The Sun in the Closet

For the entire history of the atom, the machine came first and the law came after — usually to say no. Fission was solved in a squash court under a football stadium in 1942, and within a single generation the law had sealed it inside cathedral-sized containment domes forever. A fission reactor in your garage is not a hard engineering problem. It is a felony.

Fusion inverted that sequence, and almost nobody noticed.

In June of 2026, a state health department — not the Nuclear Regulatory Commission, a health department — issued the first fusion power plant license in history, to a machine rising on the Columbia River, under the same regulatory family that covers hospital X-ray equipment. The NRC's proposed Part 57 framework treats fusion as byproduct-material-class technology, because its fuel is sieved from seawater rather than enriched from bomb feedstock. I traced that hinge at length in The Second Reactor, and it reduces to one line: fission is blocked by law, not physics; fusion is blocked by physics, not law.

Read that sentence again as a builder rather than as a citizen. It means the door is standing open and no product is walking through it. The law is ready before the machine exists. That has never happened before in the history of energy, and it will not stay true forever — regulatory openings close the way they opened, quietly, in a season of panic or a season of lobbying.

This is a prospectus for whoever walks through first.

The why This is the builder's half of a series. Fight Big Tech and Big Govt — Bring the Power Home named the machine raising your power bill and laid out the sovereign property that answers it. The Second Reactor asked whether the fire could ever come home, and refused to answer dishonestly. This piece asks the next question: what would it actually take, in dollars and years and hires, for somebody to build the thing.

The proof of demand

Skeptics say there is no market for a machine that does not exist. The record says otherwise, twice, at both ends of the market.

At the top: Microsoft signed a purchase agreement for fifty megawatts of fusion electricity before the machine that would generate it had ever worked. Fusion is, as far as I can tell, the only product in industrial history pre-sold at scale on unfinished physics. The buyer was not romantic. The buyer was desperate — for firm, clean, dense power that does not require a decade of transmission fights.

At the bottom: the household. The four biggest cloud companies will spend more on data centers this quarter than the entire journey in this prospectus costs end to end — and ratepayers are covering tens of billions of the resulting electric bill. That arrangement is the villain of this whole series, and it is minting your buyers by the neighborhood. Households are already answering with rooftop glass, wall chemistry, and shelf-sized compute — the loop that runs today, on physics nobody has to invent. Many of them are wiring their critical-loads panel with one spare breaker, labeled in pencil, for whatever comes next.

Your customer exists before your product does. That is the rarest thing in venture, and here it is sitting unclaimed.

The wall, audited

Now the honest part, because this series does not sell dreams without an audit.

The reason your garage does not already hum is a 1995 MIT thesis by Todd Rider, and it has held for three decades against every tabletop fusion scheme ever funded. Rider showed that plasmas held out of thermal equilibrium — the trick every small-fusion concept depends on — bleed energy through radiation and collisions faster than fusion returns it. The recirculating power eats you alive. Sixty years of hobbyist fusors prove him right in the most humiliating way available: real deuterium, real neutrons, genuine fusion, and a net energy balance about a billion times short. The world's most exotic space heater.

Which is what the title of this essay actually means. A sun in the closet is not a thing our species cannot build — it is a thing you can order. Sealed-tube neutron generators ship today, five and six figures, small enough to carry, lowered down oil wells every week of the year. Teenagers have fused deuterium on workbenches since the sixties. The closet sun already exists. What has never existed is one that pays its own electric bill.

But Rider's ledger had an assumption buried in it, and the assumption is the whole game: he charged full price for every joule of circulating energy, because in 1995 there was no efficient way to get it back. That is not a law of nature. That is a line item.

Direct electromagnetic recovery attacks exactly that line item. Helion's pulsed approach compresses a plasma with magnetic fields and then recaptures the energy as the expanding plasma pushes those fields back out — electricity in, electricity out, no steam, no turbine, recovery efficiencies a conventional thermal plant cannot touch. Avalanche's Orbitron holds ions in a desktop-scale electrostatic trap, stacking high-voltage milestones — 300 kV and climbing — toward the same question at appliance scale.

Add the enabling curve underneath. High-temperature superconducting tape already shrank net-gain fusion by an order of magnitude, from ITER's cathedral to SPARC's warehouse. The next octave of tape economics shrinks warehouse toward garage. And the fuel was never the problem — a kilowatt-class device burns grams of deuterium a year; heavy water is legal, unregulated at that scale, and costs less than printer ink; graphene isotope sieves point at a future where the fuel is refined from tap water on the counter.

So here is the feasibility claim, stated with both hands visible:

I am not telling you the minute sun works. I am telling you the sixty-year question is now answerable, for a knowable price, inside one company's lifetime.

Most likely outcome: Rider holds, and you bought the definitive answer to one of physics' great engineering questions. Other outcome: you built the last appliance.

The shape of the company

Five stages, roughly fifteen years, and only the first one bets against nature. Here is the whole road on one sheet — what each stage has to prove, what it costs, what must go right, and exactly where it hurts. Read the bottom two rows first if you are deciding whether to write a check; they are the ones most decks leave out.

Now walk it, because the dates in that top row are doing quiet work.

Ignite — 2026 to 2033, $300M to $1B. One device, Q>1, more electricity out than in. Non-thermal confinement research, high-voltage milestones past 300 kV, repetition-rate pulsing, and direct electromagnetic recovery of the circulating energy Rider charged full price for. You are buying plasma PhDs, national-lab partnerships, and DOE milestone awards, and you are buying seven years of investor patience against a wall that has eaten every challenger since 1995. The talent war is not with other startups. It is with hyperscalers who will pay your plasma control lead more than you can, to tune a recommendation engine.

Contain — 2031 to 2036, $200M to $500M. The milestone is not a physics result; it is 8,000 hours untouched. Sealed vacuum core, integrated cryocooler, fail-safe shutdown that is a property of the geometry rather than of a controller, twenty-year service intervals, kilowatt-scale output. The hard parts are neutron management and activation — what the materials become after years of bombardment — and cryogenic reliability in a machine nobody services. The governing constraint is a sentence: the appliance cannot need a physicist.

Comply — 2033 to 2038, $30M to $80M. The cheapest stage on the map, and the one this series has already started for you. Shape the Part 57 rulemaking, extend the health-department precedent from the Columbia River to other states, win the consumer-safety mark, clear export control. The failure modes are political rather than technical: rulemaking slips, proliferation optics, and the one that keeps founders awake — a single bad incident anywhere in the industry rebrands the whole category overnight. Born legal, not born banned.

Produce — 2037 to 2041, $500M to $1.5B, and from here on nothing is gated by nature. Tape at volume, cryocoolers, deuterium contracts, automated coil winding, per-unit QA, unit #10,000 at target cost. The risk is ordinary industrial risk: HTS tape supply concentrated in a handful of fabs, a cost curve that stalls instead of bending, recall exposure at scale. What must go right is that unit cost lands in generator territory — the price band of a thing an electrician installs, not the price band of an instrument.

Hearth — 2041 onward, funded out of margin. A stranger buys one at retail and an electrician installs it. Getting there means an installer network of licensed electricians, insurance products, service contracts, and co-op and fleet sales that prove reliability in public before a single family is asked to take one home. The obstacles are the ones no engineer can fix: fear marketing — a reactor by the water heater — insurance pricing with no actuarial history, utility lobbying, and the unglamorous economics of servicing a fleet.

Notice the overlaps. Contain starts in 2031, two years before Ignite is scheduled to finish; Comply opens while Contain is still running. That is deliberate, and it is the single most important structural claim on the map: the stages that are not gated by physics should be started before the stage that is. Regulatory groundwork laid in 2033 costs the same whether or not the plasma ever cooperates, and if it does cooperate, you have bought five years.

Total, end to end: single-digit billions. A rounding error on one AI campus.

The investor's ledger

Structure it the way the physics deserves — milestone tranches, cost-share where it fits, kill criteria written down on day one. The Ignite stage is a binary bet and should be priced like one, by people who can say the word no on schedule.

But notice the barbell, because it is the part most deep-tech pitches cannot honestly claim. Every stage after Ignite is money and time, not physics — and the supply chain of this company pays off in every branch of the future.

HTS tape capacity, cryocooler manufacturing, pulsed power electronics, deuterium supply, installer networks: you do not have to bet on the star to own the forge that would build it. The downside of the core bet is a tax-deductible answer to a sixty-year question. The upside is the appliance category that ends the utility bill. Power-law investing was invented for exactly this shape.

The engineer's seat

If you build pulsed power, high-voltage systems, cryogenics, vacuum hardware, power electronics, plasma controls, or the machine learning that now steers plasmas in real time — this is the Apollo-shaped hole in your career.

Some of you are optimizing ad auctions at a company whose data center raised your parents' power bill. There is a version of your résumé where you built the machine that ended that arrangement instead.

The citizen's lever

Not an engineer, not an investor? You hold the cheapest lever on the whole map, and it is the one that keeps the door open while the builders work.

The two filings in The Second Reactor — a comment on the Part 57 docket, and a letter defending the technology-neutral right to generate — cost a stamp. They are how the regulatory opening described at the top of this page survives long enough for somebody to walk through it. Incumbents would love fusion regulated like fission, because chains on the newcomer are a subsidy to the throne.

But the filing is only the first of five stations, and the household map runs on a completely different logic from the company's. The company's first move is a bet against nature. The household's first three moves are gated by nothing at all.

Read the certainty row on that second map against the first one and you find they run in opposite directions. The company starts at its least certain and climbs. The household starts at total certainty and only later reaches anything speculative.

Illuminate costs a stamp and is gated by nothing: the Part 57 comment, the state licensing precedent letter, the interconnection and microgrid docket at your public utility commission. Its only real hazard is that comment windows close quietly. Equip — roughly $50,000 to $70,000 gross before incentives — is also gated by nothing, because every line on it is a purchase order: eight to twelve kilowatts of glass, thirty to forty kilowatt-hours of wall chemistry, a critical-loads panel, and a three-thousand-dollar shelf node running open weights. Engage is the first gate, and it is a policy gate rather than a physics one: five to fifteen neighbors, a co-op LLC, an executed interconnection agreement, roughly five thousand dollars of legal work — and the cooperation of a utility that may prefer you did not.

Shift costs nothing but attention, and it is the station most people skip. It is a watch list with three signals on it: Helion actually delivering Microsoft's electrons; an Avalanche-class device beating Rider's wall; the tape price-and-temperature curve bending again. You keep the co-op shell alive and the panel capacity warm, and you wait. Sovereign is the only speculative station on the household side — a neighborhood storage ring, a co-op-scale fusion unit, deuterium measured in grams a year, and the breaker you labeled a decade earlier finally getting its wire.

Which is the honest summary of this entire series: stations one through three require zero physics breakthroughs and build the whole socket. Four and five are bets that may never pay — and if they never do, you still own a house that makes, stores and thinks with its own power.

The company climbs a certainty curve. The household waits at the bottom of it. Both maps are drawn. Only one of them requires permission from nature.

The breaker

Somewhere — in Salem, in ten thousand garages like mine — there is a panel with one spare breaker, labeled in pencil, two words.

It is not waiting for permission. The law came ready. The demand came pre-sold. The fuel is in the ocean and the tape is on the spool.

It is waiting for you.