Will Magnetic Cooling Replace the Dilution Refrigerator? What the Numbers Say

The argument sounds clean: adiabatic demagnetization refrigeration reaches the same millikelvin temperatures as a dilution refrigerator, uses no helium-3, has no scarce-isotope supply chain, and the technology has existed since the 1930s. So why hasn’t it replaced the dilution fridge already?

Because temperature is the easy part. The hard parts are cooling power, continuity, and scale, and the gap between what magnetic cooling delivers today and what a scaling quantum computer demands is measurable in microwatts, qubit counts, and operating hours. This article puts numbers on that gap, assesses the state of play (one company matters for quantum; I explain why), and offers a two-front verdict on the timeline. It is the companion piece to my helium-3 explainer, where I cover why helium-3 matters and why the demand projections built around it are inflated.

How Adiabatic Demagnetization Cooling Works

The physics is straightforward, even elegant. A paramagnetic material contains magnetic moments (unpaired electron spins) that can point in any direction. At thermal equilibrium, the entropy of those spins depends on temperature and the applied magnetic field.

The ADR cycle runs in three steps. First, you apply a strong magnetic field while the material is thermally coupled to a bath (a pulse-tube cooler at a few kelvin). The field aligns the spins, reducing their entropy, and the heat generated by that ordering flows into the bath. Second, you thermally isolate the material by opening a heat switch. Third, you slowly reduce the magnetic field. The spins randomize, entropy rises, and the only source of the energy needed for that randomization is the thermal energy of the material itself. The material gets cold. Depending on the refrigerant, the magnetic field strength, and the starting temperature, this can reach below 20 millikelvin.

The cooling is real and immediate. It is also finite: once the spin entropy is exhausted (all moments are randomized at the available thermal energy), the material cannot absorb more heat. It begins to warm. This is the single-shot limitation, and it is the reason ADR historically lost to dilution refrigeration for quantum computing. A dilution fridge cools indefinitely; a single ADR stage cools until its entropy budget runs out.

From Single-Shot to Continuous: cADR

The solution is architectural, not fundamental. Continuous ADR (cADR) chains two or more ADR stages in series, connected by heat switches. While one stage provides cooling, the other regenerates (re-magnetizes and dumps its heat). When the active stage approaches its entropy limit, the heat switches flip: the regenerated stage takes over, and the spent stage begins its own regeneration cycle. The result is continuous cooling at the target temperature, with no interruption.

The concept is not new, but turning it into a reliable product is hard. Heat switches are the critical component. Each switch must provide excellent thermal conduction when closed (to let heat flow between stages) and near-perfect thermal isolation when open (to maintain the adiabatic condition), and it must cycle thousands of times without degrading. Mechanical heat switches use physical contact; superconducting heat switches exploit the difference between the superconducting state (low thermal conductivity) and the normal state (high conductivity) by applying a small magnetic field. Superconducting switches are faster and have no moving parts, but they add another cryogenic subsystem. The 2026 compact cADR demonstrator used both mechanical and superconducting heat switches across its four stages, and characterizing their reliability over thousands of thermal cycles was part of the work.

The distinction between “ADR reaches millikelvin” and “cADR holds millikelvin continuously” is the distinction between a demo and a machine. For quantum computing, only the second matters.

The Materials Race: Refrigerants That Define the Floor

The base temperature an ADR system reaches depends on the refrigerant. Conventional paramagnetic salts like cerium magnesium nitrate (CMN) and chromium potassium alum (CPA) have been used since the 1930s, but they suffer from a practical problem: they contain water of crystallization, which makes them fragile and unstable under vacuum and thermal cycling. Remove the water and the salt degrades.

The recent push toward quantum-computing applications has driven a materials science effort aimed at water-free, air-stable refrigerants that maintain large magnetic entropy changes at millikelvin temperatures. Three results from the past five years define the state of the art.

KBaYb(BO₃)₂, a geometrically frustrated ytterbium borate, reaches at least 22 millikelvin on demagnetization from a modest starting field. Its frustrated magnetic network suppresses ordering to temperatures several times below the energy scale of its magnetic interactions, which is what conventional refrigerants cannot do. It is stable in air, vacuum, and at high temperatures, which means it can be baked out for cryostat integration without damage. Published by Tokiwa et al. in Communications Materials (2021).

NH₄GdF₄, a rare-earth fluoride, shows a low-field magnetocaloric effect significantly stronger than the benchmark material Gd₃Ga₅O₁₂ (GGG), with a magnetic entropy change of up to 38.2 J·kg⁻¹·K⁻¹. Published in the Journal of the American Chemical Society (2025), it represents a different materials strategy: optimizing for large entropy change at accessible field strengths rather than pushing for the lowest base temperature.

A 2025 review in Accounts of Chemical Research from Xiamen University surveys the full landscape of inorganic ADR refrigerants, framing the central trade-off: achieving both large magnetic entropy changes and low magnetic ordering temperatures simultaneously. Materials that order at very low temperatures tend to have weak magnetic interactions, which limits the entropy available for cooling. Materials with large entropy changes tend to order at higher temperatures, which raises the base temperature floor. Threading this needle for the millikelvin regime at the cooling powers quantum computing demands is the open materials problem.

One Company Matters for Quantum: kiutra

Why I Single Out kiutra — and What Everyone Else Builds

Several companies sell ADR cryostats, and they are good at what they do. The reason I focus this analysis on one of them requires explaining what the others are built for and why that does not apply to running a quantum computer.

Danaher Cryogenics (Boulder, Colorado) acquired the HPD ADR product line from FormFactor in January 2025 and now offers a full portfolio spanning 1 K down to roughly 50 mK. The HPD lineage is strong: these systems have been the workhorse at NIST, NASA, Argonne, Fermilab, Los Alamos, and Brookhaven for over twenty years. Danaher also partners with Chase Research Cryogenics (Sheffield, UK) for sorption-pumped coolers and represents Leiden Cryogenics for dilution refrigerators in North America. Entropy GmbH (Basel) sells ADR platforms for academic condensed-matter research. NASA’s Jet Propulsion Laboratory and the Mullard Space Science Laboratory (UCL) have developed flight-qualified ADR systems for X-ray astronomy missions, where radiation-hardened single-shot coolers hold detectors at 50 mK in the vacuum of space.

All of these are single-shot systems. They cool, hold, and then warm back up.

That is not a limitation for the applications they serve. A materials scientist characterizing a superconducting sample needs millikelvin for the duration of a measurement, minutes to hours, then warms the stage to swap the sample and runs another cycle. An X-ray observatory’s transition-edge sensor array absorbs individual photons at 50 mK, and the satellite’s duty cycle tolerates periodic recharging of the ADR between observation windows (hold times of 20 to 30 hours are typical for space ADR). Nuclear forensics detectors, condensed-matter transport measurements, and detector testing all follow the same pattern: cool to millikelvin, measure, warm, repeat. Single-shot ADR excels here because it is helium-free, mechanically simple, and reaches the right temperatures with no scarce-isotope supply chain.

Quantum computing breaks this model. A superconducting processor does not “cool, measure, warm.” It stays cold. Error correction runs continuously. Calibration routines fire between computation cycles without interrupting the cryogenic environment. The machine operates at 10 to 20 millikelvin for weeks or months at a stretch, and any interruption (a warm-up, a recharge) means losing the quantum state, re-cooling (which takes hours to days in a dilution fridge and would take at least as long in any ADR system), and re-calibrating every qubit. A single-shot ADR that holds for hours and then needs to recharge is fundamentally incompatible with this operating model. The quantum computer does not pause while the cooling system catches its breath.

This is why kiutra stands apart. kiutra is the only company worldwide offering continuous ADR (cADR), which is the only ADR configuration relevant for a quantum processor. None of the vendors above, not Danaher, not Chase, not Entropy, not the space agencies, have announced or demonstrated continuous millikelvin ADR. Their products are excellent for their intended use cases, and those use cases are not quantum computing.

Founded as a spin-off from the Technical University of Munich, kiutra has raised over €30 million (investors include Intel Ignite, TRUMPF Venture, and the European Innovation Council Accelerator) and ships several product lines: the L-Type Rapid (top-loading, continuous at 300 mK, single-shot to 50 mK, optimized for rapid qubit characterization), S-Type variants for optical and research applications, and custom platforms.

The landmark result came from kiutra’s SPROUT project, published in Review of Scientific Instruments in early 2026. A four-stage cADR platform, integrated into a compact 19-inch rack-sized format, achieved continuous operation below 30 millikelvin and 20 millikelvin in single-shot mode, without helium-3. Critically, it included high-density RF wiring (provided by Delft Circuits) sufficient to operate a five-qubit superconducting processor. This was not a temperature demo bolted to a physics experiment; it was a system designed to host an actual QPU, with wiring, shielding, and thermal staging, inside a form factor suitable for deployment. For the first time, cADR and superconducting qubits occupied the same cryostat architecture.

kiutra’s LEMON project, funded by the European Innovation Council, is the next step: a large-scale, highly modular magnetic cooling system designed to meet the cooling demands of full-stack quantum computers. It focuses on improving the core components (heat switches, magnets, cooling media, novel refrigerants) and demonstrating the scalability that the SPROUT platform proved in principle but not in power.

The Cooling Power Gap, Quantified

This is the section that settles the “is ADR ready?” question with data rather than adjective. Temperature is solved. Continuity is solved at small scale. What is not solved is cooling power at the qubit counts that drive helium-3 demand.

A dilution refrigerator’s cooling power is published, guaranteed, and well-characterized. The Bluefors XLD1000sl delivers more than 30 µW at 20 mK and more than 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 independent cooling units. Oxford Instruments’ Proteox series offers comparable performance. These numbers represent continuous, guaranteed cooling power available to absorb the heat load from wiring, attenuators, amplifiers, and the QPU itself.

kiutra’s commercial products do not yet publish comparable millikelvin cooling-power specifications. The L-Type Rapid operates continuously at 300 mK and reaches lower temperatures only in single-shot mode. The SPROUT demonstrator achieved continuous sub-30 mK, but its cooling power at that temperature is not quantified in the published paper, and five qubits generate a far smaller heat load than the hundreds to thousands that drive industrial helium-3 consumption. kiutra’s own framing reflects this: they describe their current platforms as characterization and testbed systems, with LEMON as the path to full-stack quantum computing.

To frame the gap concretely, I draw on the heat-load budgets from my cryogenic infrastructure guide. A rule-of-thumb budget for a 100–200 qubit superconducting system includes over 500 µW at the mixing chamber stage (20 mK). An XLD delivers more than 30 µW at that temperature. The KIDE, with three cooling units, likely delivers roughly 90 to 120 µW at 20 mK (dilution cooling power scales approximately as T², so the 3,000 µW at 100 mK implies around 120 µW at 20 mK). Even dilution refrigerators do not have unlimited headroom at millikelvin; the thermal budget must be managed aggressively through the signal chain design, wiring architecture, and staging that I detail in the build guide.

For cADR to compete at this scale, it would need to deliver comparable sustained microwatt-class cooling power at 20 mK while simultaneously managing the magnetic-field cycling that is intrinsic to its operation. The LEMON project is kiutra’s vehicle for demonstrating whether that is achievable. Until LEMON publishes power-at-temperature data for a multi-hundred-qubit wiring configuration, the gap is an inference, not a measurement, but it is an inference grounded in physics: ADR cooling power depends on the refrigerant mass, the heat-switch conductance, the cycling speed, and the entropy budget per cycle, and scaling all four simultaneously while keeping the cycling magnets from disturbing the qubits is an unsolved systems problem.

The Remaining Challenges

Beyond raw cooling power, three engineering constraints separate the SPROUT demonstrator from a production platform.

The first is the magnetic field. ADR cycles a field of 1 to 4 tesla on and off, and superconducting qubits are exquisitely sensitive to stray magnetic fields. Field shifts qubit frequencies, trapped flux causes loss, and strong fields can quench the superconducting films outright. The SPROUT system addresses this through magnetic shielding and geometric separation between the magnet and the qubit stage, and the five-qubit result shows the approach works at small scale. At larger qubit counts, the shielding mass grows, the thermal links between the shielded ADR module and the qubit stage introduce their own heat load, and the question is whether the engineering holds without eating the cooling power it is supposed to deliver. For silicon spin qubits, which operate in a deliberate applied magnetic field, this constraint is milder.

The second is refrigerant production quality. The new materials (KBaYb(BO₃)₂, NH₄GdF₄, and others) are grown as research-grade crystals. Manufacturing them in the kilogram quantities and consistent quality a product line demands, with reproducible entropy-change profiles and mechanical integrity under thermal cycling, is a materials-engineering problem that has not been addressed at scale. A review of the field describes this trade-off explicitly: achieving both large entropy change and low ordering temperature simultaneously remains the central materials challenge.

The third is the operating duty cycle. A dilution refrigerator runs for months between maintenance events. cADR cycles its stages continuously, each cycle stressing the heat switches, the magnets, and the thermal joints. The SPROUT system ran for the duration of a research campaign; no multi-thousand-hour reliability data exists. Bluefors has shipped over 1,800 dilution systems with decades of cumulative field time. kiutra’s installed base is measured in tens of units, primarily for characterization rather than continuous QPU operation.

What Else Could Displace the Dilution Fridge

Magnetic cooling is not the only technology competing for the millikelvin future. Two others bear watching, though they are complementary rather than competitive with cADR.

Cryo-CMOS co-integration attacks the problem from the opposite direction. Instead of increasing the cooling power of the fridge, it reduces the heat load at the cold stage by moving the control electronics from room temperature down to the cryogenic environment adjacent to the qubits. Intel’s Horse Ridge II, SEEQC’s single-flux-quantum logic, and Diraq’s cryo-CMOS silicon-spin integration all aim to slash the number of cables that carry heat from 300 K to 20 mK. If cryo-CMOS succeeds, the wiring wall relaxes, the heat-load budget at MXC shrinks, and the cooling-power requirement that cADR has to meet comes down. Cryo-CMOS helps both dilution fridges and cADR, but it helps cADR disproportionately because cADR’s deficit is in cooling power, not temperature.

Better helium-3 management is the unsexy competitor to both. Closed-loop recovery during fridge servicing, reduced-charge dilution circuits, and improved recycling from decommissioned systems all extend the effective supply without new technology. Combined with modest investment in additional tritium production (more lithium irradiation in commercial reactors, expanded output from Darlington), terrestrial supply could close part of the deficit on a timeline shorter than either cADR scaling or lunar mining.

Where This Leaves the Timeline

I draw the same two-front line I hold on most quantum-hardware questions.

ADR has crossed the threshold from research curiosity to engineered demonstrator. The SPROUT platform is genuine: continuous millikelvin cooling, no helium-3, wiring for superconducting qubits, rack-mountable format. kiutra is the only company with the product intent, the funding, and the track record to carry this forward. Anyone still dismissing magnetic cooling as a lab toy is behind the evidence.

ADR has not crossed the threshold from demonstrator to production-grade competitor for the machines that drive helium-3 demand. Five qubits is not a hundred, a hundred is not a thousand, and characterization duty is not continuous QPU operation. The cooling-power-at-millikelvin gap is real, the magnetic-field-compatibility question is open at scale, and the materials and reliability track record is thin. Anyone claiming cADR is about to replace the dilution fridge is extrapolating from a demonstrator to a product line without the connecting data.

My estimate of the timeline: cADR is a viable characterization and small-scale-prototyping platform today (kiutra’s L-Type Rapid is already there). For qubit counts in the tens, the SPROUT architecture is close to deployable. For the 100-to-1,000-qubit systems where helium-3 consumption actually concentrates, LEMON and its successors need to demonstrate sustained microwatt-class cooling power at millikelvin with production-grade reliability. That is a multi-year engineering program. If LEMON delivers on the 2027–2028 timeline kiutra has indicated, a transition in the 100-qubit class could begin in the late 2020s. A transition at the 1,000-qubit KIDE-class scale is further out.

The CRQC Quantum Capability Framework I maintain tracks quantum engineering progress across multiple capability dimensions. Cooling falls under E.1: Engineering Scale and Manufacturability, and the transition from He-3-dependent dilution to cADR (if it happens) is an E.1 event: it changes the supply chain, the procurement model, and the scaling ceiling, without touching the quantum logic or error correction beneath it.