IBM plans a 520-qubit Nighthawk chip and a linked system for 2027
The proposed processor would expand programmable capacity. The harder test is making those qubits work reliably across chips.
Reading tools for text size, copying, printing and your reading list
IBM plans to expand its Nighthawk quantum processor from 120 programmable qubits today to 520 in its next-generation r3 design, then connect two chips into a 1,040-qubit system in 2027. The company’s November 11 roadmap update describes future targets, not an available 520-qubit service.[1]
For developers, the practical question is whether larger calculations can run reliably across those chips. Moving quantum information between processors introduces operations of its own, so a combined qubit count is only the starting point for judging a connected machine.
A qubit is a quantum computer’s basic information unit. “Programmable” identifies the qubits an algorithm can use; other quantum elements support their operation. IBM’s current r2 has 120 programmable qubits, 218 couplers and 120 reset elements, adding up to 458 physical quantum elements. That total is not 458 programmable qubits, nor a count of error-protected .[2]
The proposed 520 programmable qubits would be about 4.3 times r2’s 120, calculated as 520 divided by 120. Useful computing power does not increase by that simple ratio: errors, connectivity and the operations a particular calculation needs also matter.
The challenge becomes clearer in an earlier IBM experiment, described in an October 21 preprint. Researchers joined two Nighthawk r1 processors with four 80-centimeter superconducting links. Only a development subset of each chip was wired and measured. The links achieved 99% average state-transfer fidelity, a measure of how closely the transferred quantum state matched the intended one.[3]
A complete operation involving qubits on different chips scored differently. Its mean fidelity was 94.75%. That operation included two state transfers plus several local gates, the instructions that manipulate qubits. The paper’s benchmarking treats that complete sequence as one remote gate. Neither percentage is a prediction of the chance that a whole application will return the right answer.[3]
This distinction matters when assessing IBM’s next step: transferring information successfully is necessary, but applications depend on what the full chain of operations delivers. The October experiment is evidence about a building block, not a performance measurement of the proposed r3 system.
IBM is also targeting 10,000 quantum operations on r3. The next useful milestone is a delivered linked system with disclosed cross-chip gate performance and workload results. The takeaway is straightforward: IBM has set a larger programmable-capacity target; demonstrating useful performance across those chips remains the test.[1]