IBM connected two refrigerator-sized quantum computing modules together this August at its Poughkeepsie, New York facility, and cleared a bottleneck that has quietly capped how big a quantum computer can get.
The problem sounds almost mundane: keeping chips cold enough to work. IBM’s superconducting quantum processors only function near absolute zero, colder than deep space, inside a dilution refrigerator called a cryostat. That cryostat also has to carry thousands of physical wires down to the chip to control and read every qubit. Add more qubits, and you add more wires and more heat leaking into an environment that has to stay below 15 millikelvin. Every single-cylinder cryostat eventually hits a hard ceiling on how much wiring it can hold.
Why One Big Fridge Was Never Going to Scale
IBM’s fix is to stop building one giant fridge and start linking smaller ones together, the same way a data center scales by adding server racks instead of building one impossibly large machine. Each new module offers about 0.53 square meters of usable wiring space and enough room to support at least 2,000 qubits, with up to 12 times the wiring capacity of IBM’s current systems. In the August test, engineers coupled two of these modules, cooled them to liquid-helium temperatures in under five days, and held them below 15 millikelvin with short, direct connections between the chips inside.
What This Buys, Eventually, for a Pharmacist Watching Drug Prices
Picture Warren, a pharmacist who has watched the price of a specialty drug climb for years, partly because finding one new molecule can take more than a decade and, by most industry estimates, over a billion dollars in trial and error. Fault-tolerant quantum computers are expected to eventually simulate molecules accurately enough to shortcut some of that trial and error. That is not a this-year benefit. It is the unglamorous infrastructure work that has to happen years before a faster, cheaper drug pipeline shows up as a smaller number on Warren’s register.
The Near-Term Payoff
IBM wants at least 1,000 programmable qubits by 2027 and a fault-tolerant machine, called Starling, by 2029. A cooling architecture that grows cell by cell, instead of hitting a wiring wall, is a prerequisite for both targets. Progress is showing up on the software side of the same push, too: a new AI-driven error-correction decoder recently beat standard benchmarks on real quantum hardware, and Brookhaven and Stony Brook proved quantum links can run over open air instead of buried cable.
None of this changes your pharmacy bill this month. But the machines that might, eventually, are being built one refrigerator cell at a time, and this week IBM proved the cells connect. Full details are in IBM’s announcement.