AMD's quantum week: a PsiQuantum stake, a Qblox control loop and the race for quantum's classical layer
Within a week, AMD joined IBM and Oak Ridge National Laboratory on an error-correction test bed, said it plans to invest in PsiQuantum, and published control-loop measurements with Qblox. Together the moves put AMD in more direct competition with NVIDIA to supply the classical computers that fault-tolerant quantum machines will need. The engineering results are real but early: design figures and test-bench round trips, not error correction running on live qubits.
What happened
On November 5, 2026, AMD said it is working with IBM and the Quantum Science Center at Oak Ridge National Laboratory (ORNL) on a new co-design platform for quantum error correction. The first deployment places IBM quantum computers, including the 120-qubit Nighthawk, next to ORNL's Umbra system, which uses AMD Instinct MI355X GPUs and 5th Gen AMD EPYC CPUs. AMD says the platform supports the US Department of Energy's Quantum Genesis Mission goal of a scientifically relevant quantum computer by 2028.
On November 12, PsiQuantum and AMD announced an expanded partnership. PsiQuantum had already selected AMD Versal FPGAs for its fault-tolerant platform; the new agreement adds Instinct GPUs, EPYC CPUs and the ROCm software stack. AMD also plans to make an equity investment in PsiQuantum. Neither company disclosed the amount or terms. The first application focus is drug discovery, and both companies will explore a presence at the Illinois Quantum and Microelectronics Park in Chicago, where PsiQuantum is the anchor tenant.
The same day, AMD published a technical article with Qblox on a direct link between a Qblox Cluster control system and the memory of AMD processors, with no host software in the path.
| Announcement | Date | What is concrete | What is not yet shown |
|---|---|---|---|
| IBM, ORNL and AMD co-design platform | November 5, 2026 | IBM quantum computers co-located with the Umbra system (AMD MI355X GPUs and EPYC CPUs) | No performance results published yet |
| PsiQuantum and AMD partnership | November 12, 2026 | Scope widened from FPGAs to GPUs, CPUs and ROCm; AMD plans an equity investment | Investment amount, terms and timing |
| PRISM decoder paper (PsiQuantum and AMD) | November 10, 2026 | Full design mapped to an AMD VP1902 FPGA at 250 MHz: 10.11 µs latency for a distance-8 block | The 6.59 µs figure assumes 400 MHz; mapping the full design at that speed is future work |
| Qblox and AMD control loop | November 12, 2026 | Median round trips of 2.70 µs to one EPYC CPU core and 4.05 µs to an Instinct MI210 GPU | Measured with a simple echo function, not a decoder on live qubits |
Why the classical layer matters
A fault-tolerant quantum computer measures its qubits again and again, sends the results to a classical decoder, and must get the correction back before the next step that depends on it. AMD's article describes this as a feedback loop that has to keep up with the qubits. Around that loop sit FPGAs for control and readout, CPUs and GPUs for decoding and calibration, and supercomputers for simulation. Our error-correction explainer covers the basics.
AMD's own pitch names all of these layers: Versal adaptive SoCs near the quantum processor, Instinct GPUs, EPYC CPUs and ROCm software. That is the market AMD is entering, and it does not require AMD to build qubits.
AMD and NVIDIA are chasing the same role
NVIDIA has been building the same links. In October 2026 it took a minority equity stake in the error-correction company Riverlane, and Qblox reported round trips below 4.0 µs on an NVIDIA RTX PRO 6000 Blackwell and below 7.2 µs on DGX Spark, using NVIDIA's NVQLink. Qblox now has published integrations with both vendors.
The two sets of latency figures are not directly comparable. They were measured on different hardware, with different frame sizes and from different points in the loop. AMD's article also argues for an open, memory-based interface so that a quantum computer does not have to live inside one vendor's fence
. Both companies' numbers are self-reported.
Hype versus substance
- Decoder speed: the PRISM paper's summary highlights 6.59 µs. The fully mapped design runs at 250 MHz with 10.11 µs latency, just above the authors' own 10 µs target for this block. The faster figure is projected from single layers built at 400 MHz.
- Control loop: the 2.70 µs and 4.05 µs medians come from a built-in echo function. A small user function took 6.33 µs on the runtime's general path, and the decoders tested were small demonstration codes.
- No qubits in the loop yet: none of these results corrects errors on a running quantum computer. They show that the classical parts can be fast enough, not that the full system works.
- Investment: AMD "plans" an equity investment. Until terms are disclosed, its size and conditions are unknown.
- Chemistry claims: PsiQuantum says its chemistry methods already perform beyond the state of the art on AMD hardware. The announcement does not include benchmarks, so treat this as a company claim.
What to watch
AMD and PsiQuantum say they will share more at SC26 in Chicago, November 15–20, 2026. Beyond that, the signals to watch are the size and terms of AMD's investment, a full PRISM mapping at 400 MHz, first error-correction results from the IBM and ORNL platform, and the Qblox loop running a real decoder on live qubits. For how to read announcements like these, see our guide to reading a quantum computing announcement.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum™.
- PsiQuantum and AMD Collaborate to Build the Computing Foundation for Utility-Scale Quantum Computing — PsiQuantum (Nov 12, 2026) psiquantum.com
- AMD Builds the Classical Foundation for Quantum's Future — AMD Newsroom (Nov 5, 2026) newsroom.amd.com
- Closing the Quantum Feedback Loop with a Composable Runtime: Qblox and AMD — AMD (Nov 12, 2026) amd.com
- Machine-learning decoders for fusion-based quantum computing in real time (Barter et al., PsiQuantum and AMD, 2026) arxiv.org
- Qblox Integrates NVIDIA DGX Spark With Quantum System Controller — The Quantum Insider (Oct 22, 2026) thequantuminsider.com
- NVIDIA Takes Strategic Stake in Riverlane — Quantum Computing Report (Oct 22, 2026) quantumcomputingreport.com