Marin Ivezic
In September 2025, Bluefors, the Finnish manufacturer that builds most of the world’s dilution refrigerators, agreed to buy helium from the Moon. The isotope is helium-3, the contract is reported at more than $300 million, and it commits a space-resources company called Interlune to deliver up to 10,000 liters a year between 2028 and 2037.
The commentary that followed split three ways, and each of the three gets something wrong. One group treated the whole thing as theater, on the grounds that helium is helium and ordinary helium-4 is cheap and plentiful. A second group pointed at magnetic cooling and declared the deal obsolete before it was signed. A third treated helium-3 scarcity as a threat to quantum computing as a whole.
We use this case in our hardware teaching because it cannot be settled by reading press releases. The physics decides whether a substitute exists, the modality decides who is exposed, and the arithmetic decides whether the demand forecasts hold together. Work through those three in order and the answer comes out narrower than any of the confident positions: helium-3 is irreplaceable for one slice of quantum hardware, the supply is genuinely tight, and the numbers used to justify mining it off-world do not survive a unit conversion.
Helium-4 cannot do helium-3’s job
Isotopes are atoms of the same element carrying different numbers of neutrons. Helium-3 and helium-4 are chemically identical, so it’s tempting to treat them as two grades of one commodity. Within a few degrees of absolute zero they behave like different substances governed by different rules.
The reason is quantum statistics. A helium-4 atom holds an even number of constituent particles, two protons, two neutrons, and two electrons, and behaves as a boson, a type of particle that will happily pile into the same quantum state as its neighbors. A helium-3 atom is one neutron lighter, carries half-integer spin, and behaves as a fermion, which is forbidden from sharing a state. That one difference sets what each isotope can do near absolute zero.
Cooling with helium-4 alone runs out of road early. The liquid boils at 4.2 kelvin at atmospheric pressure, and pumping away the vapor to lower the pressure gets you to roughly 1 kelvin before the cooling stalls. Pumping on pure helium-3 does better, reaching about 0.3 kelvin in a single-shot system, because helium-3 holds a higher vapor pressure at low temperature. Neither reaches the range that superconducting and spin qubits require, which is 10 to 20 millikelvin, or thousandths of a kelvin, a factor of fifty or more colder.
A dilution refrigerator closes that gap using a property helium-4 does not have. Below about 870 millikelvin, a mixture of the two isotopes separates into two layers: a helium-3-rich concentrated phase floating on a helium-3-poor dilute phase. The useful detail is that the dilute layer never empties. Even as the temperature approaches absolute zero it still holds about 6.6 percent helium-3. Pushing helium-3 atoms across the boundary from the concentrated side to the dilute side absorbs heat, much as evaporation does, and because there is always room on the dilute side, the cooling does not stop. The machine holds millikelvin temperatures continuously, for months at a time.
Helium-4 contributes nothing to that cycle. Below 2.17 kelvin it becomes an inert superfluid background. Remove the helium-3 and a dilution refrigerator stops being a dilution refrigerator. The two isotopes are not competing products, they are two rungs on the same ladder, and only the top rung has no substitute.
Where the scarcity comes from
Helium-3 is a trace component of terrestrial helium. The atmospheric ratio of helium-3 to helium-4, the reference value geochemists write as Ra, is about 1.4 parts per million, far too dilute to separate economically from the helium-4 supply. Almost every gram in circulation comes from one artificial source: the decay of tritium.
Tritium is a heavy isotope of hydrogen with a half-life of 12.3 years, and when it decays it becomes helium-3. The world’s tritium was not made for civilian purposes. It was produced for thermonuclear weapons, and the helium-3 supply is a byproduct of maintaining and dismantling those stockpiles. In the United States, weapons tritium production stopped in 1988, when the last Savannah River production reactor shut down. The National Nuclear Security Administration resumed it at limited scale in 2007, irradiating lithium-bearing rods in the Tennessee Valley Authority’s Watts Bar Unit 1 reactor and extracting the gas at the Tritium Extraction Facility at Savannah River. Canada runs the other significant Western civilian source, where Ontario Power Generation’s subsidiary Laurentis Energy Partners recovers helium-3 from the tritium that builds up in the heavy water of the CANDU reactors at Darlington and is stripped out at the station’s tritium removal facility. Russian supply, historically substantial, has been effectively out of Western markets since 2022.
The Lowy Institute’s 2025 synthesis, drawing on Edelgas Group data, puts global terrestrial production at 22,000 to 30,000 liters per year against demand of 40,000 to 60,000 liters from quantum computing, medical imaging, neutron detection, and fusion research. The gap is covered by drawing down stockpiles, which works until the stockpiles are gone.
What makes this structurally awkward is that production cannot be accelerated. Helium-3 appears at whatever rate tritium decays, about 5.6 percent of any holding per year, and no engineering improvement touches a decay constant. Holding a larger tritium reserve does not help this year; you collect only what this year’s decay yields. There is also a competing claim on the same tritium, since tritium is the primary fuel for deuterium-tritium fusion, and every gram burned in a reactor is a gram that never decays into helium-3.
Quantum computing is not the main buyer, either. Helium-3 is the best neutron-detection material known, which made it the material of choice for the radiation portal monitors deployed in bulk after 2001 to scan cargo for smuggled nuclear material. Neutron detection and safeguards still dominate consumption. A 2011 US Government Accountability Office review, Managing Critical Isotopes: Weaknesses in DOE’s Management of Helium-3 Delayed the Federal Response to a Critical Supply Shortage (GAO-11-472), found that weak federal management of the isotope had slowed the response once the shortage became apparent, and that episode is why low-temperature physicists have been nervous about their supply lines ever since.
Which quantum computers actually need it
This is the point most helium-3 commentary skips, and it does the most damage to the forecasts. Dilution refrigeration is mandatory for some quantum computing approaches and irrelevant to others.
Superconducting and spin qubits
Superconducting processors, the kind built by Google and IBM, and silicon spin-qubit machines pursued by Intel and several startups, have qubit transition frequencies in the gigahertz range. Keeping those qubits in their ground state means holding the surrounding environment colder than the qubit’s own energy scale, which forces operation near 10 millikelvin. That forces a dilution refrigerator, which requires helium-3. There is no way around it at present.
Trapped ions and neutral atoms
These machines hold their qubits in electromagnetic or optical traps and run at or near room temperature. No dilution refrigerator sits anywhere in the system, and helium-3 exposure is zero.
Photonics, and why PsiQuantum confuses the picture
Photons do not feel heat. A photonic qubit carries no thermal-noise penalty and needs no cooling to preserve coherence or entanglement, so the logic layer of a photonic quantum computer has no cryogenic requirement at all. What does need cooling is the readout. Photonic machines detect qubit states with superconducting nanowire single-photon detectors, or SNSPDs, which only work below their superconducting critical temperature, typically between 1 and 4 kelvin. That is ordinary liquid helium-4 territory. Bluefors sells a helium-4-only cryostat for exactly this use.
PsiQuantum’s Omega platform looks like a counterexample, and it gets cited as one. Omega is monolithically integrated: the detectors, the single-photon sources, and the electro-optic switches all sit on one silicon-nitride die, fabricated on a standard 300 mm CMOS line at GlobalFoundries. Because the detectors live on the chip rather than in a separate module, the whole die goes into a cryostat at a few kelvin, and from outside it reads as “photonics needs cryogenics.”
The causation runs only through the detectors. Silicon-nitride waveguides route light equally well at any temperature, the photon sources have no thermal requirement, and the barium titanate switches work at room temperature. PsiQuantum selected barium titanate because it keeps a strong Pockels effect at 4 kelvin, meaning its refractive index still responds to an applied voltage when cold, unlike most electro-optic materials. Monolithic integration turns “the detectors need 4 kelvin” into “everything on this die must tolerate 4 kelvin,” which is a qualification problem rather than a physics one. The temperature is still helium-4 range. A photonic architecture with off-chip detectors linked by fiber would not need to cool the logic at all.
Shane Mansfield, chief research officer at the photonic company Quandela, made the same distinction publicly to SpaceNews: most platforms lean on cryogenics, but the dilution refrigerators and their helium-3 belong specifically to superconducting processors. He has commercial reasons to draw the line, and the physics draws it in the same place.
A charge, not a fuel
The second correction is more consequential than the first. In a dilution refrigerator, helium-3 is not consumed. It circulates in a sealed loop, crossing the phase boundary and being pumped back around, indefinitely. Demand from cryogenics is the one-time charge that fills each new machine, plus whatever is lost to leaks and servicing. It is not an annual burn rate.
Neutron detection and fusion are different, and there the gas genuinely is used up. Treating all three the same way inflates the quantum figure badly, because it counts the same atoms every year for the life of the machine.
So cryogenic demand is bounded twice over. It applies only to the superconducting and spin-qubit share of an industry that also runs room-temperature and 4-kelvin platforms, and within that share it scales with the number of new machines built rather than with how much computing they perform. Before it is a supply question, helium-3 is a modality-selection question.
Do the unit conversion before you trust the forecast
Helium-3 gets quoted in liters of gas, liters of liquid, moles, grams, and kilograms, and the conversions are not intuitive. Doing them is the fastest way to test a demand claim.
One liter of helium-3 gas at standard temperature and pressure, the reference condition used for quoting gas volumes, weighs about 0.135 grams. Bluefors publishes a charge of roughly 40 liters for its XLD1000sl, so a large dilution refrigerator holds on the order of 5 grams of the material. A kilogram is about 7,400 liters of gas. Interlune’s chief executive, Rob Meyerson, has cited a commercial price near $20 million per kilogram, which puts the helium-3 in one large fridge at roughly $100,000, in a fluid that has to be recovered rather than vented every time the machine is serviced.
Now run the same conversion on the forecasts. A figure that circulates in the lunar-mining case puts quantum-driven demand at 300 to 400 kilograms per year. At 5 grams per machine, that would charge 60,000 to 80,000 new dilution refrigerators annually, against a global installed base in the low thousands. The number isn’t slightly optimistic, it is wrong by more than an order of magnitude, and it almost certainly imports the consumed-fuel assumption from fusion.
The same arithmetic cuts the other way on the supply side. Read as gas at standard conditions, 10,000 liters a year is around 250 fridge charges, or about 1.35 kilograms, which would be well under one percent of that 300 to 400 kilogram figure. Two numbers from the same pitch only reconcile if nearly all the projected demand is fusion rather than cryogenics. A 2010 Congressional Research Service report, The Helium-3 Shortage: Supply, Demand, and Options for Congress (R41419), warned about precisely this hazard, noting that mixing volume and mass figures from different analysts has the potential to create confusion once totals are added together. The warning has held up.
Magnetic cooling is closer than the skeptics say, and further away than the boosters claim
The serious challenge to helium-3 does not come from the Moon. It comes from a cooling method that uses no helium at all.
Adiabatic demagnetization refrigeration, or ADR, uses the magnetocaloric effect. Place a paramagnetic material in a strong magnetic field and its magnetic moments align, lowering its entropy and releasing heat to a bath. Isolate it thermally, then reduce the field slowly. The moments randomize, entropy rises, and the material draws the required heat from whatever it is touching. A 2025 review in Accounts of Chemical Research describes ADR as the only helium-free refrigeration technology able to reach below 1 kelvin, and the temperatures are not in dispute. The frustrated magnet KBaYb(BO3)2 reaches at least 22 millikelvin on demagnetization, and new refrigerant materials keep arriving.
Temperature was never the hard part. A dilution refrigerator is defined by four properties at once: how cold it goes, whether it cools continuously, how much heat it removes at that temperature, and whether it runs for months without intervention. ADR has historically been weak on the last three.
Continuity is now largely solved, and anyone still calling ADR a single-shot laboratory curiosity is working from old information. Continuous ADR, or cADR, chains several stages so that one cools while another recharges, and the German company kiutra has commercialized the approach. A recent four-stage cADR platform has been demonstrated holding continuous base temperatures below 30 millikelvin, reaching 20 millikelvin in single-shot mode, using no helium-3, and carrying the radio-frequency wiring needed to operate a five-qubit superconducting processor.
Scale is the remaining gap, and it is wide. That demonstration wired five qubits. The machines generating helium-3 demand carry hundreds to thousands of physical qubits now and are scaling further. Cooling power is the binding constraint, measured as heat removed per second at a given temperature. Bluefors’ published specifications give more than 30 microwatts at 20 millikelvin and more than 1,000 microwatts at 100 millikelvin for the XLD1000sl, and more than 3,000 microwatts at 100 millikelvin for the KIDE platform aimed at systems above 1,000 qubits. kiutra’s commercial L-Type Rapid offers continuous cooling at 300 millikelvin and single-shot operation to 100 millikelvin, with continuous operation in the 20 to 30 millikelvin band living in the SPROUT research demonstrator rather than in a shipping product.
Both fronts held together: ADR is the most credible long-term route off helium-3 and it is further along than the dismissive reading admits, and it has not yet displaced dilution refrigeration at the heat loads and qubit counts that create the demand. The engineers building these systems are candid about the motivation. One 2024 paper calls helium-3 a scarce and geopolitically problematic resource whose declining availability drives price shocks and dependencies, and the field is working its way off it.
What the lunar contract is, and what it isn’t
The case for lunar helium-3 is not foolish. Scarcity is real, prices are high, and the one segment of quantum computing that needs the gas is the segment the largest hardware builders are scaling hardest. Interlune has signed real customers, reportedly including Bluefors, Maybell Quantum, and a US Department of Energy purchase.
The weaknesses stack up on the demand side. The cryogenics figure is inflated by the modality error and the fuel error described above, and magnetic refrigeration is maturing on a comparable timeline. The demand carrying the largest projections is not cryogenics at all, it is fusion, and helium-3 fusion reactors do not exist. The flagship deuterium-tritium project, ITER, has moved its first deuterium plasma to 2034 and full deuterium-tritium operation to 2039, and deuterium-helium-3 fusion needs plasma temperatures several times higher still.
There is also a quiet circularity. Helion, the leading deuterium-helium-3 fusion developer, states on its own site that helium-3 is naturally rare and that the company plans to produce it through its own fuel cycle. The most prominent helium-3 fusion company is not a lunar helium-3 customer, and its alternative, breeding the isotope from deuterium-deuterium fusion, is itself unproven at commercial scale.
Read plainly, the lunar contract is an option. It bets that demand growth outruns both magnetic refrigeration and a terrestrial supply that could expand, wrapped around a longer-dated bet on a fusion market that has not arrived. That is a reasonable thing to buy an option on. It is not the certainty that either the announcements or the dismissals describe.
How to reason about it in your own program
Start with modality, not with supply. A superconducting or spin-qubit roadmap carries helium-3 exposure. A photonic, trapped-ion, or neutral-atom roadmap largely does not, and photonic systems with monolithic detectors are exposed to bulk liquid helium-4 and cryoplant capacity instead. That asymmetry rarely appears in technical due diligence, and it should.
Separate charges from consumption. Ask whether a stated demand figure counts machines built or gas burned. If a forecast grows with utilization rather than with installed units, it is treating a closed-loop working fluid as a fuel.
Convert everything to grams before comparing. Most helium-3 forecasts fall apart at this step, and the conversion takes a minute.
Watch continuous magnetic refrigeration, not launch schedules. cADR reaching production cooling powers for large qubit arrays would reshape helium-3 demand faster than any harvester reaches the regolith.
Treat it as a chokepoint case, not a crisis. Helium-3 is a small, sharp illustration of a broader pattern in quantum supply chains, where a strategic material sits in very few national hands. Supply today comes overwhelmingly from American and Canadian nuclear programs. That is a procurement and sovereignty question worth planning for, and it is not a reason to expect quantum computing to stall.
Where to go next
Cryogenic infrastructure sits at the point where physics, procurement, and program planning meet, and it is one of the places where a confident-sounding claim is easiest to check and easiest to get wrong. Quantum Academy’s quantum engineering programs work through cooling architectures, modality selection, and the supply-chain dependencies that follow from each, with the arithmetic done in the room rather than taken on trust. You can review the current program catalog at quantumacademy.com/.
For the longer technical treatment of the lunar contract, the fusion demand case, and the geopolitics of the tritium supply, see the full analysis on PostQuantum.com.