The Atom and the Chip 2.0

The Rush to Nuclear, and the Water Left Behind

Author: Adam Wood | Publication: Blue Collar Analytics | Series: Fortress America, Paper V | Date: July 2026 | Version: 2.0

Revision note (v2.0): This version makes two kinds of changes. First, it adds three new sections — The Demand (Section 2), The Fuel Bottleneck (Section 6), and The Military Buildout (Section 5) — built from three weeks of continued research after version 1.0’s publication. None of that material corrects the original paper; it confirms it. The named hyperscaler contracts, the enrichment awards, and the three parallel military reactor programs documented below are the direct evidence version 1.0’s central bet could only infer, surfacing at a rate that is itself proof of the original argument: capital and government moving faster than public visibility can track. Second, this version corrects an error version 1.0 introduced on its own: Section 5 of the original paper restated Paper III’s water findings using figures from Paper III’s pre-v1.2 text — the single 10-million-gallon fab figure rather than the audited 3-to-10-million range, a TSMC reclamation figure that describes the company’s Taiwan operations rather than Arizona, an Intel “zero-liquid-discharge” characterization that Paper III’s claim-by-claim audit replaced with the accurate and narrower “net positive,” and a description of Sherman’s wastewater plant as fab-specific when it is citywide infrastructure. The water section (now Section 7) is rebuilt to match Paper III v1.2’s audited findings. The Mechanism section is expanded from two federal postures to three, reflecting the contracting model the military programs introduce. Substantive conclusions are unchanged; the corrections align this paper with the standard the rest of the series has already been held to.

Second, on TerraPower’s Natrium plant at Kemmerer, this version adds material rather than correcting it: version 1.0 noted only that Natrium uses conventional steam-cycle generation, and said nothing about its water siting. Section 7 now documents that the plant is built beside the Naughton generating station — frequently described in public materials as a “retiring coal plant,” a characterization accurate as to coal but incomplete, since Naughton is converting to natural gas and continuing to operate. The point is not a correction to the earlier version but an illustration of a pattern this paper flags throughout: a claim can be literally true and still leave a materially incomplete impression, and such claims are treated here as soft claims and stated in full.

EXECUTIVE SUMMARY

The federal government and private capital are betting, at scale and at speed, that the AI and semiconductor buildout this series has already documented will need far more electricity than the grid currently supplies — and nuclear power is the answer they have chosen to fund. Version 1.0 of this paper made that claim by inference: hundreds of billions of dollars moving toward generation capacity sized to match an industrial demand curve, with no single reactor traceable to any single fab. The inference is no longer necessary. The demand side of the bet now has names, dollar figures, and contract terms.

Because this paper tracks a buildout that produces new announcements almost daily, it rates every commitment by the enforceability of the instrument behind it, not the size of its headline number — and in this buildout, those two run in opposite directions. From strongest to weakest: a binding contract or executed transaction — a signed power-purchase agreement, a deed — where money and obligation are both real; a conditional commitment, real money obligated but gated on a future decision; a memorandum of understanding, a framework to cooperate with no purchase obligation; a letter of intent, a stated intent to transact that binds no one; and a pipeline figure, a self-reported demand number with no instrument behind it at all. Throughout this paper, every deal is tagged against this ladder. The reason it is necessary shows at the extremes: the largest firmly contracted commitment in the entire buildout is Amazon’s 1.92 gigawatts, a signed agreement — while the largest gigawatt figures anywhere belong to the bottom rung, self-reported pipelines attached to reactors that do not yet exist. The numbers grow as the commitments soften.

The four largest AI infrastructure operators have each signed binding, long-term power-purchase agreements tying named nuclear plants to their own demand, running seventeen to twenty years against restarted, existing, and not-yet-built capacity. One brings a federal utility inside the deal as the contracted buyer; another sits alongside a federal loan to the plant it draws from. These are the top of the ladder: signed, enforceable, decades long. No company signs a twenty-year power contract against a demand curve it believes might flatten.

The fuel that demand requires has its own federal response, bipartisan in origin: a ban on Russian enriched uranium that unlocked billions in domestic enrichment funding, followed by major federal awards to three enrichment companies. And the military is running three distinct reactor programs under a contracting model that makes the government a guaranteed customer rather than an owner — a third federal posture alongside the deregulation and equity-adjacent lending version 1.0 documented.

The original paper’s findings stand unchanged beneath these additions. Four reactors reached criticality under new, deregulated federal testing pathways, all privately financed, meeting and exceeding an executive order’s target of three by July 4, 2026. The Department of Energy issued a $17.5 billion conditional loan commitment to Westinghouse — real and funded, but gated on a final investment decision, and structured through vehicles and utility partners not yet named. It sits high on the ladder as a commitment and remains, as of this writing, invisible in its particulars.

That combination — enormous commitment, minimal disclosure — is the paper’s throughline, and the new material widens the gap rather than closing it. The five sites that will receive the Westinghouse billions are not public. What the hyperscalers are paying, in most cases, is not public. The military’s final vendor and site selections are pending. And the water requirement of the new-build reactors in this paper is, with a single partial exception, unaddressed in any document this paper has reviewed — one project has visibly changed course to confront the constraint; the rest have not. That gap is the sharpest, because the hyperscaler deals overwhelmingly attach to existing or restarting plants whose water footprints were settled decades ago. The water question concentrates almost entirely on the new-build track — the same track where disclosure is thinnest.

None of the material added in this version contradicts the paper it revises. All of it confirms the original thesis at a scale and speed the first publication could not capture. Everything documented here is a floor, not a ceiling — and the rate at which new, dated, independently sourced material surfaced in three weeks is itself evidence for this paper’s core argument: pace outrunning visibility.

SECTION 1 — THE BUILDOUT

The deregulatory and capacity-building groundwork for this buildout is bipartisan, and the clearest evidence of that sits at the beginning of the timeline. The ADVANCE Act passed both chambers of Congress in mid-2024 by margins rarely seen on energy legislation — 88-2 in the Senate, 393-13 in the House — directing the Nuclear Regulatory Commission to cut review fees for advanced reactor applicants and build a faster, technology-neutral licensing framework. President Biden signed it into law on July 9, 2024.[1] Four months later, on November 12, 2024, his administration issued the U.S. Nuclear Energy Deployment Framework, setting a target of roughly 300 gigawatts of nuclear capacity by 2050 — triple the current fleet.[2]

On May 23, 2025, President Trump signed four executive orders that built directly on that foundation. EO 14300, “Ordering the Reform of the Nuclear Regulatory Commission,” set a target of 400 gigawatts by 2050 — four times today’s roughly 100 — alongside the licensing timelines discussed in Section 4.[10] EO 14302, “Reinvigorating the Nuclear Industrial Base,” directed the Department of Energy’s Loan Programs Office — since renamed the Office of Energy Dominance Financing — to prioritize nuclear restarts, uprates, and new construction.[3] EO 14299, “Deploying Advanced Nuclear Reactor Technologies for National Security,” directed the rapid deployment of reactors at DOE sites supporting AI infrastructure and, on the national-security track later implemented through the Army’s Janus program, a reactor operating at a domestic military base by September 30, 2028.[9] EO 14301, “Reforming Nuclear Reactor Testing at the Department of Energy,” created the Reactor Pilot Program and set the target this section traces: at least three advanced reactor designs reaching criticality outside the national laboratories by July 4, 2026.[6]

That target was not merely met but exceeded, and by more programs than one. Four privately developed reactors reached criticality by the July 4 deadline. Antares Nuclear’s Mark-0 was first, achieving zero-power criticality at Idaho National Laboratory on June 4, 2026 — the first privately developed non-light-water reactor to do so in the United States in more than forty years; Mark-0 uses a sodium heat-pipe design.[4] Zero-power criticality — a self-sustaining chain reaction at essentially no measurable output — proves a reactor’s core physics before any attempt to generate power, and it was this milestone, not power generation, that the executive order’s July 4 target required. Valar Atomics’ Ward 250 followed on June 18 at the Utah San Rafael Energy Lab in Emery County, becoming the first reactor built and operated entirely outside a national laboratory; it is helium-cooled, a third approach distinct from Mark-0’s sodium and from the ordinary water used in the AP1000 design discussed later.[5] Deployable Energy’s Unity reached criticality on June 30 — the third to do so, and the first under the Nuclear Energy Launch Pad, the successor pathway the DOE established in March 2026 to extend Reactor Pilot Program authorization to developers beyond the original cohort.[13a] Aalo Atomics’ Critical Test Reactor was the fourth and last, going critical at 12:20 a.m. on July 4 itself.[13b] Three of the four came through the Reactor Pilot Program; Unity came through its successor — four reactors, two programs, against a target of three.

Only one of the four has so far gone beyond cold criticality. In the days after June 18, Valar began power ascension, reaching its rated output of roughly 100 kilowatts thermal, and on July 1 briefly powered an Nvidia AI chip from that output — the first time a U.S. advanced reactor has done so, and an early proof of concept far below any commercial scale. Valar’s own framing captures the distinction this paper relies on throughout: “Cold ≠ Hot: Cold proves the physics. Hot proves the power.”[11] By that standard, the July 4 milestone was a cold one, met four times over; the hot demonstration, so far, belongs to Valar alone.

Ward 250 was also the subject of the first aerial transport of a reactor in U.S. history. In February 2026, U.S. Air Force C-17s carried the disassembled, unfueled unit from March Air Reserve Base, California, to Hill Air Force Base, Utah — two states, not a cross-country journey — in an operation named Windlord, a multi-wing effort drawing on the airlift capability that routinely transports U.S. nuclear weapons.[12] The mission used military aircraft and crews, though the Wall Street Journal reported that Valar paid the cost of the flight. All four criticality reactors were developed with private capital — reported figures include roughly $130 million raised by Valar toward Ward 250 — and required no federal money to reach the milestone, only a faster federal permitting pathway than the Nuclear Regulatory Commission’s standard licensing process.[14]

The federal financing story is a separate and much larger track, running in parallel. On June 23, 2026, the Department of Energy’s Office of Energy Dominance Financing issued a conditional loan commitment of up to $17.5 billion to Westinghouse Electric Company, to finance the fixed-price purchase of long-lead equipment for up to ten AP1000 reactors — a large, roughly 1,100-megawatt pressurized water reactor design, and the only large-scale advanced reactor currently licensed for commercial operation in the United States — across up to five two-reactor project sites.[7] The announced structure routes the loan through a Westinghouse special purpose vehicle expected to administer up to five project funding vehicles, each jointly owned by Westinghouse and a utility partner and each required to commit $500 million in equity — $1 billion per project — before any federal funds are released, with the loan repaid from the sale of the long-lead items. As of this writing, none of the five sites, utility partners, or vehicles has been named; Westinghouse has stated it signed letters of intent with seven candidate utilities, to be narrowed to five.[7] “Conditional” matters here: the commitment depends on conditions not yet satisfied, including a final investment decision that has not been made, and is distinct from money that has actually been disbursed. On the paper’s commitment ladder, this is a conditional commitment — strong, funded, and signed, but gated and, in its particulars, not yet public.

This $17.5 billion loan operationalizes part of a larger $80 billion partnership the Department of Commerce announced in October 2025, under which — after a final investment decision — the federal government is entitled to 20 percent of Westinghouse distributions in excess of $17.5 billion, and may require an initial public offering and receive warrants for an equity stake of up to 20 percent if Westinghouse is valued at $30 billion or more by January 2029.[7a] This is the same equity-adjacent architecture Paper IV traced through MP Materials, Vulcan Elements, and Trilogy Metals, now applied at a larger dollar scale to the company building the reactors. Westinghouse itself is majority owned by Brookfield, a private asset management firm, with Cameco, a uranium mining company, holding the remainder — the same upstream-supply-chain pattern Paper IV documented, in which a critical-materials company’s equity holders benefit directly from federal backing.[22]

Geographically, the buildout is concentrated but not confined to two states, and those states have begun coordinating directly. Wyoming hosts TerraPower’s Natrium reactor near Kemmerer, a sodium-cooled fast reactor that received its NRC construction permit in March 2026 — the first commercial reactor construction permit the agency has issued in nearly a decade, and the first ever for a commercial non-light-water power reactor.[15] Construction on the plant’s non-nuclear components was already under way, and the project is targeted for completion around 2030. Utah hosts the densest concentration of nuclear activity in the country: at least eight companies — TerraPower’s second site, Valar Atomics, Anfield Energy, Nusano, General Matter, Atlas Atomics, Curio, and the revived Blue Castle project near Green River — operating under Governor Spencer Cox’s Operation Gigawatt, an initiative to double the state’s energy production by 2034.[16] Idaho, Wyoming, and Utah have formalized their coordination through a tri-state agreement, signed by the three governors in April 2025, tying nuclear development to shared energy goals, with Idaho National Laboratory — where Antares went critical — serving as the shared technical anchor.[17] South Carolina, Ohio, and Tennessee each host at least one additional project: the abandoned V.C. Summer AP1000 units, which Santee Cooper voted to restart with Brookfield in October 2025; a new Elementl Power plant in development; and Kairos Power’s Hermes reactor under construction at Oak Ridge, respectively.[18,19,20]

None of this geography overlaps with the southern spine this series mapped in Paper III. Arizona, Texas, and California host the semiconductor fabs and water infrastructure documented there; no reactor in this paper sits near any of them. That absence does not weaken this paper’s central claim. The bet being made is about total generation capacity matching total projected demand, not about any single reactor’s transmission lines reaching any single fab. And as the next section documents, the demand side of that bet no longer has to be inferred.

SECTION 2 — THE DEMAND: THE HYPERSCALERS

Version 1.0 of this paper stated its central limitation plainly: whether any specific reactor ever powers any specific fab was not the point, because no contract connected them. That limitation no longer describes the record. Between early 2024 and mid-2026, the four largest AI infrastructure operators in the world each signed long-term agreements tying named nuclear plants to their own demand, in deals whose duration — seventeen to twenty years — are themselves a finding. No company signs a two-decade power contract against a demand curve it believes might flatten.

Because this buildout produces new announcements almost daily, this paper rates every commitment by the enforceability of the instrument behind it, not the size of its headline number — and those two run in opposite directions. From strongest to weakest: a binding contract or executed transaction — a signed power-purchase agreement, a deed — where money and obligation are both real; a conditional commitment, where real money is obligated but gated on a future decision; a memorandum of understanding, a framework to cooperate carrying no purchase obligation; a letter of intent, a stated intent to transact that binds no one; and a pipeline figure, a self-reported demand number with no instrument behind it at all. The ladder is necessary because in this buildout the gigawatt figures grow as the commitments soften. Amazon’s 1.92 gigawatts sit at the top as a signed contract. At the bottom sits a company like Deep Fission, whose reactor-development pitch advertises a customer “pipeline” that grew from 12.5 to 18.5 gigawatts between late 2025 and mid-2026 — a figure larger than any contracted commitment in this paper, attached to a company whose reactor, as of this writing, exists as a design and a non-nuclear prototype. Deep Fission has drilled a single data-acquisition well and is still seeking permits for its full-scale test borehole; the canister it has delivered to its Kansas site is, by the company’s own description, non-nuclear — built to rehearse the underground installation workflow before any fuel is introduced. The largest number on the board belongs to a company that has not yet drilled a reactor hole, let alone filled one. Every deal below is tagged to its rung.

The four hyperscaler agreements are the top of that ladder — binding, enforceable, decades long. Microsoft signed a twenty-year power-purchase agreement for the output of the Crane Clean Energy Center, the 835-megawatt plant formerly known as Three Mile Island Unit 1, whose restart is targeted for 2027 — accelerated from an original 2028 estimate — and backed by a $1 billion loan from the same DOE Energy Dominance Financing Program that issued the Westinghouse commitment in Section 1.[24] The federal lending instrument and the private demand contract meet inside a single project. Meta signed a twenty-year agreement for 1,121 megawatts from Constellation’s Clinton Clean Energy Center in Illinois, beginning June 2027 — an existing plant whose economics the contract effectively underwrites for a generation.[25] Amazon’s commitment came in two parts: in 2024 it acquired the data-center campus adjacent to Talen Energy’s Susquehanna plant in Pennsylvania for $650 million, and in June 2025 it signed a seventeen-year agreement for 1.92 gigawatts of that plant’s output — the largest single hyperscaler-nuclear commitment on record — collapsing the distance between generation and load to the width of a property line.[26] Google’s arrangement is structurally the most novel. It signed a Master Plant Development Agreement with Kairos Power in October 2024 for up to 500 megawatts of new reactor capacity by 2035 — among the first corporate commitments to reactors that do not yet exist — and a follow-on agreement in August 2025 brought in the Tennessee Valley Authority, a federal utility, as the contracted buyer of the power from the first plant.[27] A federal entity now sits inside a private hyperscaler’s nuclear procurement chain as the off-taker, a blend of public and private roles that fits no prior category cleanly.

Two further agreements sit one rung down, and the distinction is the point. Equinix and Prometheus Hyperscale have each signed not power-purchase agreements but letters of intent with Oklo — Equinix for 500 megawatts, accompanied by a $25 million pre-payment, and Prometheus for 100 megawatts.[28] These are real signals of demand, and the Equinix pre-payment puts money behind one of them, but a letter of intent binds no one to buy or sell; it belongs in the same category as the seven utility letters of intent behind the Westinghouse loan. Oklo — a company that will reappear in this paper’s military and fuel sections — sells power it intends to generate itself, which makes these customer commitments rather than plant-output contracts, and softer ones than the four above.

Two features of this record matter more than any individual deal. The first is timing. The Google-Kairos agreement dates to October 2024; the joint Request for Information that Google, Microsoft, and Nucor issued to aggregate demand for advanced clean-firm power — nuclear among several named technologies, alongside next-generation geothermal, clean hydrogen, and long-duration storage — dates to early 2024, both predating the May 2025 executive orders by more than a year.[29] The demand signal was not conjured by the current administration’s deregulation; it preceded it, which is the same continuity finding this series has made in every sector it has examined. The second is coordination. The joint RFI means the hyperscaler bet is not parallel isolated deal-making but a pooled industrial signal — three of the largest capital allocators in the country telling the clean-power industry, in one document, that the demand is real enough to build against.

This is the missing direct evidence for this paper’s central thesis. Version 1.0 argued that capital was betting on nuclear to match AI and chip demand. The bet now has counterparties, contract durations, megawatt figures, and, in Amazon’s case, a deed.

SECTION 3 — THE MECHANISM

Version 1.0 of this paper identified two federal postures operating inside this buildout. The record now requires three, and they should not be mistaken for one another.

The first is deregulation. Executive Order 14301 created the Reactor Pilot Program, a Department of Energy authorization pathway that lets private developers construct and operate full-scale test reactors outside the standard Nuclear Regulatory Commission licensing process, using the Department’s own authority under the Atomic Energy Act rather than a commercial NRC license.[6] Antares Nuclear, Valar Atomics, and Aalo Atomics moved through this pathway, and Deployable Energy through its successor, the Nuclear Energy Launch Pad — all developing their reactors with private capital, with a compressed federal review standard and shared DOE-NRC staff as the program’s actual contribution.[21] Operation Windlord, the February 2026 airlift of Valar’s unfueled reactor aboard Air Force C-17s, is the apparent exception that proves the rule: though the mission used military aircraft and crews, the Wall Street Journal reported that Valar paid the cost of the flight — estimated at under $1 million — so even this most visibly governmental moment was, in the end, privately funded.[12]

The second is direct federal financing through an equity-adjacent instrument — the posture this series documented at length in Paper IV. The Department of Energy’s Office of Energy Dominance Financing issued its $17.5 billion conditional loan commitment to Westinghouse on June 23, 2026, to finance the fixed-price purchase of long-lead reactor equipment.[7] The loan is expected to route through a Westinghouse special purpose vehicle administering up to five project funding vehicles, each jointly owned by Westinghouse and a utility partner committing $500 million in equity per project before any federal funds are released, with repayment drawn from the sale of the equipment — a structure whose sites, partners, and vehicles remain unnamed as of this writing. The equity-adjacent terms sit in the parallel $80 billion partnership the Department of Commerce announced in October 2025: after a final investment decision, the government is entitled to 20 percent of Westinghouse distributions above $17.5 billion, and may require an IPO and take an equity stake of up to 20 percent at a $30 billion valuation by January 2029.[7,7a] This is the same architecture Paper IV traced through MP Materials, Vulcan Elements, and Trilogy Metals, applied at a larger dollar scale to the company building the reactors. Westinghouse itself is majority owned by Brookfield, with Cameco holding the remainder — the same upstream-supply-chain pattern documented in Paper IV, in which a critical-materials company’s equity holders benefit directly from federal backing.[22]

The third is the guaranteed customer, and it is new to this paper. Two of the three military reactor programs documented in the next section — the Army’s Janus program and the Advanced Nuclear Power for Installations (ANPI) program — share a single contracting model: reactors that are commercially owned and operated (COCO), contracted through Other Transaction Authority, paid against milestones, and explicitly modeled on NASA’s Commercial Orbital Transportation Services program, the framework that produced SpaceX’s cargo and crew capability.[38] Under this posture the government neither steps back as deregulator nor buys in as shareholder. It makes milestone payments as each development stage is met, and commits to buy the resulting power — funding the reactor’s progress in stages while guaranteeing itself as the eventual customer. It is a generalization of the same economic function the Department of Defense’s offtake agreement performed for MP Materials in Paper III, now built into a standing contract architecture. The third military program, Project Pele, does not share this model: it is an earlier, government-directed Defense Department demonstration, built by BWX Technologies (BWXT) under a conventional contract, that Janus and ANPI cite as the precedent they were designed to move beyond.

Taken together, the three postures let the federal government occupy every position around this buildout at once: the regulator stepping aside, the lender-shareholder stepping in, and the customer standing at the end of the line. The distinction among them separates what has already happened from what has only been promised. The Reactor Pilot Program’s reactors were developed with private capital and have gone critical; that work is done and cannot be unwound. The Westinghouse program is a conditional commitment, dependent on a final investment decision not yet made, structured around sites and partners not yet named. The military programs sit in between: contracts signed or shortlisted, sites named, deadlines statutory — but the reactors that will fulfill them, Janus and ANPI alike, are not yet built.

SECTION 4 — THE SPEED

What changed in 2025 was the pace at which the ADVANCE Act’s groundwork got implemented. Executive Order 14300 set explicit numeric deadlines: eighteen months for the NRC to issue a final decision on a new reactor license, and roughly twelve months for license renewals.[10] On March 25, 2026, the NRC finalized “Part 53,” the first entirely new commercial reactor licensing framework since 1989, when the agency created Part 52. Part 53 does not replace the existing Part 50 and Part 52 pathways; it stands alongside them as a third, optional framework developers can choose instead.[23] The order behind this reform cites the underlying bottleneck directly: between 1954 and 1978, the NRC authorized 133 reactors that were completed; since 1978, only two have entered commercial operation.[10] The rule was published in the Federal Register five days after the vote and took effect April 29, 2026.[41] The NRC also cut its hourly review fee for advanced reactor applicants from $318 to $148, a reduction of more than half, effective October 2025 — implementing a fee structure the ADVANCE Act had directed it to build the year before.[41]

Energy Secretary Chris Wright has supplied the historical scale that makes this reform meaningful: in the roughly twenty-five years after the first reactor generated civilian electricity at Idaho National Laboratory in the 1950s, the United States permitted and began construction on more than a hundred reactors before the industry, in his account, “ground to a halt” on regulatory grounds.[17] The current target — ten large reactors under the Westinghouse program, plus a handful of pilot microreactors — is a small fraction of that historical pace, even with review timelines compressed by years. This is the civilian speed this paper has referenced throughout: real, congressionally authorized, bipartisan in origin, and now backed by a specific regulatory framework and fee structure, applying to every developer in this buildout regardless of company or state.

Military speed is a different case, and version 1.0 undersold how much of one. Operation Windlord demonstrated what an actor operating outside the civilian licensing structure entirely can accomplish, using military aircraft, crews, and the Department’s own authority under the Atomic Energy Act — even as the private developer, not the government, paid for the flight. Section 5 shows that demonstration was not isolated: three parallel programs, statutory deadlines, and a contracting model built for speed. None of this says anything about whether the civilian licensing system has gotten faster — that case is made on its own terms above. But the thread this paper flagged in version 1.0 as “worth watching” has, in three weeks of subsequent research, turned out to be a program of record three times over.

SECTION 5 — THE MILITARY BUILDOUT

Version 1.0 treated the military as a single data point — one airlift, Operation Windlord, offered as evidence of what an actor outside the civilian licensing system could do. Continued research shows that framing was too small. The military track is not one demonstration; it is three distinct, parallel programs, each with named vendors, named or shortlisted sites, and dated deadlines.

Project Pele is the foundation — a transportable microreactor program run by the Defense Department’s Strategic Capabilities Office, with BWX Technologies as integrator, targeted for operation in 2028. It predates the others in this paper’s timeline and established the premise they build on: that a reactor can be a deliverable unit rather than a construction site.[35]

The Advanced Nuclear Power for Installations program — ANPI — launched in June 2024, under the Biden administration and a year before the May 2025 executive orders, run jointly by the Defense Innovation Unit, the Army, and the Air Force. Eight companies were named eligible in April 2025: Antares, BWX Technologies, General Atomics, Kairos, Oklo, Radiant, Westinghouse, and X-energy. In April 2026 the program announced its first vendor-site pairings: Radiant at Buckley Space Force Base in Colorado, Westinghouse at Malmstrom Air Force Base in Montana, and Antares at Joint Base San Antonio in Texas.[36] Separately — and often conflated with ANPI in public accounts — the Air Force is running an older, standalone microreactor pilot at Eielson Air Force Base in Alaska, dating to a 2019 congressional mandate, with Oklo’s Aurora as its selected reactor; that pilot is not part of ANPI, though it is frequently reported alongside it.[36]

The Janus Program, launched October 14, 2025, is Army-specific and directly fulfills Executive Order 14299’s September 30, 2028 deadline for a reactor operating on a domestic military base. Nine candidate sites have been named: Fort Benning, Fort Bragg, Fort Campbell, Fort Drum, Fort Hood, Fort Wainwright, Holston Army Ammunition Plant, Joint Base Lewis-McChord, and Redstone Arsenal.[37]

Two observations from this record carry weight beyond the program details.

The first is cross-program overlap — the same finding Paper IV made about federal agencies, now visible among vendors. Oklo appears in the hyperscaler letters of intent of Section 2, the fuel chain of Section 6, and the Eielson pilot at once. Antares appears in the Reactor Pilot Program’s criticality race and in an ANPI site pairing. Westinghouse holds both the $17.5 billion civilian loan commitment of Section 1 and an ANPI pairing at Malmstrom. A small set of companies is being threaded through every federal posture simultaneously.

The second is a precision this paper commits to plainly, because the temptation to overstate is obvious and the series’ credibility depends on resisting it. In their current program definitions, these are fixed, on-base power installations. The documented record describes no combat or weapons application, and this paper describes none. It is worth noting only what the Army’s own solicitation language says: the reactors are to power “installations and nonpermanent operations.” The paper takes that phrase no further than the Army does, and draws no inference from it here.

The continuity finding arrives one more time, because the record keeps producing it: ANPI launched in June 2024, under the prior administration, a year before the executive orders usually credited with the military nuclear push. Like the ADVANCE Act, like the uranium import ban, like the hyperscaler Request for Information, the institutional motion was already under way before the administration most associated with it took office. Three administrations’ worth of the same direction is the oldest finding in this series, and the military track now exhibits it too.

SECTION 6 — THE FUEL BOTTLENECK

A reactor without fuel is a pressure vessel. The buildout documented in Sections 1 and 2 runs, at every point, through a supply chain this series’ readers will recognize: a critical material whose enrichment is dominated by a strategic rival, a bipartisan legislative response, and federal money moving at scale to build a domestic replacement.

The legislative foundation, like the ADVANCE Act, predates the current administration. The Prohibiting Russian Uranium Imports Act, signed by President Biden on May 13, 2024, bans imports of Russian low-enriched uranium through 2040 and unlocked $2.72 billion in domestic enrichment funding that had been conditioned on the ban’s passage.[30] The parallel to Paper II’s rare earth story is exact in structure: an adversary’s dominance of a processing tier, addressed not by tariff but by prohibition plus subsidized domestic replacement.

The money followed on January 5, 2026, when the Department of Energy awarded $900 million each to three companies — Centrus Energy, General Matter, and Orano — to build domestic enrichment capacity, with Centrus and General Matter tasked to produce high-assay low-enriched uranium (HALEU) and Orano to produce conventional low-enriched uranium.[31] The allocation is itself a signal worth reading. A fourth award the same day went to Global Laser Enrichment for $28 million — a developer of laser enrichment, an alternative to the centrifuge technology the three larger awards fund. The government placed roughly $2.7 billion behind centrifuge enrichment and $28 million behind the laser alternative, a ratio of nearly a hundred to one, with no public explanation of why the balance fell so heavily on one side. Whatever the reasoning, the capital reveals a decisive preference; the record does not disclose its basis.

One of the three larger recipients should be familiar: General Matter appears in Section 1’s list of companies operating under Utah’s Operation Gigawatt, meaning the same firm now holds a position in a state-level reactor buildout and a federal enrichment award simultaneously — the cross-program overlap documented in Section 5.

Centrus is the most fully documented of the three. Alongside its federal award, the company announced a $560 million expansion of centrifuge manufacturing at Oak Ridge in January 2026, has disclosed a commercial contract backlog exceeding $2 billion, and has signed a non-binding memorandum of understanding with Korea Hydro & Nuclear Power and POSCO International to explore potential investment.[32] The MOU sits on the ladder’s middle rung — a framework to cooperate, not a commitment to fund — but its direction matters: the domestic enrichment buildout is already reaching toward the same allied-nation partnership structure Paper IV documented in critical minerals.

All of this funds future capacity, and that timing exposes the sharpest fact in the fuel chain: the reactors that reached criticality this year are not running on it. Domestic commercial enrichment barely exists — Centrus, the only U.S. producer, has delivered on the order of 900 kilograms of HALEU to date, against a Department of Energy estimate that domestic demand could reach 50 metric tons per year by 2035.[31b] The fuel actually powering the buildout comes instead from a finite bridge: surplus highly-enriched uranium — weapons-grade and research-reactor material — downblended into the HALEU range. DOE is downblending surplus HEU at the Savannah River Site, recovering more from legacy EBR-II fuel at Idaho National Laboratory, and in May 2026 the National Nuclear Security Administration transferred 1.7 metric tons of surplus HALEU from Japan, the largest single international uranium shipment in its history.[31a] Each of these is described in DOE’s own materials as a bridge — a stopgap until commercial enrichment reaches scale. And the size of that bridge cannot be independently assessed, because the total surplus inventory available for downblending is classified. The visible pieces are a few tonnes here, a shipment there; the whole is not public. The buildout is running, for now, on a stockpile of downgraded weapons material whose remaining depth no outside analyst can measure.

There is a proposed way off that bridge that is not new enrichment at all: recycling spent nuclear fuel to recover usable material. Curio — one of the eight companies in Section 1’s Operation Gigawatt roster — is developing such a process. But it is early-stage: no commercial spent-fuel reprocessing facility operates in the United States, and Curio’s effort is a design, not a plant. It belongs in this section as a stated intention, not a supply. It also points at a constraint this paper does not take up: the spent fuel these reactors will eventually produce is both a disposal question and, potentially, a future fuel source — which is precisely why recycling sits at the seam between the fuel bottleneck and a waste problem that does not yet exist at scale, and will.

The demand side and the fuel side formally connected on June 18, 2026 — the same day Ward 250 reached criticality — when Oklo and Centrus signed a letter of intent for Centrus to supply HALEU for up to five of Oklo’s Aurora reactors at its planned 1.2-gigawatt Ohio campus, with deliveries beginning in 2029.[33] The companies described the agreement as addressing “one of the central constraints facing the advanced nuclear sector.” Precision requires noting what the document is: a letter of intent, not a supply contract — a letter-of-intent instrument, the same category as the seven utility letters of intent behind the Westinghouse loan, real as a signal and unenforceable as a commitment. The connection it describes is real; the obligation it creates is not.

The pattern across this section is the one the whole series has traced. An adversary controls a processing tier. The government responds with prohibition and subsidy, bipartisan across administrations. Domestic capacity gets funded at scale, reaching toward allied partners as it grows. And the instruments doing the reaching — an MOU here, a letter of intent there — are, on inspection, softer than the headlines announcing them, while the fuel actually in the reactors comes from a stockpile no one outside government can size. This is why the paper tags each instrument to its rung rather than counting them all as done, and why “supply secured” is not the same as fuel in a core.

SECTION 7 — THE WATER PROBLEM

Every reactor design in this paper solves the same engineering problem differently and arrives at the same downstream requirement regardless. Sodium, helium, or ordinary water can carry heat out of a reactor core. But once that heat is extracted, every design converts it to electricity the same way every thermal plant has since the nineteenth century: by boiling water into steam and driving a turbine with it. The exotic coolant changes how heat moves from the core to the water; it does not change the water-intensive steam cycle that follows. A sodium-cooled fast reactor and a helium-cooled high-temperature reactor both still need to reject the steam’s heat afterward — and rejecting that heat is where the water is spent. The only way to escape it is to reject the heat to air instead, through an air-cooled condenser, which is the exception this section ends on, not the rule it begins with.

The industry’s engineering response, audited claim by claim in Paper III, resolved into one fully creditable standard and a set of softer ones. TSMC’s first Arizona fab draws 4.75 million gallons a day and recycles 65 percent through its existing water resource center. Its dedicated reclamation plant — planned before the fab opened, but broken ground only in August 2025 and not operational until 2028 — is designed to reach a 90 percent recycling goal aimed at the hardest version of the problem: the ultrapure water that touches the wafer, not merely the water that cools the building. The sequence is worth stating exactly, because it is the honest version of the claim. TSMC did not solve its water problem before switching on the fab. The fab has drawn water at 65 percent recycling since late 2024; the plant that reaches 90 percent will not exist until 2028. What TSMC did was commit to the hardest version of the problem, state plainly what it had built and what it had not, name the date, and start building against demonstrated precedent in its own Taiwan operations. Measured against that standard, the other claims held up less well. Intel’s Arizona “net positive” status is real but is an end-of-pipe watershed return, not core-loop recycling. Sherman, Texas’s new wastewater plant is genuine, completed, citywide infrastructure that returns treated water to a creek — not a fab-specific system. GlobalWafers’ federally tracked 50 percent commitment is qualified by “commercially reasonable efforts”; Texas Instruments’ 70 percent aspiration carries no federal tracking and sits far above the company’s own disclosed 27-to-29-percent record.[40]

That corrected restatement matters here because the comparison holds — it just holds more precisely. The semiconductor industry saw its water constraint coming, and its most advanced operator committed capital to the hardest version of the problem while its fab was already running, and said plainly what it had and had not yet built. The nuclear buildout, with limited exceptions, has not done the same — and the exceptions are instructive precisely because of how few and how partial they are.

Start with what the criticality milestones did and did not prove. The reactors that reached criticality this year demonstrated core physics at zero or near-zero power; none ran the steam cycle that consumes water, because that cycle only operates at power, which none of them has yet sustained. The water-consuming systems of these designs are, as of this writing, undemonstrated — not because anyone has hidden them, but because the tests conducted so far do not reach them. The part of the machine that determines its water footprint has not yet been switched on.

Two projects have engaged the water constraint directly, by opposite routes, and both are worth stating exactly. The first is Blue Castle, near Green River, Utah — the one project that has visibly changed course because of water. Its original design, two large AP1000 reactors, depended on water rights the company secured around 2012 and then lost amid a mix of financing trouble, regulatory delay, and unresolved litigation. Its 2026 revival did not restore that position; it moved toward Holtec SMR-300 units, whose available air-cooled condenser configuration is marketed for exactly this kind of arid siting. The redesign identifies air cooling as its intended answer to the water problem — though, per the most recent local reporting, the project has not confirmed that the air-cooled configuration is finalized. It is the clearest case in this paper of the water constraint forcing a change in reactor technology, and even it is not yet settled.

The second is TerraPower’s Natrium plant at Kemmerer, Wyoming, and it engages the constraint the other way — by inheriting existing infrastructure rather than redesigning around it. Natrium is sited beside the Naughton generating station, and its developers cite the existing plant’s cooling-water intakes among the reasons for the location. Public materials frequently describe Naughton as a “retiring coal plant,” which is accurate as to coal and incomplete as to the plant: Naughton’s coal units ceased operation at the end of 2025, but the station is converting to natural gas and continuing to run. Natrium is therefore not stepping into a departing plant’s freed-up water; it is adding a new reactor’s steam-cycle demand alongside a fossil plant that keeps drawing. Its stated water plan reaches no further than the shared existing intakes — which is thinner than it sounds, because those intakes still serve an operating gas plant.

Everything else in the new-build track is quieter still. The hyperscaler agreements of Section 2 overwhelmingly attach to existing or restarting plants — Susquehanna, Clinton, Crane — whose water footprints were permitted decades ago and are already sunk into their watersheds. The water question therefore concentrates almost entirely on the genuinely new build: the ten Westinghouse AP1000s at five undisclosed sites, the Kairos fleet Google’s agreement contemplates, and the military microreactors of Section 5, whose cooling architectures are the least publicly specified systems in this entire paper. The track where the water question is unanswered and the track where disclosure is thinnest are the same track. And several of these plants use the same conventional steam-cycle generation in some of the same water-stressed basins Paper III already documented absorbing demand from semiconductor fabrication — the Colorado River, on which Utah’s projects draw, has been in measured, sustained decline for over two decades, independent of anything in this buildout.[29-W]

The semiconductor industry’s best operator did not wait for its water constraint to become a crisis. It named the gap while its fab was already running, committed capital to closing it, and said publicly how far it had and had not gotten. This buildout, with one redesign and one inherited intake, has not yet shown that it has done the same — and the plants where the question is hardest are the ones about which the least has been said.

CONCLUSION

This paper set out, in its first version, to determine whether the speed and structure of the federal government’s nuclear buildout matched the pattern this series identified in critical minerals. It does. Three weeks of continued research have not altered that finding; they have supplied the evidence the original publication date could not yet capture, at a rate that is itself part of the finding.

The demand side of the bet now has names. Four hyperscalers have signed binding, long-term agreements against named plants — restarted, existing, and not yet built — with a federal utility sitting inside one deal as the offtaker and a federal loan inside another. Below them, softer instruments carry larger numbers, which is why this paper rated every commitment by the enforceability of the instrument behind it rather than the size of its headline figure: the ladder runs from a signed contract down to a self-reported pipeline, and in this buildout the gigawatts grow as the commitments soften.

The fuel side has a bipartisan prohibition, billions in unlocked and awarded funding, and a lopsided bet — roughly a hundred to one — on proven centrifuge enrichment over the laser alternative, placed without public explanation. And beneath the new commercial capacity being built sits a quieter fact: the reactors that reached criticality this year are running not on that capacity but on a finite bridge of downblended surplus weapons and research uranium, whose remaining depth is classified and cannot be measured from outside.

The military side is not one airlift but three parallel programs, with statutory deadlines and a contracting model — the government as guaranteed customer — distinct enough from deregulation and equity-adjacent lending that this version names it a third federal posture in its own right. Taken together, the three postures let the federal government stand in every position around this buildout at once: the regulator stepping aside, the lender-shareholder stepping in, and the customer waiting at the end of the line.

The institutional record still runs continuously across the handoff between administrations, and the new material deepens that continuity rather than complicating it: the ADVANCE Act, the uranium import ban, the hyperscaler Request for Information, and the ANPI military program all predate the executive orders usually credited with this buildout. Three administrations, one direction — the oldest constant in this series, now confirmed in a fourth sector.

And the disclosure gap has widened in proportion to the activity. Which five sites will receive the $17.5 billion is not public. What the hyperscalers are paying, in most cases, is not public. The military’s final vendor and site selections are pending. The depth of the fuel stockpile now powering the reactors is classified. And the water requirement of every genuinely new-build reactor in this paper — beyond one project that redesigned toward air cooling and one that inherited a fossil plant’s intakes — goes unaddressed in any document this paper has reviewed. That last gap is the sharpest, because the hyperscaler deals overwhelmingly attach to existing or restarting plants whose water footprints were settled decades ago. The water question concentrates almost entirely on the new-build track — the same track where disclosure is thinnest.

None of the material added in this version contradicts the paper it revises. All of it confirms the original thesis at a scale and speed the first publication could not capture. Everything documented here is a floor, not a ceiling — and the rate at which new, dated, independently sourced material surfaced in three weeks is itself evidence for this paper’s core argument: pace outrunning visibility.

REFERENCES

Sources are primary wherever available — SEC filings, the Federal Register, agency press releases and program records, and company disclosures. Where a claim could only be reached through secondary reporting, that is noted in the entry.

[1] ADVANCE Act of 2024, Public Law 118-67, signed July 9, 2024; U.S. Senate Committee on Environment and Public Works, “Signed: Bipartisan ADVANCE Act to Boost Nuclear Energy Now Law,” July 9, 2024. (Senate 88-2; House 393-13.)

[2] U.S. Department of Energy, “U.S. Sets Targets to Triple Nuclear Energy Capacity by 2050” and the U.S. Nuclear Energy Deployment Framework, energy.gov, November 12, 2024; DOE, “Pathways to Commercial Liftoff: Advanced Nuclear” (~100 GW in 2024 to ~300 GW by 2050).

[3] Executive Order 14302, “Reinvigorating the Nuclear Industrial Base,” May 23, 2025, Federal Register.

[4] U.S. Department of Energy, “Department of Energy Celebrates First Advanced Reactor Achieving Criticality” (Antares Nuclear, Mark-0), energy.gov, June 4, 2026.

[5] U.S. Department of Energy, “Department of Energy Celebrates Second Advanced Reactor Achieving Criticality” (Valar Atomics, Ward 250), energy.gov, June 18, 2026.

[6] Executive Order 14301, “Reforming Nuclear Reactor Testing at the Department of Energy,” May 23, 2025, 90 Fed. Reg. 22591.

[7] Westinghouse Electric Company; Brookfield Asset Management; and U.S. Department of Energy, Office of Energy Dominance Financing, conditional loan commitment announcement, June 23, 2026; Cameco Corporation disclosure of the American Nuclear Supply Chain Loans structure (Westinghouse SPV; up to five project funding vehicles; $500 million equity per project), June 2026.

[7a] Cameco Corporation, Form 40-F (FY2025), U.S. Securities and Exchange Commission; Brookfield Asset Management, Form 8-K, Exhibit 99.1, October 28, 2025; “United States Government, Brookfield and Cameco Announce Transformational Partnership,” joint press release, October 28, 2025 (the $80 billion partnership; 20% of distributions above $17.5 billion; IPO/equity terms).

[9] Executive Order 14299, “Deploying Advanced Nuclear Reactor Technologies for National Security,” May 23, 2025, 90 Fed. Reg. 22581.

[10] Executive Order 14300, “Ordering the Reform of the Nuclear Regulatory Commission,” May 23, 2025, 90 Fed. Reg. 22587 (400 GW by 2050; 18-month/12-month licensing deadlines; the 133/two-reactor figure appears in the order’s preamble).

[11] Valar Atomics, technical materials and public statements, valaratomics.com, 2026 (“Cold ≠ Hot” framing).

[12] U.S. Air Force / DVIDS, Operation Windlord release, February 15, 2026 (multi-wing airlift, March ARB to Hill AFB); The Wall Street Journal, reporting that Valar paid the cost of the flight (estimated under $1 million), as cited in Partnership for Global Security and The National Interest, 2026. Marked WSJ-reported-via-secondary.

[13a] U.S. Department of Energy, statement on third advanced reactor criticality (Deployable Energy, Unity, Nuclear Energy Launch Pad), energy.gov, July 1, 2026; Deployable Energy, criticality announcement, July 1, 2026.

[13b] U.S. Department of Energy and Aalo Atomics, criticality announcement (Critical Test Reactor, July 4, 2026), energy.gov, July 2026.

[14] Valar Atomics funding disclosures, 2025-2026 (~$130 million raised toward Ward 250, per company statements and contemporaneous reporting).

[15] U.S. Nuclear Regulatory Commission, press release No. 26-028, “NRC Issues First Commercial Reactor Construction Permit in Nearly a Decade” (TerraPower Natrium), March 2026; U.S. Department of Energy, “NRC Issues Construction Permit for TerraPower’s Natrium Advanced Reactor.”

[16] State of Utah, Office of Governor Spencer Cox, “Operation Gigawatt” initiative materials, October 8, 2024 and 2025-2026.

[17] State of Utah, Office of Governor Cox, tri-state (Utah-Idaho-Wyoming) nuclear coordination agreement, April 2025; Cowboy State Daily, coverage of Energy Secretary Chris Wright remarks, December 2025.

[18] Santee Cooper, V.C. Summer site status statements, October 2025 (board approval to restart with Brookfield); American Nuclear Society reporting, 2025-2026.

[19] Elementl Power, southeast Ohio project announcement, June 2026; American Nuclear Society reporting.

[20] Kairos Power, Hermes reactor construction status, Oak Ridge, Tennessee; U.S. Department of Energy Advanced Reactor Demonstration Program records.

[21] U.S. Department of Energy, Reactor Pilot Program materials, 2025-2026; Nuclear Innovation Alliance analysis of the DOE 45-day safety-document review standard and embedded NRC staff.

[22] Brookfield Corporation and Cameco Corporation, Westinghouse Electric Company ownership disclosures, public filings.

[23] U.S. Nuclear Regulatory Commission, press release No. 26-035, “NRC Approves Final Rule for Advanced Reactor Licensing” (Part 53), March 25, 2026; Part 53 final rule, 91 Fed. Reg. 15696, published March 30, 2026 (effective April 29, 2026).

[24] Constellation Energy, Crane Clean Energy Center (Three Mile Island Unit 1) 20-year PPA with Microsoft, press release, September 2024, and Constellation Form 10-K, U.S. Securities and Exchange Commission (835 MW; 2028 in-service estimate); U.S. Department of Energy, Energy Dominance Financing $1 billion loan (closed November 18, 2025), energy.gov (restart accelerated to 2027 per subsequent reporting).

[25] Constellation Energy, Clinton Clean Energy Center 20-year PPA with Meta, press release, June 3, 2025 (1,121 MW; begins June 2027).

[26] Talen Energy, Form 8-K, U.S. Securities and Exchange Commission, June 2025 (1,920 MW Susquehanna PPA, ~17 years to 2042); Amazon/Talen $650 million data-center campus acquisition, 2024.

[27] Kairos Power, Google Master Plant Development Agreement, October 14, 2024 (up to 500 MW by 2035); Kairos Power / Tennessee Valley Authority / Google follow-on agreement, August 2025 (TVA as contracted buyer, Hermes 2).

[28] Oklo Inc., Form 8-K and shareholder materials, U.S. Securities and Exchange Commission, 2025-2026 (non-binding letters of intent: Equinix 500 MW with $25 million pre-payment; Prometheus Hyperscale 100 MW).

[29] Nucor Corporation, joint Request for Information with Google and Microsoft for advanced clean-firm power (nuclear, next-generation geothermal, clean hydrogen, long-duration storage), press release, March 19, 2024.

[29-W] U.S. Bureau of Reclamation, Colorado River operating-condition news releases and 24-Month Studies, usbr.gov, 2025-2026 (two decades of drought; Lake Powell ~25% full; Level 1 Shortage; Post-2026 Operational Guidelines Draft EIS reframing the crisis as a fundamental hydrologic shift).

[30] Prohibiting Russian Uranium Imports Act, signed May 13, 2024 (ban through 2040); associated $2.72 billion domestic enrichment appropriation, U.S. Department of Energy; U.S. Department of State.

[31] U.S. Department of Energy, enrichment capacity awards of $900 million each to Centrus Energy, General Matter, and Orano, January 5, 2026, energy.gov; Centrus Energy Form 8-K, U.S. Securities and Exchange Commission (Global Laser Enrichment received a separate $28 million award the same day).

[31a] U.S. Department of Energy, HALEU Availability Program materials and Savannah River Site / EBR-II downblending descriptions, energy.gov; U.S. Department of Energy, National Nuclear Security Administration, “U.S. Secures Largest-Ever HALEU Shipment” (1.7 metric tons transferred from Japan), May 2026.

[31b] U.S. Department of Energy, Office of Nuclear Energy, “What Is High-Assay Low-Enriched Uranium (HALEU)?” and “HALEU Frequently Asked Questions,” energy.gov (domestic HALEU demand could reach 50 metric tons per year by 2035).

[32] Centrus Energy, press releases: $560 million Oak Ridge centrifuge manufacturing expansion (January 23, 2026); commercial contract backlog disclosures (exceeding $2 billion); non-binding memorandum of understanding with Korea Hydro & Nuclear Power and POSCO International, August 25, 2025.

[33] Centrus Energy, Form 8-K, U.S. Securities and Exchange Commission, June 18, 2026; Oklo Inc., joint announcement (letter of intent, HALEU for up to five Aurora reactors, 1.2-gigawatt Ohio campus, deliveries beginning 2029).

[35] U.S. Department of Defense, Strategic Capabilities Office, Project Pele program materials; BWX Technologies (BWXT) integrator role; TRISO fuel delivery and 2028 operational target.

[36] U.S. Department of the Air Force, Advanced Nuclear Power for Installations (ANPI) program: eight-company eligibility (April 2025) and April 2026 vendor-site pairings (Radiant/Buckley SFB; Westinghouse/Malmstrom AFB; Antares/Joint Base San Antonio); World Nuclear News and Partnership for Global Security reporting. The Eielson AFB microreactor pilot (Oklo Aurora), dating to a 2019 NDAA mandate, is a separate standalone program, not part of ANPI.

[37] U.S. Army, Janus Program announcement, army.mil, October 14, 2025; nine candidate sites (Fort Benning, Fort Bragg, Fort Campbell, Fort Drum, Fort Hood, Fort Wainwright, Holston Army Ammunition Plant, Joint Base Lewis-McChord, Redstone Arsenal); Army solicitation language (“installations and nonpermanent operations”).

[38] U.S. Army, Janus Program materials, army.mil (commercially owned and operated (COCO) structure, Other Transaction Authority, milestone-based contracting explicitly modeled on NASA’s Commercial Orbital Transportation Services program); Partnership for Global Security on the ANPI COCO/PPA structure.

[40] Blue Collar Analytics, “Where Fortress America Lands,” Fortress America Series, Paper III, v1.2, Section 4, and its underlying primary sources: TSMC Arizona sustainability disclosures, tsmc.com (4.75 MGD; 65% recycling; IRWP August 2025 groundbreak, 90% goal by 2028); Intel newsroom (Arizona “net positive” water); City of Sherman, Texas, Post Oak Wastewater Treatment Plant (16 MGD citywide); National Institute of Standards and Technology CHIPS award pages (GlobalWafers 50% “commercially reasonable efforts”; Texas Instruments); Texas Instruments corporate disclosures (70% aspiration; 27-29% recorded).

[41] U.S. Nuclear Regulatory Commission, Part 53 final rule, 91 Fed. Reg. 15696 (effective April 29, 2026); NRC advanced-reactor hourly fee reduction from $318 to $148, effective October 2025, implementing ADVANCE Act Section 201.

FORTRESS AMERICA SERIES

Paper I: Fortress America — Hemispheric Consolidation, Infrastructure Cascades, and the Regional Transformation of the American Economy

Paper II: If It Can Work, It Can Fight — The Case for Domestic Humanoid Robotics Manufacturing

Paper III: Where Fortress America Lands — The Southern Spine and the Race to Build It

Paper IV: The Government Stake — Equity, Speed, and the Limits of Disclosure

Paper V: The Atom and the Chip — The Rush to Nuclear, and the Water Left Behind

This paper represents independent analytical and systems research and is the fifth paper in the Blue Collar Analytics Fortress America series. Nothing contained herein constitutes financial advice, investment recommendations, or legal counsel. Readers should conduct their own due diligence and consult qualified professionals before making investment decisions.