It is a blunt instrument where the grants are a precision one. The two instruments interact: a firm can claim the credit on the same project that receives grant funding, which means the effective federal subsidy for a single fab can be substantially larger than the grant amount alone. The credit also reaches firms that never touch the grant program, which broadens the subsidy’s footprint while thinning its conditionality. Everything the grant program does with conditions, the tax credit does without them, and that asymmetry is worth remembering when the verdict is weighed.

A third layer of subsidy sits outside the federal statute entirely. States and localities competed to host the fabs with their own packages: tax abatements, infrastructure spending, workforce training funds, expedited permitting. The federal grants did not prohibit this stacking, and in practice the largest projects combined federal, state, and local money. The stacking complicates the attribution question further. When a fab rises in Arizona or Ohio or Texas, the federal grant is one contributor among several, and the marginal effect of the federal dollar is harder to isolate. It also complicates the politics. The conditions that Congress attached to the federal money do not automatically attach to the state money, which means the effective conditionality of the total package is diluted by the layers that carry no strings.

The construction timelines impose a discipline of their own, and they require caution about attribution. A leading edge fab takes three to five years from groundbreaking to volume production. The first major awards under the program were announced in 2024, which means the facilities they fund will not be producing chips at scale until late in the decade at the earliest. Any claim about what the program has achieved must be read against that clock. Announcements are not output. Groundbreakings are not output. Output is wafers, and wafers are years away. This is not a criticism of the program. It is a property of the instrument. Capital subsidy operates on the timescale of construction, and construction is slow. The caution cuts both ways. Critics who declare the program a failure before the fabs are built are making the same error as supporters who declare it a success at the ribbon cutting. The honest position is that the money has been committed, the conditions have been accepted, and the results are not yet measurable.

There is a second caution about the instrument, and it concerns the counterfactual. The semiconductor division pays firms to build in the United States. Some of those firms were already planning American expansions before the statute existed. The question of how much domestic investment the subsidies actually caused, as opposed to how much they subsidized investment that would have happened anyway, is genuinely hard to answer. Economists call this the problem of additionality, and it is the permanent vulnerability of every subsidy program. The statute tries to address it by requiring applicants to demonstrate that the federal money is necessary for the project to proceed as proposed, but necessity is easy to assert and hard to verify. A firm with a global capital budget can always say that the American project was marginal without the grant. The Commerce Department can demand financial models and stress test them, but it cannot observe the parallel universe in which no subsidy existed. This does not make the program worthless. It makes its benefits uncertain in a specific and irreducible way. The agency model has its own version of this problem, the question of whether NASA’s contractors would have developed the same technologies without the space program, but the agency model at least produces a government capability whose existence is not in doubt. The subsidy model’s product is a change in private investment behavior, and changes in behavior are harder to see.

The deeper point about the instrument is philosophical. Capital subsidy is an admission that the government does not know how to produce the outcome it wants. If the federal government knew how to manufacture leading edge semiconductors, it would not need to pay TSMC or Intel or Samsung to do it. It would build the fabs itself, the way it built the interstate highways or the way NASA built the Saturn V. The subsidy model concedes the government’s ignorance and rents the capability instead. There is nothing shameful in that concession. Governments are often bad at commercial manufacturing, and the history of state owned chip fabs is not encouraging. But the concession sets the terms of the comparison. The subsidy model trades control for speed and for access to private competence. The agency model trades speed for control and for the creation of public capability. Neither trade is free. The question is which trade the problem requires.

The speed side of that trade deserves one more note, because it is the subsidy model’s strongest practical argument. Building a government agency to manufacture semiconductors would have taken a decade before the first wafer, assuming the government could recruit the talent at civil service salaries and manage a commercial operation through appropriations cycles. The subsidy model reached term sheets with private firms within two years of enactment, because the firms, the engineers, and the supply chains already existed. The government did not have to build competence. It had to redirect it. That redirection is faster than creation by an order of magnitude, and in a competition measured in product cycles, speed is not a luxury. The cost of that speed is everything this section has described: the uncertainty about additionality, the thin enforcement record, the dependence on firms whose interests only partly overlap with the public’s. The agency model pays the opposite cost. It builds slowly and owns what it builds. For the Moon, slowness was affordable and ownership was the point. For chips, slowness was the risk and ownership was never on offer.

Conditions on the Subsidy

If the instrument is the money, the conditions are the leverage, and the conditions are where the 2022 model is genuinely newer than anything that came before it. The statute does not simply hand out grants. It attaches a set of statutory strings that bind recipients for years after the check clears, and the strings are designed to solve three distinct problems: the geopolitical problem, the labor problem, and the political problem of the giveaway.

The geopolitical conditions are the guardrails. Any recipient of manufacturing incentives is barred, for ten years after receiving the award, from materially expanding semiconductor manufacturing capacity for advanced chips in designated countries of concern. The statute names the mechanism precisely: a recipient that undertakes a prohibited expansion must repay the full amount of the federal assistance. This is the clawback, and it is the teeth of the guardrail. The bar covers new construction and capacity expansion for leading edge and advanced semiconductors, with defined thresholds for what counts as advanced and with exceptions for legacy chips that serve established commercial markets. The Commerce Department was given authority to define the details, including which countries count as countries of concern and what transactions trigger the clawback. The logic of the guardrail is straightforward. The United States was paying firms to build in America because it feared dependence on foreign supply chains. It would be perverse to let those same firms take the money and then expand their most advanced capacity in the country whose leverage the program was meant to reduce. The ten year horizon is the notable feature. Most federal grant conditions expire when the grant period ends. These follow the recipient for a decade, which makes them less like a contract term and more like a regulatory commitment. Commerce gave the guardrails their operative detail through rulemaking, defining what counts as a material expansion, setting the process node thresholds that separate advanced semiconductors from legacy ones, and specifying the carve outs for established commercial uses. The legacy exception is the pressure valve: a recipient may continue producing older generation chips in a country of concern where those chips serve ordinary commercial markets, because the statute’s target was the frontier, not the entire industry. The rulemaking drew the line between the two, and the line is where future disputes will be fought. Every threshold in a regulation is an invitation to engineer around it, and the firms subject to these rules employ some of the best engineers in the world. The guardrail’s strength will be tested not by open defiance but by close cases: capacity additions that sit just below the threshold, technology transfers that stay just outside the definition, expansions characterized as maintenance rather than growth.

The workforce provisions address the labor problem. Semiconductor manufacturing requires large numbers of skilled technicians, engineers, and construction workers, and the domestic pipeline for those workers was thin. The statute requires applicants to submit workforce development plans describing how they will recruit, train, and retain workers, including partnerships with educational institutions and strategies for reaching economically disadvantaged individuals. For the largest awards, those exceeding $150 million in direct funding, the requirements go further: applicants must provide a plan for affordable, accessible, reliable childcare for construction and facility workers. The childcare provision attracted the most public attention of any workforce term, and it illustrated the theory behind the conditions. The government was not just buying fabs. It was buying a labor market capable of staffing them, and it was using the grant as leverage to force planning that the market might not do on its own. Whether the provision was wise is contested. Supporters called it a practical recognition that construction workers with children need childcare to show up. Critics called it social policy smuggled into an industrial bill. Both descriptions can be true at once.

The workforce terms go beyond childcare. Applicants were expected to show concrete partnerships with community colleges, technical schools, and labor organizations, commitments to registered apprenticeship programs, and plans for recruiting from groups historically underrepresented in semiconductor manufacturing. The theory was the same across all the provisions: the grant buys not just a building but the human infrastructure to run it, and the application process forces firms to plan for labor supply with a seriousness the market had not demanded. Skeptics raised two objections. The first was competence: a semiconductor firm knows how to build fabs, not how to run a childcare or apprenticeship system, and forcing it to do both risks doing neither well. The second was mission creep: every additional requirement narrows the pool of willing applicants and slows the disbursement of money that was supposed to move quickly. The statute’s answer to both objections was the $150 million threshold, which concentrated the heaviest requirements on the largest awards. Whether that calibration was right is one of the judgments the program’s history will eventually render.

The upside sharing provisions address the political problem of the giveaway. For awards above $150 million, recipients must agree to share with the government a portion of any profits that exceed the projections in their application by an agreed threshold. The mechanism is sometimes described as a clawback, but it is more accurately a profit sharing agreement: if the subsidized project turns out to be far more lucrative than the applicant claimed when asking for public money, the public gets some of the upside. The provision exists because of the additionality problem described earlier. If a firm lowballs its expected returns to make the subsidy look necessary, and then earns windfall profits, the profit sharing term recaptures part of the transfer. It is an elegant idea on paper. Its enforcement record is, by the nature of the instrument, thin. The fabs are not yet producing. No profits have been shared. The Commerce Department has not yet had to determine whether a recipient’s returns exceeded the threshold, or to collect. The same is true of the guardrails. No recipient has yet tested the ten year bar in a way that triggered a clawback determination. The conditions are therefore best understood as a newer form of leverage with an unproven enforcement record. They represent a genuine innovation in how the federal government writes industrial subsidies, attaching behavioral strings that outlast the grant and reach into corporate strategy. But innovation is not the same as effectiveness, and effectiveness cannot be judged until the strings are pulled.

One more feature of the conditions deserves emphasis: they are negotiated, not formulaic. Two recipients can receive similar grant amounts and face different specific obligations, because the award terms are hammered out in due diligence between Commerce and each applicant. The statute sets the floors, the guardrails, the workforce plans, the upside sharing for large awards, but the precise commitments, the milestones, the reporting cadence, the remedies short of clawback, are bespoke. This gives the Department flexibility to fit the conditions to the project, and it gives critics a transparency problem, because the public sees the announced award amount but not the full text of the obligations behind it. Bespoke terms are harder to evaluate in the aggregate than uniform rules. They are also harder to game, because each negotiation is a fresh contest. Which of those effects dominates will depend on the quality of the negotiators, a variable no statute can fix.

There is a structural reason the enforcement record is thin, and it is worth naming. The conditions bind the recipient, but the recipient’s compliance is largely self reported. The Commerce Department can audit, can demand records, can investigate suspected violations, but it is a small agency administering a program that touches some of the largest and most sophisticated corporations in the world. The information asymmetry is severe. A multinational semiconductor firm knows more about its global capacity planning than any federal auditor can learn from the outside. The statute tries to compensate with severe penalties, full repayment for guardrail violations, but severe penalties only deter if detection is credible. This is not an argument that the conditions will fail. It is an argument that their success depends on an enforcement capacity that has not yet been tested. The agency model, by contrast, enforces its terms through the procurement relationship itself: the government is the customer, it inspects what it buys, and payment follows performance. The subsidy model’s enforcement is necessarily more indirect, because the government is not buying anything. It is paying for behavior, and behavior is harder to verify than delivery.

Invention Ownership Contrast

The two models differ most sharply, and most consequentially, on a question that rarely makes the headlines: who owns the inventions made with the money. The Space Act took an aggressive position. Section 305 of Public Law 85-568 provided that inventions made in the performance of NASA contracts would be the exclusive property of the United States, unless NASA waived the government’s rights. The presumption ran toward public ownership. The theory was that the taxpayer had paid for the research, the research served a public mission, and the resulting intellectual property should belong to the public. NASA could and did grant waivers, particularly where private ownership would better serve commercialization, but the default was government title. The waiver practice is worth pausing over, because it shows the agency model was never as rigid as its statute looked. NASA developed an elaborate internal process for deciding when to release title, weighing the contractor’s investment, the commercial potential of the invention, and the government’s need to use it. Over time the waivers became common enough that the default mattered less than the exceptions. But the default still shaped bargaining: the contractor had to ask, the government could say no, and the starting point of every negotiation was public ownership.

The 1980 statutes reversed the starting point. The Bayh-Dole Act, Public Law 96-517, let universities, small businesses, and nonprofit organizations elect to keep title to inventions made with federal funding, subject to a government license for federal purposes and to march in rights the government could exercise if the owner failed to commercialize. The march in rights were the safety valve, and they have never been exercised, a fact that both sides of the technology transfer debate cite. Supporters of Bayh-Dole read the nonuse as proof the system works: owners commercialize, so the government never needs to intervene. Critics read it as proof the valve is decorative. Either way, the statute moved the default from public to private ownership and the commercialization numbers that followed, the surge in university patenting and licensing after 1980, are the evidence its defenders invoke. The Stevenson-Wydler Technology Innovation Act, Public Law 96-480, attacked the problem from the laboratory side, requiring federal labs to manage technology transfer actively and, through later amendments, authorizing cooperative research and development agreements that let labs work directly with private firms. Together the two laws were Congress’s verdict on the Space Act default: government ownership had protected the public’s rights and starved the public’s benefit.

The subsidy model takes the opposite default, and it does so without much fanfare. The semiconductor division contains no equivalent of Section 305. Recipients of CHIPS incentives retain ownership of the inventions, processes, and trade secrets developed in their subsidized fabs. The statute’s research and development provisions contemplate collaboration, including the National Semiconductor Technology Center, but they do not transfer title to the government. The money buys capacity, not intellectual property. This is consistent with the instrument: if the government is subsidizing an input and hoping the market produces the outcome, it has no claim on the knowledge generated along the way. The knowledge stays where the capability stays, inside the private firm.

The contrast is illuminated by the two statutes Congress passed in 1980 to fix what it saw as the defect of the agency model. The Bayh-Dole Act and the Stevenson-Wydler Technology Innovation Act both addressed the same problem: the government owned a vast portfolio of patents generated by federal funding, and almost none of them were being commercialized. Bayh-Dole allowed universities, small businesses, and nonprofit contractors to elect to retain title to inventions made with federal money, on the theory that the inventing institution would do a better job of bringing the technology to market than a federal patent office would. Stevenson-Wydler pushed federal laboratories to transfer technology to the private sector through cooperative agreements. Together they represented a judgment that the Space Act’s default, government ownership of federally funded invention, had produced a transfer problem: public capability that never became private product. The subsidy model avoids the transfer problem entirely by never taking title in the first place. There is nothing to transfer because the government never owned it.

This is a real advantage, and it should be counted as one. But it comes with a corresponding cost that the verdict section must weigh. When the government owns the invention, it can direct its use, license it broadly, and ensure that the public investment serves public purposes beyond the original mission. GPS, developed for military navigation, became a civilian utility in part because the government controlled the signal. When the private firm owns the invention, the public’s return is limited to whatever the conditions extract: the guardrails, the workforce terms, the profit sharing. The subsidy model is cleaner on ownership and thinner on public rights. The agency model is messier on ownership and thicker on public rights. The 1980 statutes were Congress’s attempt to get the best of both, private ownership to drive commercialization, with government march in rights retained for public purposes. The semiconductor division did not need that compromise because it never took ownership at all. Whether that was wisdom or abdication depends on what the public needed from the investment, which returns to the customer question.

The research side of the semiconductor division complicates the ownership picture slightly, and honesty requires noting it. Alongside the manufacturing incentives, the statute funded collaborative research institutions, including a National Semiconductor Technology Center meant to bring firms, universities, and government researchers together on precompetitive problems. In a collaborative research setting, invention ownership is governed by the agreements among the participants rather than by a statutory default, and the government’s leverage over those agreements comes from its funding share rather than from a Space Act style title clause. The result is a hybrid: private ownership as the norm, with negotiated intellectual property terms at the margins. The hybrid is consistent with the post 1980 consensus that private title commercializes better than public title. It is also consistent with the subsidy model’s deeper logic. The government is paying to accelerate a private industry, not to acquire its knowledge.

The Verdict

The comparison needs its deciding factor, and the brief names it: whether the government is the customer for the technology in question. This is the customer test, and it cuts through the ideological fog around industrial policy with a single structural question. When the government is the only meaningful buyer of a technology, procurement drives development directly. The government specifies what it wants, pays for it, and the technology comes into existence because the customer demanded it. When the government is one buyer among many in a global commercial market, procurement provides no leverage at all. The government can buy all the chips it wants and the industry will barely notice, because government purchases are a rounding error next to consumer electronics, automobiles, and data centers. In that situation, subsidy with conditions is the only available instrument. The government cannot command the market as a customer, so it must influence the market as a subsidizer.

Apply the test to the two cases. In 1958, the government was the customer. There was no commercial market for human spaceflight, no private buyer for a Saturn V, no consumer demand for lunar orbit rendezvous. The only entity that wanted to go to the Moon was the United States government, acting for reasons of prestige, science, and Cold War competition. Procurement was therefore the natural instrument. NASA could define the mission, fund the development, and create the capability, because the capability had no other reason to exist. The agency model fit the market structure perfectly: a single customer, a defined mission, a technology with no commercial alternative. The enormous capability that resulted, the industrial base, the systems engineering discipline, the materials science, was a product of that fit.

In 2022, the government was not the customer. Semiconductors are the most commercial technology in human history. The global market runs to hundreds of billions of dollars a year, driven by smartphones, cars, servers, and consumer devices. The federal government’s own chip purchases are a small fraction of that total. No procurement program, however large, could move the industry’s investment decisions. The government could not buy its way to domestic fabs because it was not a big enough buyer to matter. Subsidy was therefore the only instrument that could work. The government had to pay firms to change their behavior, because it could not command their behavior as a customer. The semiconductor division’s design, cash for construction plus conditions on strategy, follows directly from that structural fact.

The history of the integrated circuit itself confirms the test, because the customer relationship changed over time and the effective instrument changed with it. In the 1960s, the federal government, through the Defense Department and NASA, was the dominant buyer of integrated circuits, purchasing the large majority of early production for missiles, spacecraft, and military systems. Procurement drove the industry directly: government demand funded the learning curve, pushed reliability standards, and pulled unit costs down until commercial markets could take over. The agency and procurement model worked because the government was the customer. By the 1980s the commercial market had dwarfed federal demand, and procurement lost its leverage. The government could no longer steer the semiconductor industry by buying chips, because its purchases were a shrinking share of the total. The 2022 model is the response to that changed structure. When the customer relationship that made procurement powerful no longer exists, the government must find another instrument or accept irrelevance. Subsidy with conditions is the instrument it found.

The verdict is therefore not ideological but structural. The agency model is the right instrument when the government is the customer. The subsidy model is the right instrument when the government is not. Neither model is superior in general. Each is superior for the market structure it fits. This is the defended verdict the brief demands, and the deciding factor is named: whether the government is the buyer. A reader who can state the two models, identify which statute exemplifies each, and apply the customer test to a new case has understood the article.

The counter-reading must be addressed, because it is tempting and partially true. The counter-reading says that one model is simply more effective than the other, and that the comparison is really a contest. The evidence for the agency model’s effectiveness is formidable. NASA’s procurement machine produced the Apollo program, the Space Shuttle, the deep space probes, and a cascade of spinoff technologies that reshaped the civilian economy. No subsidy program has ever produced anything comparable. But the agency model’s effectiveness came with the transfer problem that the two 1980 statutes were written to solve. Government ownership of inventions, government operation of capabilities, government direction of research: all of it produced technology that sat on federal shelves, patented but unlicensed, developed but uncommercialized. Bayh-Dole and Stevenson-Wydler were admissions that the agency model, for all its power to create capability, was weak at diffusing it. The subsidy model avoids that problem by design. It never takes ownership, so there is nothing to transfer. The capability is born inside the firm that will commercialize it.

But the subsidy model’s avoidance of the transfer problem creates problems of its own, and intellectual honesty requires naming them. The first is the authorization problem. The agency model builds permanent institutions. NASA still exists. Its capabilities compound over decades. The subsidy model spends a fixed appropriation over a fixed period, and when the money is gone the leverage is gone. The ten year guardrails outlast the grants, but the grants themselves are one time. A future Congress must decide whether to renew the program, and renewal will be judged on results that are still years away. The second is the enforcement problem described earlier. The conditions are only as strong as the Commerce Department’s ability to monitor some of the world’s most sophisticated firms, and that ability is untested. The third is the capture problem. A competitive award process that distributes $39 billion to a handful of firms will attract lobbying, political pressure, and the eternal suspicion that the awards follow influence rather than merit. The agency model has its own capture dynamics, the congressional politics of NASA centers is a case study, but procurement at least ties payment to delivery. Subsidy ties payment to promises.

There is a fourth problem, and it is the mirror image of the agency model’s transfer problem. The subsidy model keeps capability in private firms, which solves commercialization, but it also keeps the knowledge needed to evaluate the program inside those same firms. The government that subsidizes without owning cannot easily audit what it bought. It depends on the recipient’s data, the recipient’s accounting, the recipient’s characterization of its own global strategy. The information asymmetry is not a flaw in implementation. It is built into the instrument. Every subsidy program faces it, and the usual answer is some combination of auditing, penalties, and transparency requirements. The semiconductor division has all three on paper. Whether they suffice in practice is the experiment the program is running.

None of this refutes the verdict. It qualifies it, which is what a defended verdict should survive.

Weigh the two sets of problems against the customer test and the verdict holds. The agency model’s problems, the transfer problem, the institutional rigidity, the political geography of its centers, are the problems of a model that works when the government is the customer. The subsidy model’s problems, additionality, enforcement, capture, impermanence, are the problems of a model that works when the government is not. The choice between them is not a choice between effectiveness and ineffectiveness. It is a choice dictated by market structure. Where the government buys, it should build. Where the government merely hopes, it must subsidize, and it should attach every condition it can enforce. The 2022 model understood this. Its authors could not make the government the customer for semiconductors, so they made it the subsidizer, and they wrote the conditions as tightly as the politics allowed. Whether the enforcement will match the drafting is the open question, and it will be answered not in Washington but in the fabs, in the years ahead, when the wafers start coming out and the auditors start asking what the money bought.

The table that follows compresses the comparison to its working parts. It is meant to be read as a reference, not as a substitute for the argument. Each axis names the dimension along which the models differ, and each cell names the design choice the statute made. The row that matters most is the last, because the market structure selects everything above it. A reader who can reproduce this table from memory, and who can say which statute exemplifies each column, has absorbed the article’s machinery. The verdict then supplies the principle that puts the machinery to work.

The Two-Model Comparison Table

Axis	1958 agency model	2022 subsidy model
Capability location	The government holds the capability: the agency employs the scientists and engineers, owns the laboratories and test facilities, and directs the mission.	Private firms hold the capability: the firms build and operate the fabs, hire the workforce, and manage supply chains; the government holds a checkbook.
Instrument	Mission procurement: the government defines the mission, funds development, buys hardware and services under procurement law, and takes delivery.	Capital subsidy: the government offers grants, loans, and a tax credit to shift private investment decisions; it takes no delivery and defines no mission.
Invention ownership	Government title by default under Section 305, with waivers available; the default was reversed toward private title by the 1980 Bayh-Dole and Stevenson-Wydler statutes.	Private firms retain title to inventions, processes, and trade secrets; the government’s conditions attach to the award agreement, not to the patent.
Conditions and enforcement	Contract terms enforced through procurement law: specifications, inspection rights, and termination for default or convenience; payment follows performance.	Statutory grant conditions: ten-year guardrails on foreign expansion with full repayment, workforce plans, and upside sharing; negotiated per award, with a thin enforcement record.
Market structure where each performs	The government is the only meaningful customer and no commercial market exists, as with human spaceflight in 1958.	The government is one buyer among many in a global commercial market and procurement has no leverage, as with semiconductors.

Each stop on that path teaches something the comparison alone cannot. The 1958 profile shows the inheritance that made the agency model possible, the NACA’s laboratories and engineers, and with it the lesson that state capability must be inherited or built over a generation; it cannot be summoned by appropriation alone. The Bayh-Dole profile shows how a single default rule about invention ownership can starve commercialization for a generation, and how Congress corrected it without surrendering the government’s rights. The CHIPS treatment shows a legislature writing conditions as tightly as the politics allowed, with an enforcement record that remains the experiment’s open question. The survey shows that authorizing statutes routinely promise funding trajectories that appropriators do not deliver, which is the context in which the 2022 law’s science division must be read. None of these lessons contradicts the customer test. Each of them fills in the machinery that the test selects.

Where This Comparison Fits in the Study Path

This article carries the cluster’s study-path recommendation: read the comparison first, then the anchor statutes, then the survey. The reason for that order is the customer test. A reader who meets the deciding factor early, that procurement fits when the government buys and subsidy fits when it does not, can use it as a lens for everything that follows instead of discovering it at the end. Read this article to fix the two models and the verdict in mind. Then read the 1958 Space Act profile for the agency model at full strength, including the invention title rule that the model carried to its logical conclusion. Then read the Bayh-Dole profile for the statute that reversed that default and for the transfer problem it was solving. Then read the CHIPS and Science Act treatment for the subsidy model in full, with the appropriations, the guardrails, and the unproven enforcement record examined statute by statute. End with the science policy survey, which shows the authorization-without-appropriation pattern that the 2022 law’s science division repeats and which places both models inside seventy years of federal science policy.

The habit to build along the way is the customer test itself. For each statute, ask who takes delivery: the government, a firm, or no one at all. The Space Act’s answer is the government, and the agency, procurement, and title rules all follow. The semiconductor division’s answer is no one at all, and the grant, the conditions, and the private title all follow. The test works beyond the cluster. Apply it to weapons systems, where the government buys and procurement still dominates, and to clean energy, where the government mostly subsidizes, and the statutes in both fields resolve into the same two instruments. That portability is why the comparison earns its place early in the study path rather than late: it is a tool for reading every technology statute that follows.

Two companion tools support that work. The legislation study notebook gives readers a structured place to record each statute’s model, instrument, and customer test result as they move through the cluster. The civics study tool tests the concepts the comparison introduces, from appropriations to agency design. Work through the cluster in the order above, notebook open, and the full sequence is mapped in the US legislation study guide.

Frequently Asked Questions

Q: What is the difference between the Space Act model and the 2022 chips law?

The Space Act model builds state capability: the statute created NASA, hired the engineers, owned the laboratories, bought missions through procurement contracts, and originally took title to the inventions its contracts produced. The government was the customer, defining the requirements and taking delivery of the mission. The semiconductor division of the 2022 law does the opposite. It leaves capability in private firms and pays them to build fabrication plants in the United States. The Commerce Department awards grants, loans, and guarantees with statutory conditions attached, but it takes no delivery of chips and claims no title to the inventions. The difference runs through four axes: where capability sits, which instrument the government uses, who owns the inventions, and how conditions are enforced. The models fit different market structures. The agency model fits when the government is the only buyer, and the subsidy model fits when the government is one buyer among many in a commercial market.

Q: Should the government build the capability or subsidize firms like the Space Act did?

The comparison’s verdict is structural, not ideological: build when the government is the customer, subsidize when it is not. In 1958 no commercial market for human spaceflight existed, so the government created the agency, employed the engineers, and procured the mission. No subsidy could have summoned a lunar program from firms that had no capability to redirect. In 2022 the semiconductor industry was mature, global, and commercial, and the government’s own chip purchases were a rounding error next to the total market, so procurement could not move investment decisions and subsidy was the available instrument. The deciding factor is whether the government can reach the technology as a buyer. Where it buys the product, as with launch vehicles, procurement drives development and the agency model works. Where it merely wants a national outcome, such as domestic fabrication capacity, subsidy with enforceable conditions is the instrument that fits.

Q: Why did the Space Act create an agency instead of grants?

Congress and President Eisenhower judged that spaceflight in 1958 was not a market that needed incentives. It was a national undertaking for which no market existed, no customer besides the government existed, and no private firm could finance the capability. The institutional answer inherited the National Advisory Committee for Aeronautics, with its laboratories and roughly 8,000 employees, giving the new agency a functioning workforce on its first day. A grant program would have had no recipients capable of doing the work, because the knowledge, the facilities, and the engineers did not exist in private hands at the required scale. The civilian agency design also served strategy. A civilian program could conduct science openly, publish its results, and cooperate internationally without feeding the arms race dynamic that a military program would have fed. The statute therefore created an organization that would do spaceflight rather than offering rewards to private actors for doing it.

Q: Does the government own inventions differently under the Space Act than under chips subsidies?

Yes, and the contrast is the sharpest in the comparison. Section 305 of Public Law 85-568 provided that inventions made under NASA contracts were the exclusive property of the United States unless the Administrator waived the government’s rights, a title-taking rule that presumed public ownership because the taxpayer had paid for the research. The semiconductor division contains no equivalent provision. Recipients of the incentives retain title to the inventions, processes, and trade secrets developed in their subsidized fabs, and the government’s conditions attach to the award agreement rather than to the patent. The agency model had to learn the cost of its default. Thousands of federally held patents sat unlicensed in government files, and Congress reversed the default with the Bayh-Dole Act of 1980 while keeping a government license and march-in rights for public purposes. The subsidy model never takes title, so it never faces a transfer problem, but its public return is limited to whatever the conditions extract.

Q: Which model works when the government is the only customer, the Space Act approach?

The agency model. When the government is the only meaningful buyer, procurement drives development directly. The government specifies what it wants, pays for it, and the technology comes into existence because the customer demanded it. That was the market structure of human spaceflight in 1958, and it is the structure of weapons systems generally. The history of the integrated circuit confirms the test. In the 1960s the federal government bought the large majority of early integrated circuits for missiles, spacecraft, and military systems, and that procurement funded the learning curve until commercial markets could take over. Subsidy cannot replace procurement in this situation because there is no private investment decision to change. Paying firms to do what only the government wants is procurement wearing a subsidy costume, without the inspection rights and delivery discipline that make procurement work. The customer test is the comparison’s deciding factor precisely because it sorts these cases cleanly: where the government buys, build the capability and buy the mission.

Q: Did the Space Act deliver technology faster than subsidy programs?

The comparison does not support a speed ranking between the models, because speed depends on the starting point and the timescale of the instrument. The agency model moved fast in 1958 because it inherited the NACA’s workforce and laboratories. It did not have to create competence, only to redirect it, which is why NASA could announce Project Mercury within a week of opening for business. The subsidy model also moves fast to commitment, reaching negotiated awards within about two years of enactment, because the firms, the engineers, and the supply chains already exist. The honest difference is what each instrument’s clock measures. Mission procurement produces missions on appropriations time. Capital subsidy produces fabs on construction time, and a leading edge fab takes three to five years from groundbreaking to volume production, so the funded facilities will not produce at scale until late in the decade. Announcements are not output and ribbon cuttings are not output. The article’s claim is fit, not speed: each model suits its market structure.

Q: Are conditions on chips subsidies like Space Act contract terms?

They are the analog, but the enforcement machinery differs. Under the Space Act model the government procures a mission, and its conditions arrive as contract terms: technical specifications, delivery schedules, inspection rights, testing requirements, data rights, and termination clauses, all enforceable through procurement law against a deliverable the government will take into its own hands. Under the subsidy model the conditions attach to grants. Recipients face ten-year guardrails against advanced capacity expansion in countries of concern with full repayment as the penalty, restrictions on stock buybacks and dividends, workforce and childcare plans for the largest awards, and upside sharing when profits exceed projections. These conditions are a genuinely newer form of leverage, binding corporate strategy for a decade, but their enforcement record is unproven because the fabs are not yet producing and no violation has been tested. Procurement enforces by inspecting delivery. Subsidy enforces by auditing behavior, and behavior is harder to verify than delivery.

Q: Which should a student study first, the Space Act or chips subsidies?

This comparison first, then the Space Act, then the subsidy statute. The comparison is the cluster’s organizing frame. The customer test sorts every later statute into the model it fits, so meeting the deciding factor early lets a student use it as a lens rather than discovering it at the end. After the comparison, study the 1958 statute for the agency model at full strength, including the invention title rule and the 1980 statutes that corrected it, because the transfer problem teaches what the model costs. Then study the semiconductor division for the subsidy model, with attention to the guardrails, the upside sharing, and the enforcement questions the program has not yet answered. End with the wider science policy sequence, which shows the authorization patterns both statutes belong to and places the two models inside seventy years of federal science policy. This order builds from the tool to the examples to the history, and each step is richer for the ones before it.

Q: What are the guardrails in the semiconductor incentives and how long do they last?

The guardrails bar recipients of manufacturing incentives from materially expanding advanced semiconductor manufacturing capacity in designated countries of concern for ten years after receiving the award, with full repayment of the federal assistance as the penalty for a violation. They cover new construction and capacity expansion for leading edge and advanced chips, with defined process node thresholds separating advanced semiconductors from legacy ones, and with carve outs for legacy chips that serve established commercial markets. The ten-year horizon is the notable feature. Most federal grant conditions expire when the grant period ends, while these follow the recipient for a decade, making them closer to a regulatory commitment than a contract term. The Commerce Department defined the operative details through rulemaking, including what counts as a material expansion and which countries count as countries of concern. The guardrails’ strength has not been tested. No recipient has triggered a clawback determination, so their deterrent power remains theoretical until a close case reaches the department.

Q: Can the government take back incentive money after a fab is built?

The statute gives the Commerce Department clawback authority, but recovery depends on the violation and the negotiated award terms. The strongest clawback attaches to the guardrails. A recipient that undertakes a prohibited foreign expansion must repay the full amount of the federal assistance. For awards above 150 million dollars, upside sharing lets the government capture part of profits that exceed the projections in the application, which functions as a profit-sharing term rather than a penalty. Beyond those, each award agreement carries compliance schedules, reporting requirements, audit rights, and remedies short of revocation. The practical limit is that money spent on concrete and steel cannot be unspent. Revoking an award punishes the recipient but does not produce the fabrication capacity the program wanted, which leaves the underlying outcome unbuilt. The agency model differs at this point. Terminating a procurement contract ends the government’s own demand, while taking back a subsidy ends the leverage without creating the outcome the money was meant to buy.

Q: Why does the semiconductor tax credit carry fewer conditions than the grants?

The 25 percent advanced manufacturing investment credit works through the tax code rather than through the Commerce Department, so it is available to any qualifying firm that builds or equips a semiconductor facility in the United States, whether or not it receives a grant. There is no competition, no application to Commerce, and no negotiated conditions beyond the tax code’s own rules. It is a blunt instrument where the grants are a precision one. The design reflects different theories. The grants buy leverage over specific projects through bargaining, while the credit broadens the subsidy’s footprint to firms that never touch the grant program, at the price of thinner conditionality. The two instruments interact, since a firm can claim the credit on the same project that receives grant funding, which makes the effective federal subsidy substantially larger than the grant amount alone. That asymmetry matters when weighing the verdict. The conditions this article analyzes attach to the grants, not to every federal dollar supporting the fabs.

Q: What is additionality and why does it challenge the subsidy case?

Additionality is the question of how much domestic investment the subsidies actually caused, as opposed to how much they subsidized investment that would have happened anyway. Some recipient firms were already planning American expansions before the statute passed, and a firm with a global capital budget can always assert that the American project was marginal without the grant. The statute requires applicants to demonstrate that the federal money is necessary for the project to proceed as proposed, but necessity is easy to assert and hard to verify, because the Commerce Department cannot observe the parallel universe in which no subsidy existed. Money is fungible, and attributing any particular investment decision to the subsidy requires a counterfactual no audit can fully establish. This does not make the program worthless, but it makes its benefits uncertain in a specific and irreducible way. The agency model’s product is a government capability whose existence is not in doubt. The subsidy model’s product is a change in private behavior, and changes in behavior are harder to see.

Q: How did state ownership of Space Act inventions slow commercialization?

The Section 305 default assigned every invention made under a NASA contract to the government as the exclusive property of the United States, with waivers available through the Administrator. The default signaled to contractors and universities that inventions made under NASA contracts belonged to the government, so firms had little incentive to invest in developing products built on patents they did not own and could not license exclusively. NASA’s waiver practice was described as liberal and expeditious, but the waiver was an escape hatch rather than a design feature. It required the request, the review, and the delay, and each step taxed the commercialization decision at the margin. Defaults govern behavior more than exceptions do. By the late 1970s the orphan portfolio was notorious, with roughly 28,000 to 30,000 federally held patents and only a small fraction ever licensed to anyone who might bring the inventions to market. Congress concluded that the title rule was defeating the purpose of the research it was meant to protect and rewrote the default in 1980.

Q: How did the 1980 statutes fix the Space Act invention default?

Congress acted twice in the same year. The Bayh-Dole Act, Public Law 96-517, allowed universities and small businesses to elect to retain title to inventions made with federal funding, subject to a worldwide royalty-free government license and the government’s march-in rights against nonuse. It replaced the patchwork of agency-specific rules with a uniform federal patent policy, so the treatment of an invention no longer depended on which agency funded the research. Congress later extended the election-of-title policy beyond universities and small businesses, including through 1984 amendments. Section 7(b) of Bayh-Dole directly amended the codification of the Space Act’s Section 305 title rule, rewriting the provision itself. The Stevenson-Wydler Technology Innovation Act, Public Law 96-480, attacked the problem from the laboratory side, declaring technology transfer a government function and requiring federal laboratories to manage it actively, with the 1986 amendments adding cooperative research and development agreements. The balance was deliberate. The institution closest to the market owns the patent and the incentive to develop it, while the government keeps the license and march-in authority it needs as customer.

Q: Which model handles program failure better, the agency or the subsidy?

Each fails in its instrument’s characteristic way. The agency model handles technical catastrophe institutionally. When the Apollo 1 fire killed three astronauts in January 1967, NASA investigated with its own engineers, redesigned the command module’s systems, and returned to flight, because the technical judgment needed for both the investigation and the redesign lived inside the agency. A funder without capability cannot do that. It can commission a report, withhold the next tranche of money, or terminate the award, but it cannot reach into the design and fix it. The subsidy model fails differently. Its failure modes are windfall, subsidy capture, and compliance theater, in which conditions are technically satisfied while the underlying investment decision changes little. Enforcement through revocation punishes the recipient without producing the fabrication capacity. The agency model’s failures are expensive and visible, with the customer paying for its own changing requirements. The subsidy model’s failures are quiet, with conditions satisfied on paper and behavior unchanged in substance.

Q: Why was the civilian character of NASA central to the 1958 design?

The military-versus-civilian question was the central design decision, and Congress considered a military space agency seriously. The Army had von Braun and the only American team with a satellite in orbit, and the Air Force had the missile programs that would supply the heavy launch vehicles. A military program would have been faster to stand up and easier to fund through defense appropriations. The civilian choice won because the administration and the President’s Science Advisory Committee judged that a military program would feed the arms race dynamic the nation was trying to manage, while a civilian program could conduct science openly, publish its results, and cooperate internationally. The statute encoded that judgment in its declaration that the nation’s space activities should be devoted to peaceful purposes for the benefit of all humankind and in its requirement that the agency disseminate information as widely as practicable. The NACA’s civilian character and technical reputation made it an acceptable vehicle, and its laboratories and wind tunnels were already the facilities any space program would need. Openness was not decoration. It was the civilian model’s operating theory and a strategic calculation.

Q: Could the government have built a federal chip fab instead of subsidizing firms?

The comparison treats this as the agency model applied to semiconductors, and the market structure answers against it. Building a government agency to manufacture semiconductors would have taken a decade before the first wafer, assuming the government could recruit leading engineers at civil service salaries and manage a commercial operation through appropriations cycles. More fundamentally, a government-run fab would compete with the private fabs the policy is trying to encourage, and the state’s engineers would be managing a commercial enterprise rather than a mission. The subsidy model reached negotiated term sheets within about two years of enactment precisely because the firms, the engineers, and the supply chains already existed. The government had to redirect competence, not create it. The agency model fits when the government is the customer and no commercial market can sustain the capability. Semiconductors were the most commercial technology market in human history, so the customer test points to subsidy rather than to a federal fab. The instrument follows the transaction, not the other way around.

Q: What happens to the subsidy model when the appropriations run out?

The agency model builds permanent institutions while the subsidy model spends a fixed appropriation over a fixed period. NASA still exists, and its capabilities compound over decades. The semiconductor division’s manufacturing incentives are one-time money, and when the money is spent the leverage is spent too. The ten-year guardrails outlast the grants, but the grants themselves do not renew on their own. A future Congress must decide whether to authorize another round, and renewal will be judged on results that are still years away, since the first fabs will not produce chips at scale until late in the decade. This impermanence is the subsidy model’s structural cost, the mirror image of the agency model’s transfer problem. It also shapes the politics. The awards are negotiated once, the conditions bind for a decade, and the program’s defenders must argue for renewal before the outcome the money bought can be fully measured. An instrument that cannot be renewed without a new fight is weaker than an institution that endures between appropriations.

Q: How do Space Act cost-plus contracts differ from semiconductor grant conditions?

Both are the conditions each instrument attaches, but they operate in different legal systems and on different sides of the transaction. In the Apollo years NASA relied on cost-plus contracts, under which the government bore the development risk because the technology was too novel to price firmly. The conditions were specifications, inspection regimes, and termination rights that a bidder priced into the offer, and the government remained the buyer, inspecting deliverables and paying only for conformance. Semiconductor grant conditions attach to money rather than to a purchase: guardrails on foreign expansion, workforce and childcare plans, upside sharing, all negotiated project by project with bespoke milestones and reporting cadences. The cost-plus contractor’s incentive is to manage the customer relationship. The grant recipient’s incentive is to satisfy the conditions while maximizing the firm’s global position. The information asymmetry also differs. In procurement the government knows what it is buying. In subsidy the firm knows its true investment calculus and the government does not, which gives the firm the advantage in every negotiation over how binding the conditions can be.

Q: Which statutes should a reader study next after this comparison?

Read the anchor statutes in the order the study path recommends. Start with the 1958 Space Act profile for the agency model at full strength, including the invention title rule that carried the model to its logical conclusion. Then read the Bayh-Dole profile for the statute that reversed the title default and for the commercialization problem it was solving. Then read the CHIPS and Science Act treatment for the subsidy model in full, with the manufacturing incentives, the tax credit, the guardrails, and the unproven enforcement record examined statute by statute. End with the science policy survey, which places both statutes inside seventy years of federal science policy and shows the authorization-without-appropriation pattern the 2022 law’s science division repeats. Work through them with the customer test in hand. For each statute, ask who takes delivery, and the model, the instrument, and the conditions will resolve into place. That habit is the real product of this comparison, and it outlasts any single statute in the cluster.