銳報SHARPPOST
Technology & Business

SharpPost Editorial Team | News Analysis | 23 September 2026

IonQ plans to install a Superion 256 quantum computer at the NVIDIA Accelerated Quantum Research Center in 2027, linking it to a GB200 NVL72 system through NVQLink. The US company said on 23 September that it would be the centre’s first on-site quantum computing deployment.

Product image of NVIDIA GB200 NVL72 racks and cabling
NVIDIA GB200 NVL72 product image. Credit: NVIDIA.

The partners will develop hybrid software and system prototypes for potential uses in financial risk modelling, materials science and drug discovery. No financial terms or benchmarks showing an advantage over existing classical methods were disclosed. Researchers must establish whether gains from quantum algorithms can outweigh the cost of moving data and coordinating the hardware.

Where GPUs fit

Quantum error correction protects information from noise by encoding it across multiple physical qubits. A classical computer interprets repeated error checks to determine the corrections needed—a process called decoding. NVIDIA made faster decoding a goal for the centre: a decoder that falls behind can create a backlog and delay subsequent operations.

NVQLink connects quantum control systems with GPUs for tasks including calibration and real-time error correction. The open-source CUDA-Q platform coordinates CPUs, GPUs and quantum processors within one program, letting researchers compare simulations with runs on compatible quantum hardware. GPUs are this project’s choice, not a requirement for every quantum system.

NVIDIA illustration of linked quantum and classical computing systems
Illustration of quantum processors working with classical computing systems. Credit: NVIDIA.

Deliveries targeted for 2027

At the 8 September launch, IonQ said SkyWater had fabricated the first 256-qubit processors and prototypes had trapped their first ions. Deliveries are planned for 2027. These milestones do not establish that all 256 qubits are operating in a complete system, or that it has 256 error-corrected logical qubits.

IonQ says on-chip electronics replace lasers for qubit control, with a server-rack footprint and cooling suited to conventional data centres. Full fault tolerance remains a development target.

A platform for several suppliers

NVIDIA named Quantinuum, QuEra and Quantum Machines among the centre’s partners in 2025 and introduced NVQLink as an open architecture for multiple suppliers. More connected devices could expand demand for its GPUs, software and interconnects. IonQ gains a setting to test its hardware alongside NVIDIA’s systems.

Prospective buyers will need reproducible results on practical tasks. At the same accuracy, and including error correction, data transfers and classical processing, can the combined system deliver results faster or more cheaply? That comparison matters more than installation or qubit count alone.