What Quantum Computers Actually Look Like Inside the Fridge

The gold chandelier you have seen in every quantum computing press photo is not the computer. It is the wiring and plumbing that keeps the computer cold, and the actual processor is a small square of silicon or sapphire, usually somewhere between the size of a fingernail and a postage stamp, bolted to the very bottom plate where nobody bothers to point the camera. Everything above it exists to deliver signals down and pull heat back up.

That inverted structure explains most of what looks strange about the photos. The chip is small and the support system is enormous because a superconducting qubit only behaves like a qubit at around 10 to 15 millikelvin, a few thousandths of a degree above absolute zero and considerably colder than the background temperature of deep space. Getting there takes a dilution refrigerator, and dilution refrigerators are large, expensive, and unusually fussy about what you attach to them.

Why Quantum Computers Need to Run at 10 Millikelvin

Superconducting qubits are tiny circuits that store quantum information in the energy difference between two states. That energy gap is small, in the microwave frequency range, and any stray thermal energy floating around at a comparable scale will kick the qubit out of the state you carefully put it in. At room temperature the thermal noise swamps the signal completely. There is no quantum computation happening, just a very cold-looking piece of aluminum wiring behaving like ordinary metal.

Cooling to around 10 millikelvin pushes thermal energy well below the qubit’s transition energy, so the qubit sits reliably in its ground state until you deliberately excite it. It also puts the aluminum or niobium circuitry deep into its superconducting regime, where electrical resistance vanishes and the circuit stops leaking energy. Research on qubit coherence, including the fabrication work NIST is running with Fermilab’s SQMS Center, has consistently linked warmer operating temperatures and lossy materials to shorter coherence times, which is why nobody has found a shortcut worth taking.

What Each Plate Inside a Dilution Refrigerator Actually Does

Open the outer vacuum can and you find a stack of round plates, usually gold-plated copper, hanging from each other on thin supports. Each plate is a temperature stage, and each one is roughly an order of magnitude colder than the one above it. The typical arrangement runs about 50 kelvin, then 4 kelvin, then the still at roughly 800 millikelvin, then a cold plate near 100 millikelvin, and finally the mixing chamber at 10 to 20 millikelvin.

The first two stages are handled by a pulse tube cryocooler, a mechanical device that drives helium through a cycle and does the bulk of the heavy lifting down to liquid helium temperatures. Below 4 kelvin, mechanical cooling runs out of road and the dilution cycle takes over. A mixture of helium-3 and helium-4 separates into two phases inside the mixing chamber, and forcing helium-3 atoms across the boundary between them absorbs heat, in much the same way that evaporation cools your skin. That process runs continuously as long as the circulation pumps keep going.

Cooling power at the bottom is genuinely small. Vendor specifications for research-scale systems typically quote a few hundred microwatts at 100 millikelvin and something in the range of tens of microwatts at 20 millikelvin. That is the entire thermal budget for the chip, its cabling, its amplifiers, and every imperfect thermal joint in the assembly. A single badly heatsunk cable can eat a meaningful fraction of it.

The Wiring Problem Nobody Talks About

Each qubit needs at least one microwave line to control it, and readout resonators need their own input and output paths, so the line count climbs quickly. A few hundred physical qubits can mean several hundred coaxial cables running from room temperature down to the mixing chamber. Every one of those cables is a thermal highway connecting a 300 kelvin room to a 10 millikelvin plate, which is exactly what you do not want.

The workaround is layered. Cables switch materials as they descend, from copper at the top to stainless steel and then superconducting niobium-titanium at the bottom, trading electrical performance for thermal isolation. Attenuators are bolted to each plate, commonly around 20 decibels at the 4 kelvin stage and again lower down, which dumps the room-temperature noise riding along with your control pulses into a plate that can afford to absorb it. On the way back up, the readout signal passes through a quantum-limited parametric amplifier at the base, then a high electron mobility transistor amplifier at 4 kelvin, and only then reaches the electronics rack.

All of that hardware is why the chandelier looks the way it does. The visual density is cable management and thermal anchoring, not computation. It is also the part that scales worst, which is why so much current engineering effort goes into cryogenic control electronics that sit inside the fridge instead of feeding it from a rack upstairs.

Not Every Quantum Computer Lives in a Fridge

This is worth being clear about, because the dilution refrigerator has become shorthand for quantum computing generally and it only describes one modality. Trapped-ion systems hold individual atoms in electromagnetic traps inside ultra-high vacuum chambers, and while they use cryogenics in some designs to improve vacuum quality, the ions themselves are laser-cooled rather than fridge-cooled. Neutral atom platforms work similarly, with arrays of atoms held by optical tweezers in a vacuum chamber that sits on an optical table surrounded by lasers.

Photonic quantum computers push even further from the stereotype. The photons themselves happily operate at room temperature, and the cryogenics, where present, is confined to single-photon detectors that need a few kelvin rather than a few millikelvin. Spin qubits in silicon do use dilution refrigerators, but there is active work on operating them above one kelvin, where cooling power is thousands of times greater and the engineering gets dramatically easier.

The differences are not cosmetic, and they change the facilities question completely. Teams weighing one platform against another usually bring in quantum computing consultants at this stage, since a photonic system and a superconducting system make almost nothing in common of your floor plan, power draw, or staffing plan.

What This Means If You Are Evaluating Quantum Hardware

The practical numbers matter more than the aesthetics. A research-grade dilution refrigerator stands roughly two meters tall before you add the frame, needs a compressor drawing on the order of ten kilowatts continuously, and carries a purchase price that industry pricing generally puts in the high six figures and often past a million dollars once wiring and amplification are specified. Cooling from room temperature to base typically takes a day or two, and a full warm-up, repair, and cool-down cycle can consume most of a working week.

That maintenance rhythm is usually the deciding factor. Universities and national labs often want the fridge because the physics is the point, and having physical access to the chip is the whole reason the group exists. Enterprises running algorithm pilots rarely do, since cloud access delivers the same qubits without the helium supply contracts and the cryogenic technician on staff.

The interesting question over the next few years is whether the fridge shrinks or the qubits warm up. Both are being worked on, with modular cryostats designed to link multiple refrigerators and hot qubit research aimed at pushing operation into the one to four kelvin range. If either succeeds convincingly, the chandelier photo will start to look like a specific moment in hardware history rather than a permanent feature of the field.

If you are budgeting for anything in this space, price the infrastructure separately from the processor and ask hard questions about uptime between thermal cycles. The chip is the cheap part.