The Signal
QuEra Details AIST’s ABCI-Q Quantum-HPC Testbed
AIST’s ABCI-Q combines a GPU supercomputer with four quantum-computing platforms, giving researchers a place to test hybrid workflows. A separate AIST notice requests supplier materials and comments on a planned expansion with AI-computing
Quantum.world Team · Written with AI · fact-checked ·
Takano’s test for quantum-HPC integration
AIST is treating quantum-HPC integration as an operating problem to work through, not just a plan to place quantum machines near a supercomputer. A QuEra case study published October 26 describes a talk by Ryousei Takano of Japan’s National Institute of Advanced Industrial Science and Technology (AIST) about ABCI-Q, the center’s quantum-HPC platform.
The testbed brings AIST’s GPU supercomputer, System H, together with quantum processors and the software and workflows used to reach them. QuEra’s Gemini neutral-atom system, called System Q, is part of that setup. AIST’s official description lists System H, Fujitsu’s superconducting System F, QuEra’s System Q and an OptQC photonic platform as components of ABCI-Q.
Gemini sits 350 meters from the GPU system
The physical arrangement shows why “integration” involves more than sharing a campus. QuEra’s account says System H and Fujitsu’s System F are in AIST’s headquarters building in Tsukuba, while Gemini is in another building about 350 meters away. A dedicated network connects the neutral-atom machine to the rest of the facility; its room-temperature operation avoids the need for a cryogenic plant.
System H has about 2,020 NVIDIA H100 GPUs. For the current connection between System H and System F, users can submit work through SSH, a REST API or the browser-based Open OnDemand interface. The Octopus component manages jobs and data, while a control system handles the transfer, generates pulse sequences, runs the quantum processor and returns results.
That workflow is useful today, but the case study describes a technical limit: loosely coupled, batch-style jobs are not the same as fault-tolerant computing that needs low-latency, stateful communication among the user-facing system, control layer and HPC resources. ABCI-Q gives AIST a place to test those interfaces across machines instead of assuming a scheduler designed for conventional computing will be enough.
GPU simulations prepare work for quantum processors
Some of the platform’s reported results use the classical side alone. QuEra’s account says AIST worked with the University of Osaka on a quantum-chemistry circuit simulation using 1,000 GPUs. In a separate collaboration with Fujitsu, it tested QompSim, a distributed state-vector simulator that ran circuits up to 56 qubits across 1,000 GPUs, with reported fidelity above 99% in the cases tested.
Those calculations do not demonstrate a quantum processor outperforming classical computers. They show how a large GPU system can help researchers simulate and check circuits before committing jobs to quantum hardware. AIST’s broader setup also includes quantum frameworks such as Qiskit, CUDA-Q, Bloqade and QURI alongside conventional machine-learning tools, according to QuEra.
AIST’s next procurement targets AI agents
Access is opening beyond AIST’s own teams: the center announced in March that its superconducting and neutral-atom quantum computers were available to external users as well as to joint research projects, with access arrangements varying by system. AIST’s user page still describes the overall service as domestic-use-only and says international access is scheduled for fiscal 2026.
AIST is planning a larger step. In a procurement notice posted October 2, it requested supplier materials and comments on “ABCI-Q System A,” an expansion intended to add advanced AI computing and an execution platform for AI agents. The notice describes a proposed system with at least 3.5 exaflops of theoretical FP4 inference performance, at least 0.25 exaflops of FP16 training performance and total power consumption no higher than 300 kilowatts.
The notice sets the planned introduction for the fourth quarter of fiscal 2027 and says the requirements may change. Its technical brief calls for software supporting large-model development, quantum-circuit simulation and persistent AI-agent services. The public request for supplier materials runs through November 13, 2026.