IBM Plans Nighthawk r3 and a 1,040-Qubit Linked System

IBM said on 11 November 2026 that IBM Quantum Nighthawk r3 “will be built to accurately execute 10,000 quantum operations across 520 qubits,” and that in 2027 it “plans to link two Nighthawk r3 processors into a single system with more than 1,000 qubits.” The IBM Newsroom release and a same-day IBM Research blog on the path to Starling are the two official accounts. The new processor and the 2027 machine are plans. They have not been built.

The newsroom notice says r3 is “designed to support 10,000 quantum operations on 520 programmable qubits, with 2,034 total physical qubits,” and that it will introduce “10 chip-to-chip interconnects, called l-couplers.” Later in the same release, IBM is more specific about the 2027 step: it “plans to connect two Nighthawk r3 processors into a single 1,040-qubit system.”

Jay Gambetta, director of IBM Research, said: “Useful quantum computing hinges on whether results can be delivered with accuracy and speed, both are fundamental. Nighthawk is already showing us what that looks like today.”

Nighthawk r2’s speed claims are company and partner reports, not independent tests

The current generation, Nighthawk r2, was introduced in September. IBM says it “delivers up to 100,000 samples per second” and supports “accurate computations with more than 7,500 quantum operations on 120 qubits.” Clients and partners including E.ON, Algorithmiq and Q-CTRL “have executed 34 billion circuits on Nighthawk r2 to date, reporting performance improvements of up to 19x” over previous IBM processors, the release says. E.ON reported a 15-times speedup in quantum machine learning for grid optimisation; IBM reported a 12-times speedup on a magnetic-materials experiment with Oak Ridge National Laboratory. Both comparisons are against earlier work on IBM Quantum Heron.

Those figures are IBM’s and its partners’. They are not independent benchmark results, and they should not be read as such. The newsroom notice also points to quantum-advantage experiments first run on Heron and listed on a community Quantum Advantage Tracker, with an Algorithmiq material-simulation experiment now extended to r2. That is still IBM’s account of the comparison.

The Research blog is about the longer path. IBM Quantum Starling, it says, is “slated to launch in 2029” and “will be capable of running quantum circuits comprising 100 million quantum operations on 200 qubits.” Along the way, IBM says a revised qLDPC code now needs five connections between qubits rather than six, with “20x-100x lower logical errors,” and that the number of Josephson junctions in the relevant designs has been cut “from around 5,000 to 2,000,” which it expects will raise the share of fully functional processors from about 1 in 12 to about 1 in 3. It also says Relay-BP decoder performance “already meets the requirements for Starling,” with a real-time decoder ASIC planned on 2 nanometre technology. Those are research-blog claims about a 2029 machine.

The same newsroom notice also describes a software wrapper around the hardware. IBM Quantum Platform, it says, is now bringing high-performance classical computing and quantum computing “together into a single managed environment.” As an example, IBM and RIKEN scientists executed a sample-based quantum diagonalization workflow “roughly 5 times faster while performing 100 times more computations than in the original study.” That, too, is a company figure.

IBM Quantum Nighthawk’s l-coupler fidelity and error rates should be kept distinct

The modular bet depends on the l-couplers. The newsroom release says IBM this year connected Nighthawk r1 processors using four l-couplers, “achieving 99.3 percent quantum state transfer fidelity and progressing toward its 99.9 percent target.” It also says two cryogenic modules were linked “for the first time,” cooling below 15 millikelvin. The Research blog reports the same linkage in a different unit: “an average 1% error rate with the best result being 0.7%.” The best result in that blog line is the same 0.7% error as a 99.3% fidelity. The average, 1%, is a separate and worse figure. They should not be collapsed into one number.

IBM already has other public work on the same stack. It unveiled a Quantum System Two at Chicago’s Hyde Park Labs, and it is a partner in the Oak Ridge, IBM and AMD quantum error-correction platform mentioned again in this week’s release, which co-locates Nighthawk with ORNL’s Umbra HPC system. On the software side, IBM Quantum’s Directed Execution model is already in beta on Qiskit IBM Runtime, another reminder that the company is selling a platform around these processors, not only a roadmap slide.

The 11 November announcement is therefore a mix of a planned chip, a planned 1,040-qubit pairing, partner-reported speedups on r2, and a 2029 Starling target. The l-coupler result is real work already done, and it should be quoted with both the 99.3% best-fidelity figure and the 1% average error. The rest, including r3 itself, is still ahead of the hardware.

Sources

1. IBM Newsroom, “IBM Quantum Nighthawk Demonstrates New Levels of Speed and Quality, with Plans to Scale Beyond 1,000 Qubits in 2027,” 11 November 2026. https://newsroom.ibm.com/2026-11-11-ibm-quantum-nighthawk-demonstrates-new-levels-of-speed-and-quality,-with-plans-to-scale-beyond-1,000-qubits-in-2027
2. IBM Research, “Updating IBM’s path to IBM Quantum Starling,” 11 November 2026. https://research.ibm.com/blog/path-to-starling

Related stories

IBM Unveils Quantum System Two at Chicago’s Hyde Park Labs is IBM’s latest system-level installation.

Also relevant:

• ORNL, IBM and AMD Launch Quantum Error Correction Platform
• IBM Quantum Launches “Directed Execution” in Beta