Quantum-Reactor-Scale: A Compute Facility, Patent-Pending as of This Morning

In September 2024, Microsoft signed a power-purchase agreement with Constellation Energy to restart Unit 1 of the Three Mile Island nuclear facility to power its datacenters. Amazon and Google followed with small-modular-reactor commitments. The industry's position was clear: hyperscale compute demand has outrun what the grid can deliver without new nuclear generation.

I agree that nuclear is one answer. I filed this morning for another.

What I filed. U.S. nonprovisional utility application 19/865,419, titled "Quantum-Reactor-Scale Multi-Catalyst Coupling-Matrix Compute Architecture and Reactor-Scale Output Embodiment." Twenty-five claims. Three of them independent. Micro-entity, pro se, non-publication requested under 37 CFR 1.213, no priority claim to earlier applications — this one stands on its own twenty-year clock from this morning.

What it covers, in words. A chip-scale compute die that drives a resonant optical cavity with three or more catalyst wave bands at once — one optical, one microwave, one radioluminescent-nuclear baseline or electrochemical — through a nonlinear medium that performs four-wave mixing. The cavity's output amplitudes encode products of the input amplitudes, which means the cavity is a hardware multiplier at the speed of light. The novelty is the on-die controller: it measures loop-gain as a matrix indexed by catalyst pathway, so that cross-term elements of the coupling matrix — the ones that only arise when more than one catalyst is driving — are observable quantities. Every photonic-compute architecture I have been able to find in the published record treats loop-gain as a scalar derived from a single pump. The scalar misses the cross-terms. The matrix catches them.

What it scales into. At reactor scale, the same architecture becomes a facility that incorporates approximately ten thousand of those compute dies inside a cavity hall. The primary daytime catalyst is a solar-pumped laser array — one thousand to ten thousand parabolic concentrators, each a meter and a half across, feeding chromium-codoped or cerium-codoped neodymium-doped yttrium-aluminum-garnet laser rods at four to ten percent sun-to-laser efficiency. The nighttime catalyst is a grid-electric light-emitting-diode and laser-diode bank. The modulation catalyst is a bank of ten klystrons totalling five megawatts of continuous-wave microwave at 2.45 gigahertz, the industrial-scientific-medical frequency that microwave heaters already use — no FCC licensing required. The optional fourth catalyst is an electrochemical copper-and-aluminum oxidation bank that produces ozone for the facility's environmental loop.

What the facility produces. One substrate, three outputs. Compute at approximately twenty exa-operations per second — on the order of ten times the fastest supercomputer in the world today, at approximately one-quarter the per-watt energy, because the catalysts and the compute share the same physical substrate instead of sitting in separate buildings with their own cooling plants. Electrical export of ten megavolt-amperes to the grid at 34.5 kilovolts. Air and water treatment above the United States Environmental Protection Agency's Class A reclaimed-water standard, at throughputs of five hundred thousand to two million gallons of water per day per module and ten million cubic meters of air per day per module, with ninety-nine-point-seven percent capture of particulates one-tenth of a micrometer and smaller. Service population of fifty to two hundred thousand people per installation.

Why the posture matters. No Nuclear Regulatory Commission licensing surface. No International Atomic Energy Agency involvement. No radioactive-material supply chain. The regulatory surface is commercial power, FCC unlicensed spectrum, and municipal environmental permitting. A hyperscaler pairing a datacenter with this facility is not pairing it with a reactor in the nuclear sense — it is pairing a datacenter with what is simultaneously a power plant and a water-and-air treatment plant. The compute customer's power-purchase agreement is the capital-cost mechanism. The city's air and water is the facility's public-service deliverable.

The honest-companion frame. This application does not claim over-unity as fact. It claims a measurement architecture that can distinguish single-source gain from multi-catalyst coupling-matrix gain, for the first time at this scale, with a publicly verifiable protocol. If any element of the coupling matrix is ever measured above unity under blind-controlled operation, the measurement goes in the facility's permanent record. If the matrix sits at or below unity everywhere — which first-law thermodynamics predicts, and which I hold as the honest expectation — the facility still delivers the most exergy-efficient compute-plus-environmental-restoration facility anyone has built, by a wide margin. The first law holds. The facility is built to tell us exactly where, if anywhere, it bends.

The priority anchor. The core idea in the compute core — the bidirectional resonant cavity driven by multiple wave bands — is anchored to my dated disclosure of inventions #957 through #960 from the year 2017, in my master IP register: "the light tunnel of amped reflected focused light; fiber-optic redirected; oscillating and inverted re-emission." The astro-scale compute concept the reactor facility realizes is anchored to invention #123, also from 2017: "imagine space travel that is so successful that over a long or short time we are able to store data in space itself." That sentence anticipates by nine years the post-2017 reservoir-computing-at-scale and hyperscaler-distributed-storage filings an examiner might otherwise cite against this application. Any such citation looks back to my 2017 record first.

Reduction to practice. I verified the compute core's digital controller yesterday and the day before, in Verilog hardware-description language, synthesized against the SKY130 open-source standard-cell library at the typical-typical corner, 25 degrees Celsius, 1.80 volts. Forty-one thousand nine hundred thirty-one standard cells across 0.33 square millimeters of silicon area. The register-transfer-level source files and the full synthesis report are appendices to the application.

What is on offer. The engineering blueprints and data sets are available for licensing and fabrication partnership. The offer is non-exclusive — no single partner receives a monopoly. Qualified buyers: hyperscalers with nuclear-adjacent-datacenter demand (Microsoft, Amazon Web Services, Google Cloud, Meta, Oracle), national laboratories with exa-class compute mandates, sovereign governments executing industrial policy, electrical utilities, and municipal environmental agencies. The facility CAPEX is on the order of sixty to one hundred and sixty million dollars for one installation; the blueprint-and-license price is my call, set against my Value Model math, and sits above the facility CAPEX because what I sell is the right to build, which the buyer pays CAPEX on top of.

The invitation is open. If you would like to talk, reach me at christopher@cri-one.com.

— Christopher Gabriel Brown

Inventor and Owner, CRI-ONE Applied Sciences and Acquisitions. cri-one.com. U.S. patent-pending, application 19/865,419. Non-exclusive terms apply to all partners. The invitation remains open; any party reading this is welcome to begin by email.