IBM has cleared a scaling hurdle that has quietly limited how big quantum computers can get: how to keep chips cold enough to work without running out of room for wiring. On August 19, the company announced it successfully connected and operated two of its new modular cryogenic cooling units together at its Poughkeepsie, New York facility, a first-of-its-kind demonstration that points toward a very different way of building large-scale quantum machines.
Superconducting quantum processors, the kind IBM builds, only function near absolute zero, colder than deep space. Getting them there requires a dilution refrigerator, essentially a cryostat, that also has to carry thousands of physical wires down to the chip to control and read out each qubit. That’s the catch: every additional qubit means more wiring, and more wiring means more heat leaking into an environment that has to stay below 15 millikelvin. Traditional single-cylinder cryostats have a hard physical ceiling on how much wiring they can fit before this becomes unmanageable.
IBM’s answer is to stop trying to fit everything into one giant fridge and instead link multiple smaller cryogenic modules together, the same way data centers scale by adding racks rather than building one impossibly large server. Each module in the new architecture offers roughly 0.53 square meters of usable wiring area and 2.75 cubic meters of vacuum chamber volume, enough to support at least 2,000 qubits per cell, IBM says, with up to 12 times the wiring capacity of the company’s most widely deployed quantum systems today. In the recent test, two of these aluminum, refrigerator-sized modules were coupled and cooled together, reaching liquid-helium temperatures in under five days and settling below 15 millikelvin, while keeping short, direct interconnect paths between the chips inside.
That last detail matters as much as the temperature numbers. Short interconnects mean quantum information can move between chips in different modules with less signal loss and less added noise, which is exactly the kind of overhead that degrades fragile qubit states. It’s the difference between designing one enormous, monolithic refrigerator and designing a cooling system that can grow module by module as more processors are added, without redesigning the whole rig each time.
This is infrastructure work rather than a flashy new algorithm, but it’s the sort of unglamorous engineering that determines whether “more qubits” is actually achievable outside a lab demo. IBM has said it wants to support at least 1,000 programmable qubits by 2027 and is targeting its Starling system as a fault-tolerant quantum computer by 2029; a modular cooling architecture that can scale by adding cells, rather than hitting a wiring wall, is a prerequisite for both. Fault tolerance is widely viewed as the threshold quantum computing needs to cross before it can reliably outperform classical machines on real-world problems in chemistry, materials science, and optimization, so bottlenecks like this one are worth watching even when they don’t come with a dramatic qubit-count headline.
IBM plans to install its Nighthawk processors into this modular system for further testing later in 2026. Full details are available in IBM’s announcement.
Leave a Reply