Helium-3 and Quantum Computing’s Coldest Problem: Physics, Scarcity, and the Moon

In September 2025, Bluefors, the Finnish company that builds most of the world’s dilution refrigerators, agreed to buy helium from the Moon. The gas is helium-3, the contract is worth more than $300 million, and it commits a space-resources startup called Interlune to deliver up to 10,000 liters a year between 2028 and 2037.

The announcement set off a round of confident commentary, and most of it was wrong in one of two directions. One camp waved the whole thing away: helium is helium, the argument went, so use ordinary helium-4 and stop worrying. A second camp pointed at magnetic cooling and declared the deal obsolete before the ink dried, on the grounds that adiabatic demagnetization already reaches the right temperatures, so nobody will need lunar helium-3 anyway. A third group ran the opposite way and treated helium-3 scarcity as an existential threat to quantum computing itself.

I want to take all three apart, because the truth is more specific than any of them. Helium-3 cannot be swapped for helium-4; the two isotopes do physically different jobs, and only one of them reaches the millikelvin regime that superconducting and spin qubits require. The scarcity is genuine and the supply chain is fragile. The demand projections that justify mining the Moon, though, rest on a model of how dilution refrigerators consume helium-3 that falls apart on inspection, and the magnetic-cooling story is neither the slam dunk its boosters claim nor the irrelevance its critics assume.

The claim worth carrying into a budget meeting: helium-3 is irreplaceable for one slice of quantum hardware, its scarcity is real, and the case for mining it from the Moon is a leveraged bet on demand outpacing both magnetic refrigeration and terrestrial supply, not the certainty either camp pretends.

Helium-4 Cannot Do Helium-3’s Job

The confusion starts with the name. Helium-3 and helium-4 are isotopes of the same element, so it is tempting to treat them as interchangeable grades of one commodity, the way you might swap one octane of gasoline for another. Near absolute zero they behave like different substances under different physics.

The distinction is quantum statistics. A helium-4 atom contains an even number of fermions (two protons, two neutrons, two electrons) and behaves as a boson. A helium-3 atom is one neutron lighter, carries half-integer spin, and behaves as a fermion. That single difference decides what each isotope can do.

Cooling with helium-4 alone takes you only so far. The liquid boils at 4.2 kelvin at atmospheric pressure, and pumping on it to drop the vapor 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 approach crosses into the territory that matters. Superconducting qubits operate at roughly 10 to 20 millikelvin, a factor of fifty or more below what pumped helium delivers.

The dilution refrigerator closes that gap, and it does so through a property with no analog in helium-4. Below about 870 millikelvin, a mixture of the two isotopes separates into two phases: a helium-3-rich concentrated phase floating on a helium-3-poor dilute phase. The decisive detail is that the dilute phase never empties of helium-3. As the temperature approaches absolute zero, it still holds a finite concentration of roughly 6.6 percent. That residual solubility is the whole trick. Forcing helium-3 atoms across the boundary from the concentrated phase into the dilute phase absorbs heat, in a process analogous to evaporation, and because there is always room for more helium-3 on the dilute side, the cooling never runs out. The machine holds millikelvin temperatures continuously, for months. I detail the cooling power budgets, wiring constraints, and procurement implications that flow from this in my guide to the cryogenic infrastructure behind a quantum computer.

Helium-4 offers no equivalent. As a boson it forms an inert superfluid background below 2.17 kelvin and contributes none of the cooling. Take the helium-3 out of the mixture and the dilution refrigerator stops being a dilution refrigerator. This is why “just use helium-4” is not an argument. The two isotopes are not competing products. Helium-4 handles the upper stages of the cooling chain and helium-3 does the millikelvin work, and the second job is the one with no substitute.

The economics compound the physics. Helium-4, the gas in party balloons and MRI magnets, comes out of natural gas wells and sells for a few tens of dollars per liquid liter. Helium-3 is among the rarest stable isotopes accessible on Earth, and the reason takes us to the supply chain.

Where the Scarcity Comes From

Helium-3 barely exists in nature. Its abundance as a fraction of all terrestrial helium is about 1.37 parts per million, far too dilute to pull economically out of the helium-4 supply. Almost every gram in use comes from a single artificial source: the radioactive decay of tritium.

Tritium is a heavy isotope of hydrogen with a half-life of 12.3 years, and as it decays it turns into helium-3. The world’s tritium did not accumulate for civilian reasons. It was produced for thermonuclear weapons, and the helium-3 supply is a byproduct of maintaining and dismantling those stockpiles. In the United States, the gas is separated from weapons tritium at the Savannah River Site, and the country halted tritium production for weapons in 1988, only resuming limited output, by irradiating lithium in a commercial reactor, in 2007. Canada runs the other meaningful Western civilian source: at the Darlington station in Ontario, the Ontario Power Generation subsidiary Laurentis Energy Partners extracts helium-3 from tritium that accumulates in its CANDU heavy-water reactors. Russian supply, historically a major source, has been effectively excluded from Western markets since 2022.

The resulting supply picture is tight and worsening. The Lowy Institute’s 2025 synthesis, drawing on Edelgas Group data, estimates global terrestrial production at 22,000 to 30,000 liters per year, against demand from quantum computing, medical imaging, neutron detection, and experimental fusion research of 40,000 to 60,000 liters per year. That structural deficit of roughly 20,000 to 40,000 liters is being covered by drawing down finite stockpiles, a strategy that works until the stockpiles run out.

The underlying constraint is that helium-3 production cannot be sped up on demand. It appears at the pace tritium decays, about 5.6 percent of any stockpile per year, and no process improvement touches that decay constant. Sitting on a large tritium reserve does not help in the near term; you collect only what this year’s decay produces. There is a second claim on that tritium, too. Tritium is the prime fuel for deuterium-tritium fusion, and the fusion developers want it. Every gram reserved for a fusion reactor is a gram not left to decay into helium-3. The supply is caught between two buyers before quantum computing even enters the picture.

This thin supply collided with a demand spike after 2001. Helium-3 is the best material known for detecting neutrons, which makes it ideal for the radiation portal monitors that scan shipping containers for smuggled nuclear material, and the U.S. government installed them by the hundreds. Security and safeguards account for roughly 80 percent of global helium-3 consumption. By 2008 the country burned through about 79,000 liters in a single year, more than half its remaining stock. A Government Accountability Office review later concluded that weak federal management of the isotope had delayed the response to the shortage. That episode is why low-temperature physicists have spent fifteen years nervous about their helium-3 lines.

Price reflects all of this, though the figures need care. Historically, the U.S. auctioned helium-3 at a relatively steady $40 to $85 per liter. Current bulk market pricing has risen to $1,900 to $2,600 per liter, with purification to the grade needed for dilution refrigerators exceeding $10,000 per liter. Interlune’s chief executive, Rob Meyerson, cites a commercial price of about $20 million per kilogram, which likely reflects purified-grade economics. At the bulk end, a Bluefors XLD1000sl dilution refrigerator holds approximately 40 liters of helium-3. At $2,500 per liter, the helium-3 charge alone represents a $100,000 investment in a fluid that must be recovered, not vented, after every servicing event.

Not Every Quantum Computer Needs It

The point most helium-3 commentary misses is also the one that does the most damage to the demand projections: dilution refrigeration, and therefore helium-3, is mandatory for some quantum computing approaches and irrelevant to others.

The platforms that live at millikelvin are the superconducting processors built by Google, IBM, and others, and the silicon spin-qubit machines pursued by Intel and several startups. Their qubits have transition frequencies in the gigahertz range, and holding them in their ground state means suppressing thermal excitations until the surrounding environment sits colder than the qubit’s own energy scale. That forces operation near 10 millikelvin, which forces a dilution refrigerator, which requires helium-3.

Other modalities never enter that regime. Trapped-ion and neutral-atom machines hold their qubits in electromagnetic or optical traps and operate at or near room temperature, with no dilution refrigerator in the system at all. Photonic quantum computing is the most instructive case, and also the most misunderstood. Photons do not feel heat. They carry no thermal-noise penalty and require no cooling to maintain coherence or entanglement. A photonic qubit can be generated, manipulated, and routed through waveguides at any temperature. The logic layer of a photonic quantum computer has no cryogenic requirement whatsoever.

What does need cooling is the readout. Photonic quantum computers detect the state of their qubits using superconducting nanowire single-photon detectors (SNSPDs), which must sit below their superconducting critical temperature to function, typically in the 1 to 4 kelvin range. That is ordinary liquid-helium-4 territory (Bluefors even sells a dedicated He-4-only XLDHe system for exactly this use case). No dilution refrigerator. No helium-3.

PsiQuantum’s Omega platform is where the confusion enters, because it appears to contradict the “photonics doesn’t need cryo” claim. Omega is a monolithically integrated chip: the SNSPDs, the single-photon sources, and the electro-optic switches all sit on the same silicon-nitride die, manufactured at GlobalFoundries on a standard 300mm CMOS line. Because the detectors are embedded in the chip rather than housed in a separate module, the entire die goes into the cryostat at a few kelvin. From the outside, it looks like “the photonic chip needs cryo,” and commentators have cited it as evidence that photonics faces the same cooling problem as superconducting qubits.

It does not. The causation runs only through the detectors. The silicon-nitride waveguides route light equally well at any temperature. The heralded single-photon sources have no thermal-physics requirement. The barium titanate (BTO) electro-optic switches work at room temperature. None of these components needs cold. They are engineered to tolerate the cold that the detectors impose, which is a different problem. PsiQuantum chose BTO specifically because it retains a strong Pockels coefficient at 4 kelvin, unlike most electro-optic materials that lose their response when cooled. In a monolithic architecture, “the detectors require 4 K” cascades into “everything on the same die operates at 4 K, and every non-detector component had to be qualified for cryogenic operation.” But the temperature is helium-4 range, not millikelvin. No dilution refrigerator. No helium-3. The supply-chain exposure is to bulk liquid helium-4 and cryoplant engineering, an entirely different and far less scarce commodity.

A different photonic architecture (say, one with off-chip detectors connected by fiber) would not need to cool the photonic logic at all. Omega’s cryo requirement is a consequence of its integration strategy, not of photonic quantum computing as a modality.

This is not a fringe view. Shane Mansfield, chief research officer at the photonic quantum company Quandela, put it plainly to SpaceNews: most quantum platforms, though not all, lean heavily on cryogenics, and the dilution fridges with their helium-3 are specific to superconducting processors like Google’s and IBM’s. Mansfield works for a photonics company and has his own reasons to stress the distinction, but the physics backs him.

The deeper error in the demand projections is treating helium-3 as a fuel. In a dilution refrigerator the helium-3 is not consumed. It circulates in a sealed loop, crossing the phase boundary and being pumped back, indefinitely. Demand from cryogenics is the one-time charge needed to fill each new machine, plus whatever leaks or is lost during handling and servicing. It is not an annual burn rate. A dilution refrigerator uses a few dozen liters of helium-3, and that charge stays in the machine for its operating life. Fusion and neutron detection are different: there the helium-3 genuinely is consumed. Conflating the two inflates the quantum demand number badly, because it counts the same atoms over and over.

So the honest demand picture is bounded on two sides at once. It applies only to the superconducting and spin-qubit share of an industry that also runs room-temperature and 4-kelvin platforms, and even within that share it scales with the number of new machines built, not with how much computing they do. For a sense of where the modalities sit and why their cooling needs diverge, I keep returning to the taxonomy of quantum computing modalities. The helium-3 question is, as much as anything, a modality-selection question.

The Units Are a Mess, and So Are the Forecasts

There is a second reason to distrust the big demand numbers, and it is almost embarrassingly basic: nobody agrees on units. Helium-3 gets quoted in liters of gas at standard conditions, in liters of liquid, in moles, in grams, and in kilograms, and the conversions are not intuitive. One liter of helium-3 gas at standard temperature and pressure weighs about 0.13 grams. A dilution refrigerator’s “few dozen liters” is therefore only around 5 grams of actual material, and a kilogram of helium-3 is nearly 7,500 liters of gas.

That sounds pedantic until you watch the forecasts collapse under it. A figure that circulates in the lunar-mining pitch puts quantum-driven demand at 300 to 400 kilograms per year. At 5 grams per machine, 400 kilograms would charge roughly 80,000 new dilution refrigerators in a single year. The global installed base is in the low thousands. The number is not slightly high; it is off by more than an order of magnitude, and it almost certainly smuggles in the consumed-fuel error from fusion. A 2010 Congressional Research Service report flagged this exact hazard, warning that mixing volume and mass figures from different analysts “has the potential to create confusion” once the totals are added up. Fifteen years later the warning still holds, and the confusion still sells lunar mining.

Does Magnetic Cooling Make Helium-3 Obsolete?

The most technically 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, exploits the magnetocaloric effect. You place a paramagnetic material in a strong magnetic field, which aligns its magnetic moments and lowers its entropy, and let it dump the resulting heat to a bath. Then you thermally isolate the material and slowly remove the field. The moments randomize, entropy rises, and to supply that entropy the material pulls heat from whatever it touches. A 2025 review in Accounts of Chemical Research calls ADR the only helium-free refrigeration technology capable of reaching below 1 kelvin. The temperatures are not in question. Researchers have shown refrigerant materials reaching the low tens of millikelvin: a water-free frustrated magnet, KBaYb(BO₃)₂, reaches at least 22 millikelvin on demagnetization, and the field keeps producing new refrigerants, including rare-earth fluorides that outperform the long-standing benchmark material.

If ADR reaches millikelvin without helium-3, why is it still described as a future technology rather than a present replacement? Because temperature was never the hard part. A dilution refrigerator is defined by four things at once: how cold it gets, whether it cools continuously, how much heat it removes at that temperature, and whether it survives months of running in a real machine. ADR has historically been weak on the last three.

Continuity comes first, because this is where the critics have a real point against the old objection. Single-stage ADR is a single-shot device: it cools until the magnetic entropy is exhausted, then warms up and has to be recharged. That limitation is now largely solved. Continuous ADR, or cADR, chains multiple stages so that one cools while another regenerates, and the German company kiutra has commercialized exactly this. In early 2026 a group demonstrated a compact four-stage cADR platform that holds continuous base temperatures below 30 millikelvin, reaches 20 millikelvin in single-shot mode, uses no helium-3, and carries the radio-frequency wiring to operate a five-qubit superconducting processor. That is a genuine result, and anyone still calling ADR a single-shot lab curiosity has not been paying attention.

The gap that remains is scale, and it is wide. The cADR demonstration ran wiring for five qubits. The superconducting machines that actually drive helium-3 demand carry hundreds to thousands of physical qubits today and are scaling toward the millions a cryptographically relevant quantum computer would require, as I lay out in the CRQC Quantum Capability Framework. Cooling power is the binding constraint. A Bluefors XLD1000sl delivers over 30 µW at 20 mK and over 1,000 µW at 100 mK. The KIDE platform, designed for 1,000-plus qubits, provides over 3,000 µW at 100 mK from three cooling units. kiutra’s commercial L-Type Rapid currently offers continuous cooling at 300 mK and single-shot operation to 100 mK; continuous operation in the 20–30 mK range exists only in the SPROUT research demonstrator, not yet in a shipping product. The LEMON project is kiutra’s path to bridging that gap for full-stack quantum computers. I cover the full comparison, the numbers, and what they mean for timing in a dedicated assessment of whether magnetic cooling can replace the dilution fridge.

My read, holding both fronts: ADR is the most credible long-term path off helium-3, and it is further along than the dismissive camp admits. It has not displaced dilution refrigeration at the heat loads and qubit counts that generate helium-3 demand, and the transition, if it comes, plays out over years. Anyone who tells you ADR has already made helium-3 irrelevant is reading a lab result as a product. Anyone who tells you helium-3 dependence is permanent is ignoring a field moving quickly. The people building these systems are blunt about why: one 2024 paper describes helium-3 flatly as a “scarce and geopolitically problematic resource” whose declining availability drives price jumps and dependencies. They are engineering their way off it, and they are getting somewhere.

The Lunar Bet and the Fusion Demand That Doesn’t Exist Yet

The full story of lunar helium-3 (the physics of why it is on the Moon, how extraction would work, the contracts, the technical gauntlet, and the geopolitics) deserves and receives its own treatment in my in-depth analysis of lunar helium-3 mining. What matters for the helium-3 picture overall is where the lunar business case is strong and where it breaks down.

The bull case is not stupid. The scarcity is real, the price is high and rising, and the one segment of quantum computing that needs the gas is the segment the largest players are scaling hardest. Interlune has signed real customers: Bluefors, Maybell Quantum, and a first-of-its-kind DOE purchase.

The bear case stacks up in several places. The cryogenics demand is inflated by the modality blindness and consumed-fuel errors I flagged above. ADR is maturing on the same timeline. And the demand that actually carries the trillion-dollar pitch is not cryogenics at all: it is fusion. The hundreds-of-kilograms-per-year numbers assume helium-3 fusion reactors burning the gas as fuel. Those reactors do not exist and are not close. The flagship deuterium-tritium project, ITER, has slipped its first deuterium plasma to 2034 and full D-T operation to 2039. Deuterium-helium-3 fusion requires plasma temperatures roughly five times higher than D-T, which sits well beyond ITER’s horizon.

And there is a circularity that quietly settles the fusion case. Helion, the leading D-He-3 fusion company (backed by Sam Altman, building its Orion commercial facility), states plainly on its Polaris page: “Helium-3 is rare naturally, so Helion plans to produce it through its own fuel cycle.” The plan is to breed He-3 by running deuterium-deuterium fusion first, not to buy it from the Moon. The leading helium-3-fusion developer is not a lunar-helium-3 customer. It is worth noting that D-D fusion itself is undemonstrated at commercial scale; Helion’s Polaris prototype reached 150 million degrees Celsius with D-T fuel in January 2026, but the D-He-3 regime requires roughly 750 million degrees, and Helion has not yet published sustained energy output at those temperatures. So “terrestrial He-3 breeding from D-D fusion” is a real alternative to lunar supply, but it depends on mastering D-D fusion at scale, which nobody has done.

My verdict is neither of the easy ones. Lunar helium-3 is a leveraged option: a bet that demand growth outruns both magnetic refrigeration and a terrestrial supply that could expand, wrapped around a longer-dated option on a fusion market that does not yet exist. What it is not is the certainty either the press releases or the dismissals pretend. This is the claim-versus-evidence gap I spend a lot of time on, and it rhymes with the quantum panic industry dynamic: a phenomenon real in its physics, wrapped in projections that outrun what the data supports.

What This Means If You’re Betting on Quantum Hardware

For the readers I write for, the people allocating capital to quantum companies, running hardware roadmaps, or advising governments, helium-3 is worth understanding precisely because it is so easy to get wrong in both directions.

If you are evaluating quantum hardware investments, helium-3 exposure is a modality question before it is a supply question. A bet on superconducting or spin qubits carries helium-3 risk; a bet on photonics, trapped ions, or neutral atoms largely does not. That asymmetry rarely shows up in the diligence, and it should.

If you are tracking the supply chain itself, the variable that matters most is not the next lunar press release. It is the maturation curve of continuous magnetic refrigeration. cADR reaching production scale for large arrays would reshape helium-3 demand far faster than any harvester reaches the regolith. I cover the current state and the specific numbers in my assessment of ADR versus the dilution refrigerator.

If you care about the geopolitics, helium-3 is a small but sharp example of a larger pattern I work through in Quantum Sovereignty: the quantum supply chain runs through a handful of chokepoints, and strategic materials concentrated in a few national hands become instruments of leverage. Helium-3 today comes overwhelmingly from American and Canadian nuclear programs, with Russia historically a third source now excluded. China’s interest in lunar helium-3 is not purely about fusion; it is also about not depending on anyone else’s tritium. I trace the broader concentration risks across the full quantum supply chain.