Before AI, America Faced This Same Problem Five Times. Here's What It Did.
Before AI, America Faced This Same Problem Five Times. Here's What It Did.
The debate over who controls AI is new. The problem it is trying to solve is not.
Senator Bernie Sanders wants the public to own half of the country's largest AI companies. President Trump signed an order creating federal review for advanced models. The U.S. government directed Anthropic to suspend foreign access to its two most capable systems. The nationalization debate has arrived in public. What hasn't arrived yet is the historical context that might make it legible.
By Aaron Rose · Tech Reader Magazine · July 14, 2026
The word nationalization lands differently depending on who is saying it. To some it sounds like a reasonable policy instrument. To others it sounds like a category error applied to a technology sector that built its dominance precisely by staying clear of government ownership. The debate currently running through Washington policy circles, congressional proposals, and classified discussions inside the national security apparatus tends to generate more heat than light — partly because the participants are arguing past each other, and partly because none of them have agreed on what they mean.
What they are actually arguing about is an older question wearing new clothes. The question is this: when a private entity controls infrastructure too consequential to fail, too concentrated to self-correct, and too fast-moving for conventional regulation to track, how does a democratic government assert the public interest without destroying the thing it is trying to govern?
America has faced that question before. The answers it produced were imperfect, improvised, and durable in ways nobody anticipated. They are worth examining now — not because any of them maps cleanly onto artificial intelligence, but because the places where the parallels hold and the places where they break down are both instructive.
The Federal Reserve
The Federal Reserve was not designed. It was improvised under pressure following the Panic of 1907, formalized by the Federal Reserve Act of 1913, and has been contested, reformed, and reinterpreted in the eleven decades since. What it represents structurally is something the current AI debate has not yet found language for: a hybrid institution that is neither a government agency nor a private company, accountable to both and fully controlled by neither.
The twelve regional Federal Reserve Banks are private corporations, owned by member commercial banks, paying dividends to their shareholders. The Board of Governors in Washington is a federal agency whose members are appointed by the President and confirmed by the Senate. The Federal Open Market Committee, which sets monetary policy, draws membership from both. The result is an institution that exercises public authority — setting interest rates, regulating bank reserves, serving as lender of last resort — through a structure that is partially private, partially public, and entirely unlike anything that existed before it was created.
The parallel to frontier AI governance is not exact. But it is instructive. The Federal Reserve exists because two earlier experiments — pure private control of monetary policy through the banking system, and direct government management of currency — both produced catastrophic failures. The hybrid was not the preferred solution of any faction. It was the solution that survived the political process because it gave enough to enough interests to become law.
The Federal Reserve was not designed. It was improvised under pressure, formalized by statute, and contested ever since. AI governance is probably on the same road.
Frontier AI labs now exercise something analogous to monetary policy — not over the supply of money, but over the supply of cognitive capability. They control what models exist, what they can do, who can access them, and at what price. Those decisions have economy-wide effects that extend far beyond the labs' customer bases. The question of whether that kind of systemic influence can remain entirely within private governance, accountable only to shareholders and boards, is the same question that monetary policy raised a century ago. The Federal Reserve's hybrid structure is one answer to it. Not the only answer. Not necessarily the right answer for AI. But an answer with a working track record.
The Manhattan Project
In the summer of 1942, the United States government faced a technology problem it could not solve through conventional procurement. Building an atomic weapon required scientific expertise concentrated in universities and private research institutions, industrial manufacturing capacity spread across private companies, and operational security that no purely private arrangement could guarantee. The solution — the Manhattan Project — created something that had no legal template: a government program operated almost entirely by private contractors and academic institutions, under military command, producing a weapon the government would own and the contractors would never see deployed.
Los Alamos was operated by the University of California under contract to the Army Corps of Engineers. Oak Ridge was built and run by private construction and chemical companies under government direction. The Hanford Site in Washington State — where the plutonium for the Trinity test and the Fat Man bomb was produced — was constructed and operated by DuPont, a private corporation, under Army oversight. The scientists who designed the weapon held university appointments, not government positions. The industrialists who built the production facilities were private executives executing government contracts of a scope and secrecy that had no peacetime equivalent.
The parallel to the current AI moment is not the weapons application — it is the structural arrangement. A technology too consequential for purely private development, requiring expertise too distributed for purely government employment, operating under security constraints too severe for normal market dynamics. The Manhattan Project's answer was to blur every line simultaneously: the government funded and directed, the private sector operated and staffed, the output belonged to the state, and the institutional identities of the participants remained formally intact throughout.
That last point marks a significant difference from the current AI arrangement. The weapon the Manhattan Project produced belonged to the government outright. Frontier AI models remain private property — owned by the labs, commercialized by the labs, with the government asserting control over access and distribution rather than ownership. Whether that difference is structural or merely transitional is one of the open questions the current debate has not resolved.
That blurring left questions that took decades to answer — about liability, about intellectual property, about the obligations of scientists to institutions whose directives they sometimes found unconscionable. It also produced results no other structure available at the time could have produced. The arrangement was not elegant. It was fast.
The arrangement was not elegant. It was fast.
Enriched Uranium
Uranium enrichment is a less dramatic case than the Manhattan Project, which is part of what makes it useful. After the war, the United States needed to continue producing enriched uranium for weapons and, eventually, for civilian nuclear power. The Atomic Energy Act of 1946 placed all fissile material under government ownership and all enrichment under government operation. Private companies could use nuclear power but could not own the fuel cycle.
That arrangement persisted for three decades before partial privatization began in the 1970s. Full privatization of enrichment never arrived cleanly — the sector today remains a hybrid of government-owned facilities, private operators, government subsidies, and national security constraints that no purely commercial logic can explain. The half-life of the hybrid structure turned out to be longer than anyone anticipated when it was created. Entities built under emergency conditions, justified by temporary necessity, have a tendency to persist.
This is worth holding in mind as the current AI governance debate produces its first durable structures. What gets built now — under the pressure of export control directives, national security reviews, and the political urgency of great power competition — will not be easily dismantled when the emergency pressure subsides. The question of what kind of hybrid to create is also, necessarily, a question about what kind of hybrid to live with for the next several decades.
Entities built under emergency conditions, justified by temporary necessity, have a tendency to persist. The half-life of hybrid structures is usually longer than anyone anticipated.
Entities built under emergency conditions, justified by temporary necessity, have a tendency to persist.
COMSAT
In 1962, Congress passed the Communications Satellite Act and created COMSAT — the Communications Satellite Corporation. The legislation established a private, for-profit company, publicly traded and shareholder-owned, with an explicit statutory mandate to serve the public interest in global satellite communications. Half its board was elected by public shareholders. Half was appointed by the President of the United States. Its customers were both private telecommunications carriers and foreign governments. Its mission was commercial. Its creation was an act of Congress.
COMSAT occupied a category that had no name. It was not a government agency. It was not an ordinary corporation. It was not a regulated utility in the traditional sense. It was a private company created by statute, governed partly by its shareholders and partly by presidential appointment, operating in a domain the government considered strategically important enough to shape but not important enough to own outright.
The structure worked, in the sense that COMSAT functioned and global satellite communications developed. It eventually became harder to justify as the strategic urgency faded and commercial satellite operators proliferated. COMSAT was absorbed by a private company in 2000, thirty-eight years after its creation. The hybrid persisted for nearly four decades before the conditions that created it changed enough to make it unnecessary.
The COMSAT precedent is the least discussed of the historical analogues, which may be why it is among the most applicable. It represents a moment when the U.S. government decided that a new communications technology was too strategically important to leave entirely to market dynamics, but not important enough — or not politically viable enough — to nationalize directly. The solution was a new kind of entity that had not previously existed. Congress invented the legal structure to fit the policy need. Whether frontier AI requires a similarly new legal category — or can be accommodated within existing structures, adapted under pressure — is a question the current debate has not yet resolved.
What COMSAT also illustrates is a risk that every hybrid structure carries: over time, the private partner tends to shape the public institution more than the public institution shapes the private partner. COMSAT did not fail — it simply became irrelevant as the industry it was designed to govern matured around it. Regulatory capture in AI governance may not look like corruption or scandal. It may look like gradual accommodation, decade by decade, until the public interest function the structure was created to serve becomes indistinguishable from the commercial interest it was created to check.
Operation Warp Speed
Operation Warp Speed, launched in 2020, offers the most recent and perhaps most legible precedent. The federal government committed billions of dollars to accelerate vaccine development, guaranteed advance purchase commitments to reduce commercial risk, coordinated manufacturing scale-up across private companies, and maintained regulatory oversight through the FDA — all while the pharmaceutical companies retained ownership of the resulting products, their manufacturing relationships, their intellectual property, and their profit rights.
The government told the private sector what to make, funded the development, removed the market risk, and then stepped back from the product. The companies kept the upside. The public got the vaccines. The arrangement was not a contract in the ordinary sense, not a nationalization in any sense, and not a partnership in any sense the word usually carries. It was something improvised to fit a specific emergency that had no prior template.
The structure that appears to be emerging around frontier AI models bears a recognizable resemblance. Government funding through contracts and grants. Government direction of priorities through national security requirements. Government control of access through export control authority. Private companies retaining ownership, operational control, and the commercial upside of deployment. The public bearing the downside risk of whatever the technology produces that markets cannot price.
Where the Parallels Hold
Each of these cases illuminates a different facet of the current situation. The Federal Reserve speaks to the governance structure question — how do you create an institution with public authority that is not fully captured by either government or private interest. The Manhattan Project speaks to the national security dimension — what happens when a technology is so consequential that government cannot stay at arm's length, regardless of who nominally owns it. Uranium enrichment speaks to the long tail — hybrid structures built under emergency conditions persist far longer than their creators intended. COMSAT speaks to the invention of new legal categories — when no existing structure fits the policy need, Congress has been willing to create one. Operation Warp Speed speaks to the public mission / private profit arrangement that seems to be emerging in real time.
None of these parallels is exact. The differences matter as much as the similarities.
Frontier AI is not monetary policy, though it shares the property of systemic influence over economic decisions. It is not a weapons program, though it shares the property of dual-use capability that cannot be cleanly separated into civilian and military applications. It is not a fuel cycle, though it shares the property of being foundational infrastructure whose control raises questions that purely commercial logic cannot resolve. It is not a satellite communications company, though it shares the property of being a new technology whose strategic importance Congress recognized before markets could fully price it. It is not a vaccine, though it shares the property of being a product whose development the government is willing to fund and direct while the private sector retains the intellectual property.
It is, in some ways, all of these things simultaneously — and in the combination, something that the existing precedents do not fully capture.
None of these parallels is exact. The differences matter as much as the similarities.
The Speed of AI Development
What makes the current moment structurally different from any of the historical cases is not the nature of the technology. It is the speed at which the technology is moving relative to the institutional capacity available to govern it.
The Federal Reserve took six years from the Panic of 1907 to formal legislation. The Atomic Energy Act took a year after the war ended. COMSAT took several years of Congressional debate after Sputnik. These timelines, while fast by the standards of major institutional innovation, were calibrated to technologies whose development pace allowed for deliberation.
Frontier AI capability is advancing on a timeline that makes deliberative institution-building extremely difficult. The model that triggers a national security review today may be two generations behind the frontier by the time any formal governance structure created in response to it is operational. The structures being built now are being built for a technology that will not be the same technology by the time the structures are finished.
This is the honest version of the challenge. Not that AI is unprecedented in kind — the historical parallels are real and instructive. But that the familiar forces are operating at an unfamiliar velocity, and the institutional machinery available to respond was calibrated for slower disruption cycles.
The familiar forces are operating at an unfamiliar velocity. The institutional machinery available to respond was calibrated for slower disruption cycles.
What the Debate Is Actually About
The nationalization proposals currently circulating — Senator Sanders's sovereign wealth fund, the Trump administration's model review order, the export control directives that have already produced real consequences for real products — are not a single coherent policy position. They represent different answers to different versions of the same underlying question, arriving simultaneously from different directions.
The Sanders proposal is primarily about wealth distribution — who captures the economic value that AI generates. The model review order is primarily about capability gatekeeping — who decides what gets released and when. The export control directives are primarily about access control — who can use what the labs have already built. These are related questions, but they are not the same question, and the governance structures appropriate to each are different.
What the public debate obscures is that the executive branch is not a single actor in this process. The Pentagon, the intelligence community, the Commerce Department, the National Security Council, and the Treasury each have distinct equities in frontier AI — and distinct preferences about what governance should look like. The export control directives that produced real consequences for real products in 2026 came from Commerce, but the national security logic driving them originated elsewhere.
Which agency emerges as the primary governance authority over frontier AI will shape the hybrid structure that results as much as any legislative proposal. An arrangement designed around Defense priorities looks different from one designed around Commerce priorities, which looks different again from one designed around the intelligence community's access requirements. That internal competition has not been resolved. Its resolution will matter.
What they share is the recognition — now bipartisan, however differently motivated — that the current arrangement, in which frontier AI development is governed primarily by the boards and investors of a small number of private companies, is not the permanent state of affairs. The debate is not really about whether public authority asserts itself over frontier AI. It is already asserting itself, in real time, through export controls and access directives and model reviews. The debate is about what form that assertion takes, how it is legitimized, and who is accountable for the results.
The historical precedents suggest that the form it takes will be improvised, that the improvisation will be formalized under pressure, and that the structure produced will persist longer than anyone expects and require more maintenance than anyone plans for. They also suggest that the question of whether to build a hybrid structure is probably already settled by events. The question that remains open is what kind of hybrid, built on what principles, accountable to whom.
The Coming AI Governance Structure
The Federal Reserve did not arrive as a designed solution. It arrived as the answer that survived when the alternatives failed visibly enough that something had to change. The Panic of 1907 was the crisis that made the debate impossible to defer. Something analogous may be on the horizon for AI governance — not necessarily a single dramatic event, but an accumulation of incidents that makes the current improvised arrangements untenable.
When that inflection arrives, the historical record suggests that the structure built in response will draw on whatever precedents are closest at hand, modified by the specific pressures of the moment, and shaped by the political coalition that can be assembled to pass it. It will not be elegant. It will not be complete. It will leave questions that take decades to answer.
What it will be is durable. The hybrid structures that American policy has built when private ownership and public interest collided at sufficient scale have proven remarkably persistent — not because they were optimal, but because they were the structures that existed when the next crisis arrived, and it was easier to adapt them than to replace them.
The parallels to AI are imperfect. They are also the best available guide to what comes next. Time will tell what form it takes. History suggests it will be something nobody fully designed, governed by rules nobody fully anticipated, producing outcomes nobody fully predicted — and still operating, in some recognizable form, several decades from now.
Coming Soon at Tech Reader Magazine
The Case for Nationalization
The strongest version of the argument for public ownership or
control of frontier AI labs and hyperscalers — market concentration, the
accountability gap, national security logic, and the velocity problem. At
Tech Reader Magazine.
Tech Reader Magazine
TechReaderMagazine.com