IBM maps a path from 520 qubits to a fault-tolerant machine by 2029
IBM has unveiled Nighthawk r3, a quantum processor with 520 qubits, due in 2027. It is designed to run circuits of 10,000 quantum operations, up from 7,500 on Nighthawk r2, the 120-qubit chip that became available in September 2026. IBM says speed and accuracy match r2. In 2027 it plans to link two r3 processors into a single 1,040-qubit system.
The linking relies on 10 chip-to-chip connections called l-couplers, a technology IBM tested using Nighthawk r1 processors. IBM argues that applications increasingly call for circuits with thousands of operations on roughly 100 or more qubits, which is the range these chips target. The longer goal is Starling, planned for 2029: IBM's intended first fault-tolerant quantum computer, running 100 million operations on 200 logical qubits. Its design, from a 2025 IBM paper, uses bivariate bicycle qLDPC error-correction codes, which need fewer physical qubits than surface codes. It would be built from modules joined by l-couplers.
The systems are also being deployed. Oak Ridge National Laboratory announced last week a machine pairing an IBM Quantum System Two with Nighthawk r2 processors and an AMD-powered supercomputer called Umbra. IBM also has a RIKEN partnership in Japan linking a Quantum System Two to the Fugaku supercomputer.
For abundance, the relevance is indirect. If error-corrected quantum machines work, they could lower the cost of hard simulations in materials science and drug discovery, which touches knowledge and, downstream, health. Less qubit overhead per error-corrected unit is the kind of change that would make that affordable.
The caveats are large. Nighthawk r3 and Starling are plans, not delivered products, and the performance figures are IBM's own. Fault-tolerant quantum computing has long seemed about a decade away, and a University of Auckland professor has claimed quantum computers have not factored numbers above 21 without cheating. Watch for r3 shipping in 2027 and for l-coupler results at scale.
Source: The Register
Many Minded