Helium-3 Alternatives: Every Exit From the Shortage, Assessed

In September 2011, the Government Accountability Office handed Congress a report with a bureaucratic title and a quietly triumphant conclusion. Three years earlier, the United States had burned through roughly 79,000 liters of helium-3 in a single year, most of it inside radiation portal monitors scanning cargo for smuggled nuclear material, and had nearly emptied its strategic reserve of the rarest stable gas on Earth. GAO-11-753 reported the way out: boron-10 lined proportional detectors had passed field testing and, in the GAO’s estimation, could be acquired and deployed with confidence. The world’s largest consumer of helium-3 did not find more helium-3. It redesigned the detector so it no longer needed the gas at all.

Fifteen years later, quantum computing is the demand spike, and the detector trick does not transfer. A dilution refrigerator uses helium-3 not for a convenient nuclear cross-section, which boron-10 could imitate, but for its quantum statistics: helium-3 is the only stable fermionic working fluid in cryogenics, and the phase-boundary physics that cools superconducting processors to 10 millikelvin has no substitute species, as I explained in my helium-3 explainer. So instead of one clean substitution, the industry is running every other play at once: harvesting more tritium, proposing to breed it deliberately, filtering helium-3 out of ordinary helium, moving qubits to warmer operating points, reviving 1930s magnetic cooling, building refrigerators onto the chips themselves, and mining the Moon. This article is my field guide to all the helium-3 alternatives: what each can actually deliver, what blocks it today, and the specific trigger that would change my grade.

The bottom line: no single alternative replaces helium-3 on the timeline quantum computing is scaling on, and the realistic outcome is a portfolio in which magnetic cooling absorbs the small-system market, helium-4-only platforms absorb everything that tolerates one kelvin, expanded tritium harvesting and isotope separation feed the dilution refrigerators that remain, and lunar extraction stays a long-dated hedge.

This is the capstone to three pieces I published earlier this year: the helium-3 explainer covering why the isotope matters and why the demand projections around it are inflated, the lunar mining assessment, and the magnetic cooling analysis. Where those articles went deep, this one points and summarizes. The depth here goes to the routes I have not yet covered, and two of them (deliberate tritium breeding and terrestrial isotope separation) deserve far more analysis than anyone has published.

Three Ways Out of a Shortage

Every response to a scarce input falls into one of three buckets: make more of it, need less of it, or replace the process that consumes it. For helium-3 the buckets are unusually distinct because the gas is a working fluid, not a fuel. A dilution refrigerator recirculates its charge indefinitely; demand comes from new machine builds plus losses during service, not from consumption per computation. That framing, which I quantified in the cryogenics infrastructure guide, matters for grading the alternatives: supply-side routes compete against a demand that scales with installed base, while demand-side routes attack the installed base requirement itself. The latest USGS count puts world helium-3 production near 40,000 liters a year, most of it from Canada, Russia, and the United States, with a meaningful share of that Russian and unavailable to Western buyers; against demand running well ahead of the accessible remainder, I put the structural deficit at roughly 20,000 to 40,000 liters per year in that guide. Purified material now trades above $10,000 per liter, with bulk and subsidized gas quoted lower.

Route One: Make More Helium-3

Harvest the Tritium Already Decaying

Helium-3 on Earth is a byproduct of tritium, which decays into it with a 12.3-year half-life. The arithmetic that governs every supply-side plan is worth stating once, plainly: each kilogram of tritium sitting in storage gives off about 420 liters of helium-3 per year at standard conditions. Not per half-life. Per year, every year, declining by 5.6 percent annually unless the inventory is replenished. Helium-3 supply is an inventory game, and whoever holds tritium holds a helium-3 annuity.

Two civilian facilities harvest that annuity today. Ontario Power Generation’s Darlington Tritium Removal Facility, the world’s largest, has been detritiating heavy water from Canada’s CANDU fleet since 1990, and its subsidiary Laurentis Energy Partners began extracting helium-3 from that stored tritium in 2021 as the first civilian, non-military source, with Air Liquide as distribution partner. South Korea operates the second facility at Wolsong. Together the two produce two to three kilograms of tritium per year, and the same paper reports a figure that should reframe how everyone thinks about near-term supply: by 2017, the Darlington facility alone was estimated to have accumulated around 100,000 liters of helium-3 in its stored tritium, enough that the U.S. Department of Energy approached OPG about extracting it. By that estimate, one concrete vault in Ontario held several years’ worth of today’s entire global deficit, generated as a radiological-safety byproduct.

A third harvesting site is arriving. Romania’s Cernavoda station is building its own tritium removal facility, with commissioning tests scheduled for 2026 and trial operation to follow in 2027, and the design work explicitly contemplates helium-3 production as a revenue line. Romanian researchers have already published a feasibility study on extracting helium-3 from the Cernavoda moderator cover gas, including a cryogenic distillation design reaching 99.1 percent purity. India’s large pressurized heavy-water fleet produces tritium on the same physics, though no public helium-3 program exists there.

Why this route works: it is running today, it uses existing nuclear infrastructure, and the refurbishment wave across the CANDU world (Darlington, Bruce, Cernavoda) extends the feedstock for decades. Why it is not enough: the entire civilian tritium harvest is two to three kilograms per year, which caps new helium-3 evolution in the low thousands of liters annually against a five-figure deficit, and the tritium itself now has a competing buyer. Fusion developers want the same kilograms as startup fuel, and every gram sold to a tokamak is 7.4 liters of future helium-3 exported out of the cryogenics supply chain. What would change the picture: publicly visible, commercial helium-3 extraction and sales from Wolsong and Cernavoda matching the Laurentis model, a DOE-OPG agreement that actually taps the accumulated Darlington reserve at scale, and a deliberate policy choice about how much CANDU tritium the fusion industry is allowed to burn.

Breed Tritium on Purpose

Here is the route almost nobody has modeled seriously, and the one your government could start tomorrow. The United States already breeds tritium deliberately by irradiating lithium-6 inside Tennessee Valley Authority’s Watts Bar reactors, in Tritium-Producing Burnable Absorber Rods that are shipped to the Savannah River Site for extraction. Helium-3 recovery is an explicit mission line at Savannah River, harvested when aging weapons reservoirs are serviced. The program has run since 2003, the interagency agreement with TVA extends to November 2035, and in 2023 NNSA amended its record of decision to allow up to 5,000 TPBARs every 18 months across Watts Bar units 1 and 2. Taylor Loy of the Federation of American Scientists has argued for doubling the production goal beyond the current 2.8 kilograms per 18-month cycle to support fusion energy leadership. Nobody in that policy conversation is talking about helium-3, which is a mistake, because the isotope economics may be more attractive than the fusion-fuel economics.

Run the numbers with me. A gram of tritium yields 7.4 liters of helium-3 when it has fully decayed, roughly half of that within the first 12.3 years. At purified helium-3 prices above $10,000 per liter, a gram of tritium therefore contains something like $74,000 of eventual helium-3, against a Canadian market rate near $30,000 per gram and a U.S. production cost around $50,000. The gross margin exists at purified helium-3 prices; at the $2,000 to $3,000 per liter the gas fetched in the late 2010s, it would not, which is part of why nobody built this business earlier. The second catch is time: a breeding program is an annuity purchase, paying for irradiation now and collecting gas over half a century. A program that breeds one kilogram of civilian tritium per year reaches, at steady state decades out, an evolution rate of roughly 7,500 liters of helium-3 per year, which would rival today’s entire civilian output. But ten years in, the same program yields only about 43 percent of that. To add 10,000 liters per year within a single decade, you would need to breed on the order of three kilograms of tritium annually: roughly 60 percent more than the NNSA program’s current goal of 2.8 kilograms per 18-month cycle, and about what the maximum licensed TPBAR loading could deliver if every rod position were committed.

Why it might work anyway: the reactors, rod designs, extraction facility, and helium-3 processing line all exist, and the DOE Isotope Program has already shown it will sign long-dated offtakes for future helium-3, having contracted to buy lunar helium-3 deliverable by 2029. A comparable offtake for terrestrially bred material would be less exotic, not more. Why nobody has done it: the rods compete directly with the weapons mission for reactor positions and unobligated low-enriched uranium, the nonproliferation optics of producing weapons-usable tritium for commerce are genuinely awkward, and no private actor can finance a 12.3-year half-life on venture timelines. What would change my grade: a Loy-style expansion of TPBAR irradiation framed around fusion fuel, with helium-3 recovery contractually attached as the co-product that improves the program’s economics. The fusion lobby may end up being the accidental savior of quantum cryogenics.

Let Fusion Feed Itself

The more ambitious fusion route skips tritium entirely. Deuterium-deuterium fusion produces tritium and helium-3 in roughly equal branches, and Helion, the leading deuterium-helium-3 developer, plans to breed its own helium-3 from D-D reactions rather than buy it. If that closes, helium-3 stops being a decay byproduct and becomes a manufactured industrial gas, in quantities that would eventually dwarf every other source on this list.

The distance between plan and physics remains wide. Helion’s Polaris prototype reached 150 million degrees Celsius in January 2026, using deuterium-tritium fuel; the deuterium-helium-3 operating regime needs roughly five times that temperature, and sustained net-energy D-D operation has not been demonstrated by anyone. There is also the circularity I flagged in the explainer: until D-D breeding works, fusion is a competitor for helium-3’s parent isotope, not a producer, because every deuterium-tritium program from ITER to the private ventures consumes the tritium whose decay would otherwise become helium-3. The trigger to watch is unambiguous: sustained D-D pulses with net energy at Helion’s Orion-class machines. Until that data exists, fusion-bred helium-3 belongs in the same mental category as lunar helium-3, a real option with an unproven core step.

Sift It Out of Ordinary Helium

The least glamorous supply route may be the fastest one. Ordinary helium is not isotopically pure: atmospheric helium contains about 1.37 parts per million of helium-3, and helium recovered from natural gas typically carries 0.05 to 0.3 parts per million, poorer than air because crustal helium is mostly radiogenic helium-4 from alpha decay. Dilute, yes. But the world’s helium industry already concentrates, purifies, and liquefies enormous volumes of the carrier gas. The helium-3 is flowing through existing plants right now and being sold off in party balloons and MRI magnets at helium-4 prices.

The separation physics has been solved for generations. A 1966 Atomic Energy Commission patent describes combining superfluid filtration with low-temperature distillation; below 2.17 kelvin, helium-4 becomes a superfluid and flows through porous superleaks that normal-fluid helium-3 cannot follow, and the complementary heat-flush effect sweeps helium-3 away from heat sources. Kuz’menko and Lebedev demonstrated laboratory-scale distillation separation in 1995, Air Products patented industrial distillation schemes for helium-3 recovery from natural helium, and a 2025 review of helium isotope separation technology catalogs the full toolkit (superleak, heat flush, cryogenic distillation, adsorption) and estimates that helium-3 recovered from natural-gas helium could meet roughly half of world demand.

The economics used to be the blocker, and this is where the situation has quietly inverted. A 2020 cost analysis of superfluid-based separation concluded that at the then-price of $2,750 per liter, entropy filters would need 80 percent efficiency at real-world feed concentrations of 0.2 to 0.3 ppm, marginal at best, and modeled a hypothetical future price of $12,000 per liter at which separation would become viable from 0.45 ppm feedstock with only 10 percent filter efficiency. Purified helium-3 now trades above $10,000 per liter, with bulk and subsidized material quoted lower. For the grade a dilution refrigerator actually needs, the paper’s speculative future is the current spot market. Interlune evidently agrees: alongside its lunar program, the company took a $365,000 DOE grant specifically to develop terrestrial helium-3 separation from ordinary helium as a near-term supply bridge.

Scale check, so nobody mistakes this for a panacea: at 0.2 ppm and 50 percent recovery, netting 1,000 liters of helium-3 per year requires processing about 10 million cubic meters of helium, a mid-single-digit percentage of global helium production, through a single add-on train. That is not absurd; it is a finishing column bolted to the liquid stream of one large liquefier. It does, however, mean the route runs through the handful of companies that own liquefaction capacity (Air Liquide, Linde, Air Products, QatarEnergy, Gazprom), none of which has announced an isotope program, and it inherits helium-4’s geopolitical fragility, which the 2026 Qatar supply shock just demonstrated. What would change my grade from promising to real: a pilot separation column commissioned at one flagship liquefaction plant, or the Interlune DOE work publishing recovery-efficiency data at industrially relevant feed rates. Of all the supply routes, this is the one where the distance between physics and product is shortest, and I find it telling that a Moon-mining startup is the only company visibly pursuing it.

The Russian Wildcard

For completeness: Russia historically supplied a meaningful share of Western helium-3 from tritium operations at Mayak, and that material has been effectively excluded from Western markets since 2022. A geopolitical thaw would change the supply arithmetic overnight, and would do so in a way that damages every capital-intensive alternative on this list by collapsing the price that justifies them. Russian re-entry is not a strategy anyone can plan on. It is a scenario every investor in this space should stress-test against, because the alternatives above are all, in effect, long positions on helium-3 staying expensive.

The Moon, Briefly

I published a full assessment of lunar helium-3 mining in April and will not repeat it. The short version: the resource is real, the contracts (Bluefors, Maybell Quantum, DOE) are real, and the extraction chain is unproven at every stage past excavation prototyping, with the economics resting on variables that have nothing to do with space technology. Since April, Interlune has continued executing on the ground: a NASA-funded excavation project with the Colorado School of Mines runs through mid-2026, building on the full-scale Vermeer excavator prototype, with a full extraction demonstration targeted for 2027 and a pilot plant by 2029. Competitors including Magna Petra and XMC are earlier still. My verdict from April stands: a leveraged option worth holding, not a delivery date worth planning around, and no commercial volumes before the early 2030s under any realistic sequence.

Route Two: Need Less Helium-3

Treat the Gas as Inventory, Not Fuel

The cheapest liter of helium-3 is the one that never leaks out of a fridge. Because dilution refrigerators recirculate their charge, industry demand decomposes into new-system charges plus service losses, and both terms respond to unglamorous engineering discipline: leak-tight gas handling, closed-loop recovery during maintenance, reclamation from decommissioned systems, and circuit designs that need less gas per unit of cooling power. A mid-size system charge runs on the order of 40 liters, while large research fridges have historically shipped with far more (one 2008 order required 180 liters), so reduced-charge designs of the kind Maybell Quantum markets, and the cooling-wiring decoupling in Bluefors’ modular platform that I covered in the infrastructure guide, translate directly into avoided demand. None of this makes headlines. All of it compounds, and for anyone operating fridges today it is the only lever with an immediate payback.

The One-Kelvin Escape Hatch

Not every quantum workload needs 10 millikelvin, and fridge vendors have started segmenting their product lines accordingly. Bluefors’ XLDHe High Power, which I detailed in the cryogenics guide, delivers 200 to 700 milliwatts at 1 to 1.2 kelvin using helium-4 only, no helium-3 anywhere in the system, and targets exactly the workloads that live comfortably at that temperature: superconducting nanowire single-photon detector arrays for photonic quantum computing, and silicon spin qubit systems. Every processor that moves to a helium-4-only platform exits the helium-3 demand curve permanently. The constraint is obvious: the escape hatch only fits the modalities that can physically use it, which brings us to the question of how many can.

Qubits That Run Hot

The most consequential demand-side development is happening inside the qubits themselves. In 2020, Andrew Dzurak’s UNSW group demonstrated a silicon quantum processor unit cell operating at 1.5 kelvin, and a Delft team showed universal logic in hot silicon qubits in the same issue of Nature. In 2024, Diraq’s Jonathan Huang and colleagues closed most of the remaining quality gap, operating silicon spin qubits above one kelvin with an algorithmic initialization protocol that prepares pure states despite thermal energy exceeding the qubit energies. The reported numbers: 99.85 percent single-qubit and 98.92 percent two-qubit gate fidelity, with initialization and readout at 99.34 percent, all above one kelvin. Hole spin qubits in fin field-effect transistors have operated above four kelvin in separate work, and Diraq has since reported foundry-manufactured devices crossing 99 percent two-qubit fidelity, though so far at conventional millikelvin operating points; nobody has yet combined foundry fabrication with above-one-kelvin operation.

The physics logic is compelling: cooling power at 1 kelvin exceeds cooling power at 20 millikelvin by orders of magnitude, which is precisely why the Huang paper frames elevated-temperature operation as a scaling prerequisite rather than a curiosity, and why it pairs naturally with the millikelvin cryo-CMOS control work that attacks the wiring problem from the electronics side. My caution is the fidelity tax. A two-qubit gate at 98.92 percent is a real achievement at that temperature and still sits meaningfully below the 99.5-plus regime where surface-code overheads start to become tolerable; every lost fraction of a nine multiplies the physical-qubit count needed per logical qubit. Hot qubits do not eliminate the trade, they relocate it, exchanging cryogenic scarcity for error-correction overhead. What would change my grade: two-qubit fidelities at or above 99.5 percent, above one kelvin, in foundry-fabricated silicon. Hit that, and the one-kelvin escape hatch widens into a highway, and a large fraction of projected helium-3 demand simply never materializes. This is the modality-shift lever, and across the full taxonomy of qubit approaches it remains the single biggest variable in every helium-3 demand forecast.

What the Last Shortage Teaches

The 2008 crisis resolved through substitution, and the record is worth internalizing because it shows how fast demand evaporates once engineering attention concentrates. Security and safeguards consumed the overwhelming majority of helium-3 when the shortage hit; GAO-11-753 documented boron-10 lined proportional counters, boron trifluoride tubes, and lithium-6 scintillators moving through qualification, with the first boron-10 portal monitors deployable by fiscal 2012, roughly three years after the crisis peaked. Medicine followed the same script: hyperpolarized lung MRI migrated from helium-3 to xenon-129, which won FDA approval as a clinical imaging agent, Polarean’s Xenoview, in December 2022. The largest demand blocks on Earth were engineered away from the isotope within half a decade of the price signal.

The lesson cuts both ways for quantum. The encouraging half: markets facing a 50-fold price increase find substitutes with remarkable speed, and Routes Two and Three are quantum’s version of that search. The sobering half: boron-10 could substitute because those applications used helium-3’s neutron-capture cross-section, a property other nuclei share. Dilution refrigeration exploits the fact that helium-3 is the one stable fermion that remains a fluid at absolute zero, a role nothing else in the periodic table can fill. Quantum computing cannot substitute the isotope. It can only substitute the machine, or the operating point.

Route Three: Replace the Millikelvin Stage

Magnetic Cooling: The State of Play

I published a dedicated analysis of adiabatic demagnetization refrigeration versus the dilution fridge in April, with the physics, the materials race, and the quantified cooling-power gap, so here I compress to the state of play. Continuous ADR is the only demonstrated technology that reaches quantum-relevant millikelvin temperatures with no helium-3 whatsoever, and kiutra’s SPROUT platform, published in Review of Scientific Instruments in early 2026, proved the concept as a system: continuous operation below 30 millikelvin, 20 millikelvin single-shot, rack-scale format, with RF wiring sufficient to run a five-qubit superconducting processor. The company’s EIC-funded LEMON project targets the scale-up, its founders told Science they expected dilution-class cooling power at 20 millikelvin by the end of this year, and in April they put a datasheet behind the promise: the announced X-Type architecture is specified at 20 microwatts of cooling power at 20 millikelvin, a 12 millikelvin base temperature, and up to 768 wiring lines, with first customer deployments in 2027.

My April verdict has not moved, and the X-Type is exactly the test of it. The gap is cooling power under real wiring heat loads, not temperature: a Bluefors XLD-class fridge guarantees more than 30 microwatts at 20 millikelvin, the SPROUT demonstrator delivers just over 3 microwatts at 50 millikelvin, five qubits of wiring is not five hundred, and cycling multi-tesla magnets beside flux-sensitive qubits is a shielding problem that grows with scale. On paper, the X-Type specification closes most of that power gap. What graduates cADR from demonstrator to competitor is independent customer data at those numbers under real wiring loads, and the 2027 deployments are where that data will come from. The 1,000-qubit class stays dilution territory for years either way.

Cooling From Inside the Chip

The most conceptually interesting attack on the problem inverts it: instead of building a colder box around the processor, build refrigeration into the processor. In January 2025, a Chalmers and University of Maryland team led by Mohammed Ali Aamir, Nicole Yunger Halpern, and Simone Gasparinetti demonstrated a quantum absorption refrigerator made of superconducting circuits that cools a transmon qubit using nothing but a thermal gradient. Two auxiliary circuit elements couple to engineered hot and cold baths (microwave waveguides carrying synthesized quasithermal radiation), a three-body interaction moves entropy out of the target qubit, and the machine runs autonomously with no control pulses or feedback. Starting from a fully excited qubit, it reaches an effective temperature near 22 millikelvin, below what the surrounding cryostat provides and below what state-of-the-art active reset protocols achieve, with ground-state populations of 99.97 percent. Finnish groups at Aalto and VTT are pursuing the rival on-chip approach through electronic tunnel-junction coolers, a lineage that includes the quantum-circuit refrigerators developed there over the past decade.

Precision matters about what these devices are. They cool qubit populations, picowatt-scale entropy flows, and they do so brilliantly. They do not cool substrates, attenuators, cabling, or amplifier chains, and they do nothing about thermal photons leaking down the lines or quasiparticle poisoning, so they cannot replace the platform refrigerator. What they can do, and this is the synthesis I think the field is sleeping on, is relax the platform’s specification. If on-chip machines handle qubit initialization and residual thermal excitation locally, the surrounding bath no longer needs to sit at 10 to 20 millikelvin; a 100 to 300 millikelvin environment might suffice for the passive components. Continuous ADR already holds 300 millikelvin in shipping products, and helium-4 systems reach one kelvin without a drop of helium-3. A future architecture pairing warm-ish helium-3-free platform cooling with cold-on-chip qubit refrigeration is speculative today, unproven as an integrated system, and the single most elegant escape route on this list. The trigger: any multi-qubit processor running error correction with on-chip cooling in a bath above 100 millikelvin.

Dead Ends Worth Naming

For completeness, the routes that do not go anywhere, because each still surfaces in my inbox. Substituting helium-4 fails at the level of quantum statistics; the isotopes do physically different jobs, and I devoted a section of the explainer to burying this one. Pomeranchuk cooling reaches millikelvin through the anomalous melting curve of helium-3 itself, which makes it a consumer of the scarce isotope, not an alternative; its place in history (it enabled the discovery of superfluid helium-3) is secure, its place in this discussion is not. Laser cooling of solids stalls around 90 to 100 kelvin, four orders of magnitude in temperature above where qubits live. Thermoelectric and electrocaloric devices have no demonstrated sub-kelvin stage. And nuclear demagnetization, the microkelvin champion, sits below a millikelvin precooler rather than replacing one; it extends the temperature floor, which was never the constraint, and adds nothing to cooling power, which is.

The Scoreboard

What I Would Do With This

If you run a quantum hardware program, your helium-3 strategy for the next three years is written in Route Two: specify charge volumes, guaranteed loss rates, and recovery provisions in every fridge procurement, qualify a helium-4-only platform for every workload that tolerates one kelvin, and build the relationship with Laurentis or the DOE Isotope Program before you need it, exactly as I advised in the lunar assessment. Track kiutra’s X-Type the way you track a competitor’s roadmap, because if the announced 20-microwatt specification holds under customer wiring loads, your next characterization cryostat should not contain helium-3 at all.

If you shape national quantum or isotope policy, the two neglected files on this list are yours: civilian tritium breeding as a co-product of fusion-fuel policy, and separation capacity at allied helium liquefiers. Both are conventional industrial programs hiding behind an exotic-sounding isotope, both sit years closer than lunar extraction, and both belong in the supply-chain chapters of any serious strategy, a pattern I map across the whole quantum stack in Quantum Sovereignty. The strategic exposure is real even though, and this deserves emphasis against the commodity edition of Q-FUD now circulating, helium-3 scarcity does not gate CRQC arrival or move Q-Day. Within my CRQC Quantum Capability Framework this entire topic lives inside E.1, Engineering Scale and Manufacturability, as a cost-and-logistics term on the superconducting and spin pathways; trapped ions, neutral atoms, and photonics route around the millikelvin stage entirely. Anyone selling you a quantum timeline built on a gas shortage is selling you a story about one supply chain, not about the field.

For investors, the next 18 months resolve an unusual amount of uncertainty, and five datapoints do the resolving: independent power-at-temperature data from kiutra’s first X-Type deployments, Interlune’s 2027 extraction-and-return demonstration, any published plasma temperatures from Helion’s D-D campaigns, recovery-efficiency numbers from the DOE-funded terrestrial separation work, and Diraq’s next two-qubit fidelity figure above one kelvin. Each is a public, falsifiable milestone, and together they will sort this list into winners and museum pieces.