IBM builds a better fridge for its quantum computers
The New Stack Frederic Lardinois
IBM built and cooled the first two modules of a new fridge meant to house future quantum chips. No processors inside yet, but that's the whole point of the test.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
IBM spent Wednesday talking about refrigeration, which sounds like an odd headline for a quantum computing company until you remember that superconducting qubits are useless if they're not sitting a fraction of a degree above absolute zero. The company has now built and cooled the first two modules of a new cryogenic dilution refrigerator system, hitting temperatures below 15 millikelvin on their own and 23 millikelvin once IBM added a 30-microwatt heat load to simulate the processors, wiring, and electronics that will eventually live inside. There's no actual quantum processor in there yet. IBM says it plans to install one Nighthawk chip in each of the two modules later this year, which will be the real test of whether the hardware works and whether the modules can talk to each other.
The bigger story here is wiring, and space, and modularity. Jerry Chow, IBM's fellow and CTO for quantum-centric supercomputing, put it plainly in a press briefing: the challenge isn't just stacking more qubits, it's building all the infrastructure around them without cooking the fridge with excess heat. IBM's answer is to break the refrigeration environment into connectable rectangular cells rather than one giant chamber. The first two modules together stand about 8 feet tall and 8 feet wide, and each one offers roughly 12 times the wiring space of Quantum System One, the company's original commercial machine from 2019. That matters because IBM's current systems handle just over 100 physical qubits, while its fault-tolerant ambitions call for thousands, eventually hundreds of thousands, of physical qubits to produce a much smaller number of error-corrected logical ones.
Modularity also buys IBM some engineering flexibility. Oliver Dial, an IBM fellow and vice president of quantum systems, says the company can swap out a wiring assembly for a new processor generation without tearing apart the whole refrigerator. Dial admits a single massive vacuum chamber might be cheaper for a one-off machine, but it's a nightmare to manufacture, ship, and replicate. Building and testing individual cells before they reach a client site is the more repeatable path, even if it's not the cheapest one on paper.
The part still not solved is how processors actually talk to each other once they're sitting in neighboring modules. IBM is betting on something called the L-coupler, a superconducting cable that ferries microwave photons between chips up to a meter apart, capable of transferring quantum states or running two-qubit operations. The company already showed this off with Flamingo, a prototype linking two Heron processors, and Dial says the best fidelity achieved across an L-coupler now sits at 99.3%. IBM wants that number at 99.9% before it's good enough for a fault-tolerant machine, which tells you how much room is left between a working demo and something reliable enough to ship.
IBM expects to start deploying these modular fridges next year, with early systems using two or three cells to support around 1,000 programmable physical qubits. The company's Starling system, its planned fault-tolerant machine due in 2029, is expected to need about 12 modules. Dial names reliability, not qubit count, as the real obstacle ahead: going from thousands to hundreds of thousands of qubits, he says, means everything about the system needs to become a thousand times more reliable.
My take — AI-written commentary, not fact-checked reporting
A fridge announcement without a chip inside it is exactly the kind of thing that sounds anticlimactic until you notice what's actually happening: IBM is admitting that plumbing and wiring, not qubit counts, are the real bottleneck on the road to fault tolerance. Every quantum roadmap loves to talk about logical qubits and error correction, but nobody gets there without somewhere to put a hundred times more wires without melting the fridge. Watch the 99.3%-to-99.9% fidelity gap on that L-coupler — that's the number that decides whether 2029 is realistic or just a nice slide in a deck.
Read more about this at: The New Stack