On this page
IBM has unveiled the Nighthawk r3, a quantum processor designed to run 10,000 quantum operations on 520 programmable qubits, and says it plans to link two of the chips into a roughly 1,040-qubit system in 2027. The company announced the IBM Nighthawk r3 on 11 November 2026 at its Quantum Developer Conference in Yorktown Heights, New York. It is a plan, not a product you can use yet: IBM has given no availability date for the r3 beyond that 2027 system.
What is the IBM Nighthawk r3?
According to IBM's press release, as syndicated by PR Newswire, the r3 is built to execute 10,000 quantum operations across 520 programmable qubits, with 2,034 physical quantum components in total. It keeps the square-lattice connectivity and fast operation of the earlier Nighthawk r2, and adds ten chip-to-chip connections that IBM calls l-couplers, which are meant to let separate processors work as one machine.
The gap between 520 and 2,034 is not an error. SiliconANGLE reports that the larger figure includes supporting elements such as couplers and reset components, not only the qubits programmers can use.
How does it compare with the Nighthawk r2?
IBM introduced the Nighthawk r2 in September 2026. It is the generation clients can use now.
| Nighthawk r2 | Nighthawk r3 | |
|---|---|---|
| Status | Available now (introduced September 2026) | Announced 11 November 2026; 2027 system planned |
| Programmable qubits | 120 | 520 |
| Quantum operations | More than 7,500 | 10,000 (design target) |
| Chip-to-chip links | None cited | Ten l-couplers |
Figures are from IBM's release and from StorageReview, which puts the r3 at about 4.3 times the usable qubits of the r2 and roughly a third more operations per circuit.
What does IBM plan for 2027?
IBM says it will link two r3 processors into a single system with more than 1,000 qubits, which outlets put at 1,040. Quantum Computing Report describes the r3 as a stepping stone to Starling, the fault-tolerant machine IBM is targeting for 2029. A 1,040-qubit figure is therefore two chips joined, not one 1,040-qubit chip.
What has the r2 shown so far?
Most of the performance evidence in the announcement comes from the r2, and it is reported by IBM and its partners rather than independently verified.
- IBM says the r2 reaches up to 100,000 samples per second.
- Clients and partners including E.ON, Algorithmiq and Q-CTRL have run more than 34 billion circuits on it, reporting speedups of up to 19 times.
- E.ON reported a 15 times speedup in quantum machine learning for energy grid optimisation, and an IBM and Oak Ridge National Laboratory magnetic-materials experiment reported 12 times.
- IBM says an Algorithmiq experiment ran 8 times faster on the r2, and that leading classical tensor-network methods would need an estimated 240 times more memory for the new circuits.
Techzine notes that these are customer-reported examples, not general results, and that the report contains no benchmarks for the r3 itself.
What else did IBM announce?
The same release covered software and platform changes. SiliconANGLE says IBM Quantum Platform now brings classical high-performance computing and quantum workflows into one managed environment, with pulse-level access for researchers working on error correction. IBM also launched an error-correction platform with AMD and Oak Ridge National Laboratory that places Nighthawk processors alongside AMD GPUs and CPUs.
Quantum Computing Report adds that "Managed Workflows" in IBM Quantum Functions provide persistent storage, GPU-accelerated classical compute and parallel orchestration for large hybrid jobs.
Analysis: what should readers make of it?
This is our reading, not IBM's. The headline numbers are targets. The 2027 two-chip system depends on linking chips well, and the evidence so far comes from older hardware. According to StorageReview, IBM's linking of two r1 processors reached 99% average state-transfer fidelity, with a best of 99.3%, against a 99.9% target. Different outlets quote slightly different figures for that demonstration, so treat the exact number with care.
The speedups are also relative. Several are measured against IBM's earlier Heron processors or against specific classical methods, so they do not show that a quantum computer beats the best supercomputers at a useful task. The practical test for r3 will be whether the 2027 system arrives on schedule and whether independent teams can reproduce partner results. Until then, the announcement is best read as a roadmap milestone, with the r2 as the hardware that exists today.
Sources
- IBM press release via PR Newswire (syndicated copy)
- SiliconANGLE: IBM Quantum announces upcoming Nighthawk r3 processor
- Quantum Computing Report: IBM unveils 520-qubit Nighthawk r3 at QDC 2026
- StorageReview: IBM Quantum Nighthawk r3 targets 10,000 operations
- Techzine: IBM Nighthawk r3 processor can handle 10,000 quantum operations
FAQ
Frequently asked questions
IBM announced it on 11 November 2026 at its Quantum Developer Conference in Yorktown Heights, New York.
It is designed with 520 programmable qubits and 2,034 physical quantum components in total. Two linked r3 chips are planned to make a system of about 1,040 qubits in 2027.
No. IBM has given no availability date for the r3 beyond its 2027 plan for a linked two-chip system. The Nighthawk r2, introduced in September 2026, is the current generation.
IBM uses the term for the chip-to-chip connections that let separate processors be linked into one larger system. The r3 has ten of them.
Spotted an error? Tell us. Read how we report and correct stories in our editorial policy.
Written by
Mainstream Tech
Newsroom
News and analysis on technology, AI, business, politics and international affairs, from our newsroom in London.
More from Mainstream



