IBM maps a 1,040-qubit Nighthawk system by connecting two planned r3 processors

IBM plans to connect two Nighthawk r3 quantum processors into a 1,040-programmable-qubit system in 2027. The announcement matters to teams whose experiments exceed a single processor's capacity, but the larger machine remains a development target. Today's available Nighthawk generation is r2.

The originating IBM announcement is dated November 11, 2026, coinciding with its Quantum Developer Conference. It updates the path from running repeated experiments on an existing processor to joining larger processors in one system.

The connection between chips is part of the computer

IBM specifies 520 programmable qubits for the planned r3 chip, compared with 120 on r2. It separately lists 2,034 total physical qubits for r3. These counts describe different things; the larger physical total is not the number of programmable qubits available to an experiment.

The design adds ten chip-to-chip links called l-couplers. A pair of processors working as one quantum system needs quantum connections, rather than simply a network that sends ordinary data between computers. The question is whether those links can preserve the behavior an experiment requires as the system grows.

An October 21 research preprint reports four such connections between two older Nighthawk r1 processors. Its authors distinguish quantum-state transfer from operations performed across the modules. Their reported mean transfer fidelity is 99.00%, while mean fidelity for distant CZ gates is 94.75%. A transfer figure should therefore not be read as the quality of every operation in a future combined machine.

That experiment supplies evidence for an interconnect approach. It does not establish that the planned two-r3 system is already available or has met its application targets.

Use the current processor to establish an experimental baseline

IBM's conference overview says r2 supports circuits with more than 7,500 quantum operations and throughput up to 100,000 samples per second. These are IBM's stated capabilities, not TechKili benchmark results. Samples are repeated measurement outcomes; collecting them faster can help an experiment, but does not by itself establish a useful answer.

For developers, the conference tooling guide points to public device benchmarks, tutorials and pulse-level controls on r2. Those provide a practical starting point for checking a workload's accuracy, repetition needs and total execution time before making plans around a larger processor.

Our assessment is to track the r3 system through demonstrated results, access terms and workload-specific comparisons. Programmable-qubit counts help describe scale; a migration decision also needs evidence that the full experiment finishes with useful accuracy and acceptable resource use.

Sources

• IBM's November 11 Nighthawk announcement

• IBM Quantum conference overview

• IBM developer tools and benchmark guide

• Meter-scale quantum interconnects preprint