Reimei Upgraded to 56-Qubit H2 as Quantinuum Widens RIKEN Deal

Quantinuum and RIKEN completed the upgrade of the Reimei quantum computer to the 56-qubit System Model H2 on November 10, 2026, and used the occasion to announce a broader partnership covering a fault-tolerant "Made with Japan" quantum computer and a separate SoftBank agreement on quantum-AI hybrid services for Japanese AI data centers. The hardware move itself was old news: an April 14, 2026 announcement had already disclosed that H2 was delivered for assembly to replace the earlier H1 system. What November delivered was the finished upgrade plus new commitments — commitments that Quantinuum itself cautions may not lead to definitive agreements, joint development or commercial services.

RIKEN marked the expansion with a ceremony at its Wako campus. Mitsuhisa Sato, director of RIKEN's Quantum-HPC Hybrid Platform Division, said the upgraded machine should address larger and more complex scientific problems than the previous H1 system. A newly signed memorandum of understanding is intended to expand research into quantum computing for AI for Science — a field where quantum processors assist machine-learning workflows on scientific problems rather than replacing classical machines outright.

Why the upgrade matters

Reimei sits inside one of the most closely watched experiments in hybrid computing: a Japanese national research program pairing a trapped-ion quantum processor (qubits held in electromagnetic traps, here described by Quantinuum as fully connected and high-fidelity) with some of the world's most powerful classical infrastructure. Reimei remains part of the JHPC-quantum project funded by Japan's New Energy and Industrial Technology Development Organization, which places the current work inside a supported research program rather than a commercial deployment.

The stakes for Quantinuum are straightforward. The company needs evidence that its hardware produces results classical systems cannot, at a scale that matters, inside workflows researchers actually use. The upgrade from H1 to H2 roughly expands the machine's usable quantum resources; the partnership's credibility now rests on whether early research gains survive larger runs.

How do the three systems divide the work?

Quantinuum's technical account assigns each machine a distinct role:

• Fugaku, RIKEN's flagship supercomputer, handles large-scale classical modeling and established scientific computing.
• ROQUO performs tensor-network simulation (a classical method that compresses quantum states), AI training and massively parallel computation.
• Reimei takes selected calculations involving quantum states and correlations that become prohibitively expensive to represent classically.

Quantinuum Nexus, the company's hybrid workflow manager, coordinates the resources. Jobs running on classical infrastructure can submit quantum tasks to Reimei and fold the results back into larger calculations. The design gives the quantum processor a focused role inside a broader scientific workflow instead of forcing it to handle an entire problem — a pragmatic division given current hardware limits.

NVIDIA reported in July 2026 that ROQUO uses 540 Blackwell GPUs accessed through the GB200 NVL4 platform and connects to on-premises quantum systems, including Reimei. Researchers were beginning to explore an evolutionary AI framework, developed with NVIDIA and integrated with its CUDA-Q platform for quantum-classical computing, to generate Reimei circuits automatically rather than by hand.

What have the early quantum-AI projects shown?

Two research projects predate the public launch of Reimei's services, and both now report progress — with explicit caveats attached.

The materials project combines quantum simulation, classical tensor networks and generative AI to construct phase diagrams, which map how material properties change with temperature, pressure or magnetic field. ROQUO handles dense tensor-network simulations in regions where classical methods remain tractable. Reimei addresses deep time-evolution circuits, where long-range correlations and growing entanglement drive classical simulation costs beyond reach.

The team has completed concrete steps:

• Collected Reimei hardware data for a 6 × 6 two-dimensional spin model
• Trained a Transformer model for zero-noise extrapolation (a technique that estimates noiseless results from noisy quantum data) across the parameter space
• Applied active learning to prioritize which additional quantum data to collect
• Trained a generative model on the combined classical and quantum dataset, running on ROQUO to sample conditions not yet simulated

Quantinuum reports preliminary results suggesting the model makes more accurate predictions than AI trained only on classical simulations, and more precise predictions than Reimei alone. The company immediately qualifies the claim: the experimental findings may not be confirmed at larger scale or in commercial use. Better atomic-level understanding of magnetism could eventually inform ultra-high-field MRI and smart power grids — but those are potential applications, not deployments.

A second project, involving RIKEN, NVIDIA and Quantinuum, is developing a scalable workflow spanning quantum processors, GPUs and high-performance computing. Its initial target is zinc porphyrin, a light-absorbing molecule, with an emphasis on excited electronic states involved in light absorption rather than only ground-state searches — a harder and more chemically relevant problem. Here too the labor is split: Fugaku represents the larger molecular environment and chemistry workflow; ROQUO handles optimization, configuration-interaction calculations, AI-assisted circuit design and spectral reconstruction; Reimei searches strongly correlated regions for important electronic configurations. Classical systems then use those candidate configurations to compute energies and reconstruct experimentally meaningful properties, including the molecule's absorption spectrum. This division avoids requiring the quantum processor to represent the entire biomolecular environment.

After both projects conclude, Reimei is scheduled to begin regular operations under the Test User Program — the point at which external researchers, not just project insiders, get access.

What do the commercialization plans actually commit to?

The "Made with Japan" research plan combines Quantinuum's technology with Japanese university research and domestic component suppliers' manufacturing expertise, drawing on lessons from integrating and operating Reimei. The framing signals intent to build industrial capacity inside Japan rather than import a finished machine. No timeline, budget or qubit-count target appears in the announcement.

The SoftBank agreement is exploratory by design. The three organizations — Quantinuum, RIKEN and SoftBank — will assess infrastructure readiness and examine direct connections between quantum computers and AI data centers, including whether such links could support commercial hybrid services beyond the research community. The question of whether data-center operators need on-site quantum access remains open; the partners have only agreed to study it.

Quantinuum's own disclosure flags the limits of both documents. The memorandum and the SoftBank agreement may not produce definitive agreements, joint development or commercial services. That disclaimer is standard, but it defines the gap between November's announcements and any revenue.

The test ahead

The upgrade establishes a hardware and workflow foundation. The commercial test is twofold: reproducing the early research gains at larger scale, and converting the SoftBank exploration into services that reach beyond RIKEN's research setting. Reimei now has the qubits and the orchestration software. Whether it has customers is a question for 2027 and beyond.