IBM plans to join two quantum chips into a 1,040-qubit system in 2027

IBM said on Wednesday that it plans to connect two of its next-generation Nighthawk quantum processors into one 1,040-qubit system in 2027, a step toward building quantum computers out of linked modules rather than ever-larger single chips. It also published new performance figures for the Nighthawk r2 processor it launched in September.

The next chip: Nighthawk r3

According to IBM's announcement, Nighthawk r3 is designed to run 10,000 quantum operations across 520 programmable qubits. The chip physically holds 2,034 qubits in total, but IBM's headline number is the ones available for running programs. It keeps the square-lattice layout of the current chip, in which each qubit connects directly to its neighbours, and adds 10 chip-to-chip interconnects that IBM calls l-couplers.

Those couplers are the point. In 2027, IBM says it will use them to join two r3 processors into a single 1,040-qubit machine. The company has already tested the approach on its first-generation chips: it linked four Nighthawk r1 processors and moved quantum states between them with 99.3% fidelity. IBM's target is 99.9%, so the links still lose more information than it wants. It also connected two of its cryogenic modules, the refrigerators that keep the chips near absolute zero, into one system for the first time, cooling it below 15 millikelvin in under five days.

What today's chip is doing

IBM says Nighthawk r2 can produce up to 100,000 samples per second and handle more than 7,500 operations on 120 qubits, helped by a fast qubit reset that cuts the wait between runs. Customers and partners, including E.ON, Algorithmiq and Q-CTRL, have run more than 34 billion circuits on it, according to IBM.

The company highlighted several results, each measured against earlier work on its older Heron chips:

• E.ON reported a 15x speedup on quantum machine learning for power grid optimisation.
• Oak Ridge National Laboratory and IBM got a 12x speedup simulating magnetic materials, with results that matched neutron-scattering experiments.
• Algorithmiq ran a quantum materials simulation 8x faster. IBM estimates the leading classical method would need about 240 times more memory to match its accuracy.

IBM also says RIKEN and the Cleveland Clinic got large speedups from running quantum and classical computing together on its managed platform.

As The Next Web points out, every one of these figures comes from IBM or its partners and hasn't been independently checked. "Speedup" here means faster than IBM's own previous hardware, not faster than the best classical computers.

Why it matters

Quantum programs give one probabilistic answer per run, so useful work can take millions or billions of runs. That makes raw speed nearly as important as qubit count, which is why IBM leads with samples per second. Jay Gambetta, director of IBM Research, put it this way: "Useful quantum computing hinges on whether results can be delivered with accuracy and speed."

The bigger story is modularity. Making one chip with thousands of good qubits is extremely hard, so IBM, like several rivals, is betting on linking smaller chips together. Getting those links from 99.3% to 99.9% fidelity is one of the harder engineering problems in the field. The 2027 two-chip system will be the first real test of whether IBM's approach can scale past a single processor without losing the quality it claims today.