Earth Has 25 Kilograms of Fusion Fuel. Each Power Plant Needs 10.

Two. That is the largest number of first-generation fusion power plants humanity can build with every gram of civilian tritium on Earth. Global stockpiles sit near 25 kilograms, most of it a waste byproduct of Canadian fission reactors built in the 1970s. A single deuterium-tritium plant needs roughly 10 kilograms just to start up, before it generates a single watt for the grid. Divide the first number by the second and the industry’s fuel arithmetic turns stark: 25 divided by 10 is 2.5, which rounds down to two.

Fusion advocates like to say their fuel comes from seawater and is effectively limitless. For deuterium, that is true: about one in 5,000 hydrogen atoms in the ocean is deuterium, and it sells for roughly $13 a gram. Tritium is the other half of the D-T reaction, and almost none of it exists. It forms in trace amounts when cosmic rays strike the upper atmosphere, decays with a 12.3-year half-life, and is otherwise manufactured only as an unintended byproduct of a specific Canadian reactor design. Nineteen CANDU reactors make about half a kilogram each per year, and half of those reactors are scheduled to retire this decade.

Into this arithmetic walked Marathon Fusion, a San Francisco startup, on August 27. Its announcement: a proprietary plasma centrifuge that successfully enriched hydrogen isotopes and lithium isotopes in the same machine. One device, two bottlenecks. On the hydrogen side, the centrifuge enables “differential pumping,” selectively filtering unburned tritium out of reactor exhaust so it can be recycled, which the company says could shrink tritium flow rates and processing hardware by a factor of ten or more. On the lithium side, the same physics enriches lithium-6, the isotope breeding blankets need to manufacture fresh tritium inside the reactor. Conventional approaches treat these as entirely separate industries with entirely different equipment.

The Math Nobody Ran

Nobody in the coverage divided the fixed global stockpile by the per-plant startup inventory to ask the blunt question: how many first-generation plants can Earth actually build? Startup estimates come from serious fuel-cycle modeling. Mohamed Abdou’s team at UCLA, publishing in Nuclear Fusion in 2021, estimated 5 to 14 kilograms to start DEMO-class plants. Lawrence Livermore’s LIFE fuel-cycle design put magnetic-confinement startup needs at 8 to 60 kilograms per gigawatt. Take 10 kilograms as a mid-range working figure and run the division.

Scenario	Startup inventory per plant	Plants the 25 kg stockpile can start	Fuel bill at $30,000/g
Optimistic DEMO (Abdou low end)	5 kg	5	$150M
Mid-range DEMO	10 kg	2	$300M
Pessimistic DEMO (Abdou high end)	14 kg	1	$420M
MCF high end (LIFE)	60 kg	0	$1.8B
After ITER draws its planned 12 kg	10 kg	1	$300M
With Marathon’s 10x inventory cut	1 kg	25	$30M
Post-ITER stockpile with 10x cut	1 kg	13	$30M

Read the middle rows twice. ITER, the international research tokamak in France, plans to consume about 12 kilograms of tritium across its experimental program, somewhere between a third and half of the assured world stockpile. Subtract that draw and the remaining 13 kilograms starts exactly one mid-range commercial plant, with 3 kilograms left over. That is the entire planet’s fusion buildout: one plant. Unless breeding blankets work perfectly from day one, or unless somebody shrinks the startup inventory, the much-discussed fleet of fusion power plants is a fleet of one.

Now read the bottom rows. A tenfold reduction in tritium inventory turns the same 25 kilograms from two plant-starts into twenty-five. Even after ITER takes its share, thirteen plants can start instead of one. That is why a centrifuge test stand in San Francisco deserves more attention than the latest plasma temperature record: temperature records do not change the denominator of the only division that determines how many plants get built.

The Startup Bill, in Dollars

Tritium currently trades at $30,000 to $40,000 per gram, according to the UK Atomic Energy Authority’s fusion technology chief, and Ontario Power Generation sells its CANDU-derived supply at $25,000 to $30,000 per gram. Multiply by 10,000 grams and the fuel required to start one plant costs $300 million, spent before a single electron flows. At the pessimistic end of the DEMO range the bill is $420 million; at the LIFE high end it reaches $1.8 billion, more than some fusion startups have raised in total.

Burn economics make the picture stranger. A 1-gigawatt plant consumes tritium at roughly 55.6 kilograms per full-power year; Abdou’s modeling puts a 3-gigawatt-electric plant’s annual burn at 167 kilograms. Each gram of tritium fused releases 550 gigajoules of thermal energy, worth about $7,000 as electricity at ten cents a kilowatt-hour, according to the Bulletin of the Atomic Scientists. So every gram burned destroys $30,000 of fuel to create $7,000 of electricity. That $23,000 gap is not a rounding error; it is the reason every credible plant design assumes the breeding blanket manufactures more tritium than the plasma consumes. A breeding ratio above one is not a stretch goal. It is the entire business model, and nobody has demonstrated it at power-plant scale.

Meanwhile the stockpile is melting. Tritium’s 12.3-year half-life means the 25-kilogram inventory decays at about 5.5 percent per year, roughly 1.4 kilograms annually, comparable to Canada’s total yearly production of around 2 kilograms. ITER’s own 2018 research plan projected the global stockpile would peak before the end of this decade and then begin a steady decline as material is consumed and decays. Fusion’s fuel reserve is not just small; it is actively evaporating while the industry debates timelines.

Why One Machine Doing Two Jobs Matters

Marathon’s technical claim rests on an elegant observation about conventional centrifuges. Mechanical gas centrifuges separate isotopes by spinning rotors at extreme speed, but hydrogen isotopes move at such high thermal velocities that rotors must spin near the tearing point of carbon fiber to catch them. A plasma centrifuge has no rotor to tear: crossed electric and magnetic fields spin the charged particles themselves at supersonic speeds, and the mass difference between isotopes does the sorting. Dennis Whyte, the MIT professor whose research quantified how selective pumping changes fuel-cycle economics, called the results “impressive and important,” noting the technology could “dramatically” reduce required tritium inventory.

Corporate validation exists, within limits. Marathon is one of three private fusion companies funded through the Department of Energy’s ARPA-E VISION OPEN program, and a DOE environmental review record (CX-270890) documents federal support for the partial-ionization centrifuge work specifically aimed at tritium burn efficiency. The DOE’s Fusion Science and Technology Roadmap, published in June, named tritium processing a core challenge, writing that progress there is “critical for enabling sustained DT operation.”

Lithium is the quieter half of the announcement and possibly the nearer-term business. Natural lithium contains only about 7.6 percent lithium-6, the isotope that breeds tritium efficiently, so blankets need enriched material. A Special Competitive Studies Project report on the fusion supply chain identified lithium enrichment as the single highest-risk supply chain gap, adding that there is no domestic commercial supply. Enrichment capacity can be built with known engineering, unlike plasma physics breakthroughs, which makes it the kind of bottleneck an investor can actually price.

What This Analysis Does Not Prove

Several honest caveats apply. The $30,000-per-gram tritium price reflects gram-scale transactions; no liquid market exists for ten-kilogram purchases, and a buyer attempting one would face worse terms, longer waits, or simply no seller. Startup inventory estimates span more than an order of magnitude, from 5 to 60 kilograms, so the “two plants” figure is a mid-range illustration, not a prophecy; at the optimistic extreme the stockpile starts five plants, at the pessimistic extreme it starts none. Marathon has published no separation factors, throughput rates, or energy costs, only a press release and a test-stand photograph, and its tenfold reduction is framed as an objective (“could decrease”) with initial results described qualitatively. The fleet math also assumes the full 25 kilograms is available for plant startups, when ITER, weapons programs, and research consumers hold prior claims. Finally, breeding-blanket performance, the variable that ultimately decides whether startup inventory is a one-time loan or a recurring crisis, remains unproven at any relevant scale.

The Strongest Case Against the Tritium Panic

Skeptics of the fuel-shortage framing have a serious answer, and it deserves full strength. The tritium crunch is a bootstrap problem, and bootstraps are exactly what breeding blankets were invented to solve. Every credible D-T plant design already assumes tritium self-sufficiency, with breeding ratios modeled above one, which means the startup inventory is a one-time loan repaid by the reactor itself, not a permanent fuel bill. DEMO’s designers are attacking the startup number directly and independently: firing frozen fuel pellets deep into the burning zone to raise burn efficiency, and cutting fuel recycling time toward twenty minutes with metal-foil filters. If breeding works as modeled, the second generation of plants needs no mined tritium at all, and Marathon’s centrifuge becomes an economic optimization of a problem the industry already intends to engineer around. Valuable, but not existential. The counter boils down to this: the industry’s plan never required the stockpile to fund a fleet, only to light the first match. Whether the match lights on the first strike is the open question, and a tenfold smaller match is easier to strike.

What You Can Do With This

If you invest in or cover fusion, add one question to every diligence checklist: what is this company’s startup tritium inventory, in kilograms, and which supplier has committed to deliver it? A D-T startup that cannot name a number and a source has a timeline built on fuel it cannot buy. The credible answers will cite breeding-blanket test programs, enrichment partnerships, or inventory-reduction technology; hand-waving about future supply is a red flag measured in kilograms.

If you make policy, the actionable bottleneck is the one the SCSP report named: lithium enrichment has no domestic commercial supply, and building it requires no physics breakthroughs, only capital and permitting. That is a solvable problem on a legislative timescale, unlike plasma confinement.

For everyone else, the signal to watch is specific. Marathon’s announcement contains no separation factor and no throughput figure, and the tenfold claim lives or dies on those two numbers when they are published. A centrifuge that separates beautifully at milligrams per day is a laboratory curiosity; one that processes grams per hour at reasonable energy cost rewrites the fleet math above. Until those numbers appear, treat the 10x as a direction, not a result. And watch ITER’s tritium draw: every kilogram it consumes is a kilogram the first commercial plant cannot use, which makes the race to shrink startup inventory a race against the world’s largest fusion experiment itself.

The Bottom Line

Fusion’s hardest near-term constraint is not plasma temperature or magnet strength but a 25-kilogram global fuel stockpile that decays a little more every year. At roughly 10 kilograms per plant start, simple division says Earth can build two first-generation plants, and ITER’s planned consumption cuts that to one. Marathon Fusion’s plasma centrifuge, demonstrated August 27, attacks the denominator directly: shrink each plant’s tritium need tenfold and the same stockpile starts twenty-five plants instead of two. The company has credible backing and a genuinely clever dual-use design, but it has published no hard performance numbers yet. Fusion does not have a physics shortage. It has an inventory shortage, and inventory problems are solvable by engineers. That is the optimistic reading, and for once the math supports it.