The Underwriter
State

What Has Already Happened

Between July 2025 and August 2026, the U.S. government took direct ownership positions in the following publicly traded firms. The list is deliberately not sorted by importance, because the value of the aggregation is precisely in how routine each individual case appears when read in isolation.

A 9.9 percent stake in Intel, acquired August 22, 2025, structured as a conversion of previously awarded CHIPS Act grants into $8.9 billion of common equity, with a five-year warrant for an additional 5 percent at $20 per share triggered if Intel's ownership of its foundry business drops below 51 percent. A 15 percent stake in MP Materials, the rare earth producer, acquired by the Department of Defense in July 2025 for approximately $400 million, which made the U.S. government the company's largest shareholder. Additional positions in Lithium Americas and Trilogy Metals. A "golden share" in U.S. Steel, retained as a condition of permitting Nippon Steel's acquisition, granting the federal government veto authority over headquarters relocation, offshoring of production, and closure of facilities. Positions in at least six quantum computing firms. In aggregate: approximately $26.7 billion across roughly thirty equity or quasi-equity deals, spread across four separate agencies (seventeen through Commerce, seven through Defense, six through the International Development Finance Corporation, two through Energy), with no consolidated ledger of holdings.

The Fortune reporting that surfaced the aggregated figure is worth citing precisely because it identifies the operational tell: there is no unified accounting of what the government owns. The stakes exist. The transparency architecture that would normally accompany them does not. That is not accidental; it is the natural consequence of multiple agencies acting concurrently on adjacent but separately-justified rationales, none of which required a piece of enabling legislation because each was reachable through existing appropriations authority.

Commerce Secretary Howard Lutnick has publicly floated the extension to defense contractors, arguing that firms like Lockheed Martin, which derive nearly all their revenue from the U.S. government, are already "basically an arm of the U.S. government." The rhetorical construction there is important. He did not argue that the government should become an equity holder in Lockheed; he argued that the equity relationship already exists in economic substance, and formalizing it in law would only be catching the disclosure up to the fact.

That is the framing that will settle in over the next twelve months, regardless of whether the OpenAI deal specifically closes.

What Historical Era This Belongs To

The convention says this is unprecedented. That is only true if the sample begins in 1980.

The Reconstruction Finance Corporation, chartered by Herbert Hoover in 1932 and continued by Franklin Roosevelt until 1957, made direct equity and quasi-equity investments in banks, railroads, and industrial firms; during World War II, through its Defense Plant Corporation subsidiary, it built and owned roughly $9 billion of manufacturing capacity (approximately a tenth of all U.S. industrial plant at war's end, later sold back to the private sector). TARP, in 2008-2009, took Treasury equity positions in nine major banks and in AIG, GM, and Chrysler. The COVID-era airline warrants of 2020 continued the pattern. Sematech, in 1987, used a public-private research consortium equity structure to defend against Japanese semiconductor competition; it worked well enough that the shape of it survived into today's CHIPS Act.

Reach further back and the pattern is even less unusual. The British East India Company was a chartered instrument of state trade policy for two and a half centuries; the Dutch VOC was arguably the first modern state-backed multinational. Roman grain fleets, medieval royal monopolies on salt and tin, and the entire mercantilist framework of the seventeenth and eighteenth centuries proceeded on the assumption that critical trade categories required direct sovereign participation. Japan's MITI, from the 1950s through the 1990s, ran an industrial policy that included direct equity, credit steering, and coordinated foreign-exchange access for firms judged strategically important.

The anomaly is not the current moment. The anomaly is the roughly forty-year window, from the Reagan-Thatcher liberalizations in the early 1980s to approximately 2020, during which most large industrial democracies (notably not China, not Singapore, and not South Korea beyond a fig-leaf) conducted their strategic economic policy through regulation and subsidy rather than through ownership. That window is now closing, and I do not think anyone with a clear read of the historical baseline should be surprised.

The relevant question is not whether this is unprecedented (it is not) but what has changed such that the last forty years' arrangement no longer suffices.

Why Now

This is where the argument from №21 becomes structural rather than illustrative.

The claim I made in that dispatch was that reliability has changed units, from reserve to production rate; that the American security guarantee, in the categories that actually get consumed in modern conflict (interceptors, artillery shells, standoff strike, ISR platforms), is now denominated in output rather than inventory, and that output cannot be scaled at appropriations speed. That argument was about what the state owes its allies. It has a domestic mirror, and the domestic mirror is what "The Underwriter State" is really about.

Consider the categories in which the current administration has taken equity: semiconductors (Intel), rare earths (MP Materials), lithium (Lithium Americas), critical mineral development (Trilogy Metals), steel (US Steel golden share), quantum computing (six firms), and now reportedly AI foundation models (OpenAI, with Anthropic, Google, and Meta possibly invited). What links these categories is that each one is a chokepoint in the production of national strategic capability; each one requires capital deployment at a scale and duration that the private market either will not or cannot finance at the required velocity; and each one is a category in which allowing foreign (predominantly Chinese) dominance is judged politically unacceptable.

The regulatory-and-subsidy model of 1980-2020 assumed the private capital markets would price strategic categories correctly if the incentives were tuned. That assumption is now visibly failing on both ends. On one end, categories the government considers strategic (rare earth processing, advanced foundry capacity, indigenous lithium supply) do not attract sufficient private capital because the payback horizons are too long and the geopolitical risk premia are too high. On the other end, categories the government does not consider strategic (retail commercial real estate, mid-tier office towers, non-defense discretionary IT) have received abundant capital while the underlying assets deteriorate.

Equity ownership is the state's response to this pricing failure. It is a slow-motion admission that the regulatory-and-subsidy tools were insufficient for the throughput required. If you cannot get capital into the category through incentives, you become the capital.

What Changes When the State Becomes an Equity Holder

Four things change; only one of them is discussed in public.

The first, and the only one that receives attention. The state has a direct financial interest in the survival and appreciation of the firms it owns. This creates conflicts of interest with regulation, with antitrust, with tax policy, and with the government's role as customer; and with rival firms in the same category (Nvidia versus Intel versus TSMC-in-America; MP Materials versus Lynas versus Chinese competitors). All of this is real and is being litigated in real time by ethics lawyers, journalists, and Congress. It is the least interesting of the four.

The second. The state acquires equity-holder information rights, which are structurally different from regulatory information rights. A regulator sees what a firm is legally required to disclose; a shareholder sees what management shares with its board. In categories like foundation model training, semiconductor process technology, and rare earth separation chemistry, the difference between these information positions is substantial. This is one of the underappreciated reasons why the AI labs may actually welcome a federal equity stake: it gives the government a legible way to be inside the tent, which reduces the ambient risk of hostile export controls or unfavorable regulation. The state gets information; the firm gets predictability. This is a real value exchange for both sides.

The third. The state acquires a form of capital that private markets cannot match: patient capital indifferent to quarterly returns and tolerant of extremely long payback horizons. Sematech worked for exactly this reason in the 1980s. The CHIPS Act plus Intel equity plus IDFC investment stacking suggests the same pattern is now being reconstructed at a much larger scale. In categories where the payback horizon is measured in decades (nuclear reprocessing, gallium extraction, quantum error correction, foundational model training), the government is often the only actor who can hold the position long enough for the return to arrive. Whether that produces good capital allocation is a genuinely open empirical question. What is not open is that private markets have declined to allocate capital at the required duration.

The fourth, and the one that will matter most in retrospect. The state acquires the political incentive to prevent the failure of the firms it owns. This is the moral hazard argument that critics correctly raise, and it is not just about individual firms; it is about the structure of expectations across the entire strategic-industry perimeter. Once the government owns 10 percent of Intel and 15 percent of MP Materials and (perhaps) 5 percent of OpenAI, every other firm in adjacent categories now knows that the government is a potential buyer, a potential rescuer, and a potential political constituency for its own survival. That reshapes the risk-taking behavior of the whole ecosystem in ways that are structurally similar to (and distinct from in specifics) the too-big-to-fail dynamic that developed in banking between 1998 and 2008.

The banking analogy is worth sitting with. The 1998 rescue of Long-Term Capital Management was, at the time, framed as a one-off. It was not; it was the beginning of a decade in which the boundary between private financial-firm risk and public sovereign risk gradually became so thin that in 2008 it collapsed entirely, and the American state discovered it owned a fair fraction of the financial system whether it wanted to or not. What is happening now in industrial policy is the version of that trajectory that starts with the ownership stake rather than ending with it. Whether that is more or less dangerous than the banking version is a question I do not know how to answer without seeing the endpoint.

The Circuit That Closes

There is a specific reflexive loop worth naming, because it is the one that will determine whether this era is remembered as a successful state-directed industrial revival or as the origin story of a very expensive failure.

Step one. The state takes an equity position in a strategic firm on the theory that this catalyzes private capital.

Step two. Private capital does indeed follow, partly on merit and partly because the state's presence is read as an implicit backstop.

Step three. The firm scales, sometimes dramatically, on the combined public-private capital stack. Both the state's equity value and the firm's operational capacity grow.

Step four. The firm's growth becomes structurally important enough that its failure would carry political costs the state cannot bear. The equity stake, which was originally an instrument of policy, becomes a hostage of policy.

Step five. The state, now a hostage of the firm's survival, begins making regulatory, tax, and customer-side decisions that favor the firm. This is not necessarily corrupt; it is the natural output of an incentive structure the state built for itself.

Step six. Rival firms in the same category, and eventually adjacent categories, either fail or restructure themselves to attract the same treatment, since competing against a state-backed firm on a pure-private-capital basis is not viable.

Step seven. The perimeter of state-owned equity expands, either through explicit new deals or through implicit expectations that the state will show up when a strategic firm approaches distress.

None of these steps requires bad faith on anyone's part. Each step is locally defensible given the previous one. The endpoint is a state that owns most of the strategic industrial base and has forgotten how to price the capital it deploys. Whether that endpoint is arrived at over five years or fifty is a policy variable, not an inevitability. But the direction of travel, once the loop is running, is not easily reversed. TARP was supposed to be temporary. Sematech was supposed to be temporary. The DPC's assets were supposed to be liquidated after the war (they were, but only after a decade of legal fights).

The Fieldnotes-shaped reading here is not that the loop is bad. It is that the loop is real, and that once it starts running the temporal-mismatch problem from №21 shows up again on a different clock. The state's capacity to originate equity positions is fast; its capacity to unwind them is slow; its willingness to unwind them decays over time as constituencies form. That is not a moral judgment. It is an observation about how bureaucratic institutions actually behave with fixed-asset holdings.

What Would Falsify This

I want to be honest about how this reading fails, because the case against it is not weak.

If the federal equity positions are unwound within a five-year window with substantial capital return to the Treasury (as TARP largely was), the moral hazard argument mostly dissolves, and this becomes a story about competent countercyclical industrial policy rather than a story about creeping state capitalism. The RFC provides real historical precedent for this outcome; it existed for twenty-five years and eventually wound down.

If the OpenAI, Anthropic, Google, and Meta stakes do not materialize (the current reporting is preliminary and one-sided), the AI-labs component of the argument weakens, and the underwriter state remains confined to hard-industrial categories where the historical precedent is less controversial.

If Chinese industrial-capacity leadership erodes over the next three years (a real possibility given demographic, real-estate, and semiconductor-sanctions headwinds), the strategic urgency justifying direct state equity weakens; the political will to sustain the ownership positions may weaken with it. This is the scenario in which the underwriter state is a temporary artifact of a specific competitive moment rather than a structural transition.

If the current administration is succeeded by one that treats these positions as illegitimate and moves aggressively to divest, the whole framework I have described might not persist. The Trump administration built the current portfolio through executive action and appropriations reallocation, not legislation; a successor can dismantle it through the same tools. Whether a successor would want to, given the political constituencies these positions create, is a separate question.

If private capital returns to strategic categories in volume during 2027-2028 (which some observers of the AI capex cycle believe is likely once the current bubble compresses), the pricing-failure argument that justified state entry no longer holds, and the ownership positions become vestigial.

These are all real objections. The reading survives them because the trajectory of the last eighteen months is not consistent with any of the scenarios above. But I should note that I am reading a trend that has been visible for a specific and short time window; the confidence interval on where it lands is genuinely wide.

The Tool in the Drawer

The clearest way to read what happened between July 2025 and August 2026 is that the U.S. government, without saying so, decided that regulation and subsidy were no longer sufficient tools for maintaining the industrial capacity that its national security posture assumes. Having decided that, it did the thing that any organization does when its existing tools are insufficient: it reached for a tool that had been sitting in a drawer since 1957.

The tool worked once. It may work again. The question I have not answered, and cannot answer from where I am sitting in August of 2026, is whether the state that reaches for that tool now still remembers how to put it down.

The Pew survey I opened №21 with measured what the world thinks of America. This dispatch is about what America thinks of itself, insofar as its revealed institutional behavior discloses that. The interceptor standard argument said the American security guarantee is now priced in production rate rather than reserve. The underwriter state argument says the American industrial base is now capitalized through direct sovereign equity rather than through incentivized private capital. These are the same story told in two ledgers. Both are adaptations to the same underlying condition, which is that the throughput of critical goods is what national capacity now means; neither the guarantee nor the industrial base can any longer be run on the assumption that throughput will show up because a market or an ally has been asked nicely.

Nobody put the tool down between the RFC and TARP; there was just a long institutional pause. The pause is over. Whether the pause turns out to have been the exception or the rule, we will know in about fifteen years.