IBM Links Two Nighthawk r3 Chips To Make A 1,040-Qubit Quantum System

AI & Compute

IBM Pairs Two Nighthawk r3 Chips Into a 1,040-Qubit Quantum System

IBM will link two Nighthawk r3 processors into a 1,040-qubit system in 2027, fabbed at its Albany 300mm quantum foundry, on the road to the fault-tolerant Starling machine in 2029.

By
Sophie Lindqvist
Filed
Channel
AI & Compute
Read
4 min read

IBM says it will join two Nighthawk r3 quantum processors into a single 1,040-qubit system, with the first r3 chip due in 2027. The company announced the plan this week at its IBM Quantum Developer Conference 2026 in Chicago. Both figures are roadmap targets, not shipped hardware.

The claim continues a fast sequence of launches. IBM unveiled the original 120-qubit Nighthawk a year ago, adopting a square-lattice topology in place of its earlier heavy-hex design. Nighthawk r2 followed in September, keeping 120 qubits and adding on-chip qubit reset that lifts throughput above 100,000 circuits per second.

What has actually shipped?

Nighthawk r2 carries the confirmed numbers. Since September, users have run more than 34 billion circuits on it, IBM said. In the first three weeks after launch alone, users ran 8 billion circuits. IBM has also executed 7,500 gates on the chip, 19 times faster than Nighthawk r1 would have managed, thanks to the reset capability.

Oliver Dial, IBM Fellow and senior vice president of quantum systems, described the mechanism to journalists before the event. "This is an extra physical qubit that we glued on the chip for each programmable qubit that lets us quickly remove the energy from one of our programmable qubits, returning it to the ground state," he said.

Speed matters as much as scale and quality, executives argued. Growing applications demand thousands of operations across 100 or more qubits to generate millions to billions of samples. Systems must run such workloads fast enough, and at acceptable cost, to be useful.

How will the 1,040 qubits be built?

Nighthawk r3 is specified at more than 520 qubits and 10,000 quantum operations. It retains square-lattice connectivity and adds 10 chip-to-chip l-coupler interconnects. Two linked r3 processors produce the 1,040-qubit system.

The couplers have a measured track record. IBM connected two Nighthawk r1 processors with four l-couplers and achieved 99.3 percent quantum state transfer fidelity. The company's goal is 99.9 percent. Dial said a first is now cooling down at IBM's Quantum Data Center in Poughkeepsie, New York: two Nighthawk r2 processors in two separate refrigerators, joined by four l-couplers.

Manufacturing will move to a US fab. Nighthawk r3 will be produced at Anderon, an IBM subsidiary operating a 300-millimeter superconducting quantum wafer foundry in Albany, New York. The expansion follows a $1 billion award from the US Commerce Department in May under the federal CHIPS Act, matched by $1 billion from IBM. Separately, GlobalFoundries received $375 million from Commerce to develop a quantum foundry.

What sits between Nighthawk and fault tolerance?

The end target is Starling, a fault-tolerant system planned for 2029 with 200 logical qubits and 100 million quantum operations. It will use bivariate bicycle codes, a class of quantum low-density parity-check codes that reduce the physical qubits needed per logical qubit. Blue Jay, the successor, is slated for 2033 or later.

Dial defended the approach against rivals. "Our path to large-scale fault tolerance is the only practical approach to fault tolerance out there," he said, citing an error-correcting code "over 10x more efficient" than alternatives such as surface codes, and a modular platform built from modules no larger than those IBM already builds.

The checklist remains long. Dial listed device physics for higher connectivity via c-couplers, packaging for more signals, real-time decoders, modular QPUs with l-couplers, modular cryogenics, and modular control electronics. Starling also needs bigger devices, more wiring, a cryogenic cooling system at roughly three times today's scale, and real-time decoding.

Research papers have already outlined three quantum advantage demonstrations on Heron chips and a modular, scalable cryogenic cooling system. Partner work includes error-correction projects with AMD and Oak Ridge National Laboratory, plus hybrid quantum-classical research with RIKEN in Japan and the Cleveland Clinic.

Is quantum merging with classical HPC?

IBM also launched a managed environment on its Quantum Platform that places classical and quantum compute together. Users can integrate quantum workflows with classical resources ranging from laptops to supercomputers.

Jerry Chow, IBM Fellow and CTO of quantum-centric supercomputing, said the offering lets users "provision for a class of compute... from CPU-based all the way up to GPU-based" alongside quantum execution, with persistence across jobs and a single payment point for both resource types.

The service builds on a co-design platform announced this month, which paired Nighthawk r2 with Oak Ridge's Umbra supercomputer — powered by AMD Epyc CPUs and Instinct GPUs — in the same Poughkeepsie datacenter.

IBM has set precise markers for 2027 and 2029. Whether the 99.9 percent coupler fidelity and tripled cryogenic scale arrive on schedule will decide if the 1,040-qubit claim becomes hardware or a slide.

Source: The Next Platform

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,113 articles

Related articles

« Previous articleNext article »