IBM puts Nighthawk next to AMD-powered Umbra to test quantum error correction

IBM, the Quantum Science Center at Oak Ridge National Laboratory (ORNL) and AMD have launched a co-design platform for quantum error correction. It places IBM’s 120-qubit Nighthawk processor next to ORNL’s AMD-based Umbra supercomputer.

Quantum chips and GPUs under one roof

According to IBM, the first deployment co-locates its quantum computers, including Nighthawk, with Umbra, a high-performance computing system built on AMD Instinct MI355X GPUs and 5th Gen AMD EPYC CPUs. IBM’s own blog places the platform at its site in Poughkeepsie, New York. The AMD hardware is meant to supply the classical computing for error-correction research, decoding optimization and software orchestration.

The setup will also connect to the Oak Ridge Leadership Computing Facility in Tennessee. IBM says that gives researchers a route to study quantum as part of a wider AI and scientific computing infrastructure.

On the quantum side, IBM points to Nighthawk’s square lattice connectivity, its speed in circuits per second and real-time pulse control. Through IBM’s Quantum System API, researchers can measure qubits mid-calculation and act on the result immediately. They can also set the exact timing of signals, see what the hardware actually ran, and work with raw measurement data. The stated aim is to compare error-correction methods by the qubits, time, samples and classical compute each one needs, and to find where GPUs help most.

The workflows named include concatenated codes, space-time codes, algorithm-aware decoders and hardware-aware compilation, all in one stack.

A roughly 230-fold gap to the 2028 target

The launch is tied to the Department of Energy’s Quantum Genesis Mission. Its Q Competition targets 100 logical qubits running 100,000 hard operations by late 2028. IBM says its current experiments reach up to 70 logical qubits and 438 hard operations, a result listed on the Quantum Advantage Tracker. Closing that gap, the company says, needs about a 230-fold increase in operation count.

Travis Humble, director of the Quantum Science Center, said: “With Umbra, we will now test new ideas, evaluate approaches at scale, and build the foundation for more reliable quantum computers.” Jay Gambetta, director of IBM Research, framed the shift this way: “The question is no longer if error correction will work on quantum hardware.”

How the platform fits a crowded race

The announcement does not say who can use the platform, how access works, or what it costs. It names the QSC and AMD as initial partners, so it reads as a research collaboration rather than a product. There are also no performance results yet. Every claim about what the setup will enable comes from the companies themselves and has not been independently checked.

The competitive context is set by the DOE program. Reporting on the Q Competition, including Singularity Hub’s coverage of a $215 million race, describes a funding pool for error-corrected machines, and DOE’s webinar materials say applications were due October 19, 2026. IBM has not said in this announcement whether it entered the competition, so the platform should not be read as a bid. What it does give IBM is a place to test the classical side of the problem, which is where it says scaling now depends on.

Other vendors are also pairing quantum hardware with classical software. Earlier this fall we reported that Intel is buying Classiq to add quantum software to its silicon qubits, and IBM’s own open-source toolkit moved forward with Qiskit 2.6. The IBM and AMD pairing differs in putting a GPU-class supercomputer in the same loop as the qubits, with a national laboratory as a partner.

The scale of the classical demand is a reason to expect more such pairings. IBM’s blog says recent research on hierarchical codes suggests a trillion logical operations could be reached at a physical error rate of one in a thousand. Decoding the error signals fast enough is a heavy computing task, and that is the work Umbra is meant to study.

A stepping stone to Starling

IBM says what the teams learn may inform IBM Quantum Starling, its first fault-tolerant quantum computer, which its blog says it plans to deliver by 2029. The press release adds that future AMD compute, networking and open software will be explored alongside quantum systems.

The collaboration builds on earlier work. IBM and ORNL combined quantum computers with the Frontier supercomputer to simulate magnetic materials earlier this year. IBM and AMD have also used AMD GPUs for the classical step in sample-based quantum diagonalization, and last year showed a decoding algorithm running on AMD chips.

The open question is whether more classical computing can close the gap that remains. The first results from Umbra and Nighthawk working together will be the first real test.