IBM's Nighthawk r3 targets 10,000 quantum operations on 520 qubits as it pushes toward fault tolerance

IBM targets 10,000 quantum operations with Nighthawk r3

IBM has unveiled plans for its next quantum processor, Nighthawk r3, which the company says will be built to accurately execute 10,000 quantum operations across 520 programmable qubits. The chip will carry 2,034 physical qubits in total, keep the square-lattice layout of today's Nighthawk hardware, and add ten chip-to-chip interconnects — called l-couplers — designed to join separate processors into larger machines. In 2027, IBM intends to pair two r3 chips into a single 1,040-qubit system.

The announcement, dated November 11 from IBM's Yorktown Heights research hub, positions the new processor as the next step on the company's route toward fault-tolerant quantum computing.

IBM argues that usefulness now depends on speed as much as scale. Because a quantum computer delivers answers in samples — each run of a circuit returns one result — demanding applications need circuits with thousands of operations running on roughly 100 qubits or more, repeated millions or billions of times. "Useful quantum computing hinges on whether results can be delivered with accuracy and speed, both are fundamental," said Jay Gambetta, director of IBM Research and IBM Fellow.

Nighthawk r2 shows the speed payoff

The case for r3 rests on results from the current generation. Nighthawk r2, introduced in September, delivers up to 100,000 samples per second and supports accurate computations involving more than 7,500 quantum operations on 120 qubits, helped by fast qubit-reset technology that shortens the pause between circuit executions.

Clients and partners including E.ON, Algorithmiq and Q-CTRL have run more than 34 billion circuits on the processor, reporting performance gains of up to 19x over previous IBM quantum chips. E.ON said it saw a 15x speedup on quantum machine learning work for energy grid optimization, while IBM reported a 12x speedup rerunning a magnetic materials simulation first conducted with Oak Ridge National Laboratory, with results consistent with neutron scattering experiments. Both comparisons were against the earlier Heron generation.

An Algorithmiq experiment simulating a heterogeneous quantum material was extended to r2 as well, running circuits eight times faster than the original Heron version. IBM estimates the new circuits would require roughly 240 times more memory to simulate classically with leading tensor-network methods at comparable accuracy.

L-couplers and cryogenics point to modular scaling

Two demonstrations underpin IBM's confidence that it can join chips into bigger systems. Earlier this year the company linked Nighthawk r1 processors using four l-couplers, achieving 99.3 percent quantum state transfer fidelity — progress toward a stated 99.9 percent target. It also combined two cryogenic modules into one system for the first time, reaching operating temperature in under five days and cooling below 15 millikelvin.

Quantum and classical computing merged on one platform

Alongside the processor news, IBM refreshed the IBM Quantum Platform to combine quantum and high-performance classical computing in a single managed environment, reflecting the reality that error correction, error mitigation and post-processing all lean heavily on classical resources.

Early users report gains from the setup. IBM and RIKEN scientists ran their sample-based quantum diagonalization workflow about five times faster while performing 100 times more computations than in the original study, and Cleveland Clinic and RIKEN used the platform's automation to process 608 million samples for modeling biologically relevant structures 33 times faster than without it.

Separately, IBM has launched a quantum error correction co-design platform with Oak Ridge National Laboratory and AMD that physically co-locates Nighthawk with ORNL's Umbra HPC system, built on AMD Instinct MI355X GPUs and 5th Gen EPYC CPUs.

Why it matters

The r3 specification signals a shift in how quantum progress is measured. Qubit counts alone no longer tell the story; the number of quantum operations a chip can execute accurately, and how quickly it can generate samples, determine whether industry-relevant workloads finish in practical time and at reasonable cost. IBM's bet is that modular linking through l-couplers — rather than ever-larger monolithic chips — is the viable path from hundreds of qubits toward the machines fault tolerance will eventually demand.

There are caveats: r3 is a design target rather than a shipping processor, the demonstrated interconnect fidelity of 99.3 percent still falls short of the 99.9 percent goal, and the 1,040-qubit system remains a 2027 plan. Still, if r2's measured speedups carry over to the larger chip, the 10,000-operation milestone would put bigger, classically hard computations within practical reach.