Illustration of the IBM Quantum System Two

IBM, AMD & ORNL Co-Design Platform for Quantum Error Correction

IBM, AMD, and the Department of Energy’s Quantum Science Center (QSC) at Oak Ridge National Laboratory (ORNL) recently unveiled a co-design platform for quantum error correction.

The first deployment is at IBM’s Poughkeepsie, New York, site, where IBM’s 120-qubit Nighthawk quantum processors are co-located with Umbra, an ORNL HPC system built on AMD Instinct MI355X GPUs and 5th-gen AMD EPYC CPUs. The platform connects to the Oak Ridge Leadership Computing Facility (OLCF) in Tennessee.

The platform addresses the classical half of error correction. Decoders, the software that reads syndrome measurements and identifies which qubits erred, must keep pace with the quantum processor, yet most researchers access quantum hardware via the cloud and run classical work on separate systems.

The three partners bring both sides together in a single environment for a defined research community. The timing aligns with the DOE’s Quantum Genesis Q Competition, which targets 100 logical qubits performing 100,000 hard operations by late 2028. IBM’s most recent reported result is 70 logical qubits and 438 hard operations.

Technical Details

The platform joins three elements in a single environment:

  • IBM supplies quantum hardware and pulse-level control software.
  • AMD supplies accelerated compute for decoding and orchestration.
  • ORNL supplies the HPC ecosystem and research community.

IBM describes the design as quantum-centric supercomputing, an architecture in which quantum processors, CPUs, and GPUs function as a single system. This includes the following components and capabilities:

  • Quantum hardware: IBM Quantum Nighthawk r2 processors with 120 qubits and square-lattice connectivity. Each qubit couples to four neighbors, about 60 percent more connectivity than IBM’s earlier heavy-hex layout, with direct compatibility with the standard surface code. IBM claims industry-leading speed in maximum circuits per second.
  • Classical hardware: Umbra pairs AMD Instinct MI355X GPUs with 5th Gen AMD EPYC “Turin” CPUs. AMD compute supports QEC research, decoder optimization, and software orchestration, while the GPUs let researchers train and evaluate decoders at scale.
  • Control interface: IBM’s Quantum System API provides pulse-level control. Researchers can measure qubits mid-circuit, act on the results in real time, set the precise timing of noise-suppression signals, inspect what the hardware executed, and work with raw measurement signals.
  • Co-location: The classical compute sits alongside the quantum processors in Poughkeepsie. The layout enables low-latency quantum-classical execution for decoding and conditional execution, in which a program selects its next operation based on a mid-circuit measurement. A connection to OLCF extends the platform into ORNL’s broader AI and scientific computing infrastructure.
  • Research targets: The partners list concatenated codes, space-time codes, algorithm-aware decoders, hardware-aware circuit optimization and compilation, and software orchestration within a single stack. IBM highlights hierarchical error correction, which layers an inner code within each module and an outer code across the inner logical qubits.
  • Prior collaboration: The platform builds on IBM’s work with AMD GPUs to accelerate the classical diagonalization step in sample-based quantum diagonalization, on a 2025 demonstration of a decoding algorithm on AMD chips, and on IBM and ORNL workloads that pair quantum hardware with the Frontier supercomputer.

Analysis

For IBM, the platform extends its quantum-centric supercomputing strategy and aligns its superconducting roadmap with the DOE’s 2028 target and its own Starling system. It places IBM hardware within the national laboratory ecosystem as the DOE allocates $215 million in a competition focused on logical qubit counts. 

For AMD, the platform provides a quantum reference deployment for Instinct and EPYC. This is AMD’s counter to NVIDIA in quantum-classical computing.

For ORNL, the platform adds another vendor relationship to a portfolio that already includes collaborations with NVIDIA and HPE.

Competitive Impact

The strongest competition comes from the NVIDIA ecosystem, which already connects GPU supercomputers to a broad set of quantum builders, and from quantum vendors that have published QEC results.

The following summarizes the principal alternatives and how each compares with the IBM, AMD, and ORNL platform:

AlternativeApproachReported PositionCompared to IBM-AMD-ORNL
NVIDIA NVQLink with CUDA-QOpen interconnect that links QPUs to GPU supercomputers across many quantum vendorsNVIDIA states 400 Gbps throughput and under 4 microsecond latency. Adopted by national labs including ORNL.Wider ecosystem and published interconnect figures. IBM and AMD offer a co-located 120-qubit QPU with pulse-level access and no published latency.
Quantinuum Helios98-qubit trapped-ion system with GPU decoding through NVQLinkReported 67 microsecond real-time qLDPC decoding in November 2025 and a record logical error of 5 in 1,000,000 per round in October 2026.Commercial system with published QEC results. The IBM-AMD-ORNL platform is a research environment with no published results.
Google Quantum AI Willow105-qubit superconducting processor running the surface codeDemonstrated below-threshold scaling with a distance-7 logical memory and an error suppression factor of 2.14 per step.Leading demonstration of QEC scaling on a superconducting chip. The sources reviewed describe no comparable shared co-design platform.
QuEra and collaboratorsNeutral-atom architectureReported up to 96 logical qubits and plans a Libra system by 2028 with more than 256 logical qubits.Higher reported logical qubit counts in a different modality, compared with IBM’s 70 logical qubits.

Differentiation is strongest in physical co-location of a 120-qubit IBM processor with a national laboratory GPU system and in pulse-level access for decoder research.

It is weakest on evidence. NVIDIA and Quantinuum have published latency numbers, and the IBM-AMD-ORNL partners have not.

Final Thoughts

IBM, AMD, and ORNL have built an environment that most QEC groups cannot assemble on their own, featuring a leading superconducting processor, current-generation AMD accelerators, and pulse-level control, all in one location.

The platform aligns with the DOE’s 2028 goal and provides IBM with a concrete path from its Starling roadmap into the national laboratory system.

AMD secures a named quantum deployment, and ORNL adds a second major vendor stack to its quantum program.

There are open questions. The partners have not published latency, logical error rates, or access terms. The field must still close a gap of roughly 230 times in hard operations to meet the Q Competition target, and a testbed at one site does not close that gap on its own.

For researchers and program managers evaluating QEC infrastructure, the platform deserves a place on the shortlist now and a performance review once data becomes available. IBM and AMD have secured a seat in the DOE’s 2028 error-correction effort, and their credibility over the next twelve months rests on publishing decoder latency and logical-error results that compare favorably with NVIDIA-linked systems.

Disclosure: The author is an industry analyst, and NAND Research an industry analyst firm, that engages in, or has engaged in, research, analysis, and advisory services with many technology companies, which may include those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.