RIKEN Upgrades Reimei to 56-Qubit H2, Signs New Quantinuum MoU

RIKEN has upgraded its Reimei quantum computer from Quantinuum's System Model H1 to the 56-qubit H2, and the partners signed a new memorandum of understanding on Tuesday to expand quantum work under RIKEN's AI for Science program. The announcement disclosed no budget, timeline or financial terms. That absence matters. It leaves the promised fault-tolerant "Made with Japan" quantum computer as an aspiration, not a funded roadmap.

What does the new agreement cover?

The MoU covers three commitments. First, extended research into quantum computing for RIKEN's AI for Science initiative. Second, joint R&D on a fault-tolerant "Made with Japan" quantum computer built with Japanese universities and domestic component suppliers. Third, a separate agreement with SoftBank Corp. to explore connecting quantum computers directly to AI data centers in Japan.

The release gave no date for delivery of the domestic machine. It named no suppliers. It specified no funding. A skeptic notes what is missing before crediting what is present.

Reimei carries a short but concrete history. RIKEN installed it as an H1 system in February 2025 and launched it that spring as part of a hybrid platform with Fugaku, its flagship supercomputer. RIKEN announced in April that it had bought the 56-qubit H2 as a replacement. The system runs under the JHPC-quantum project, funded by NEDO, the technology agency of Japan's Ministry of Economy, Trade and Industry.

What have researchers actually shown?

Two research projects began on the upgraded machine before its services open to outside users. Both produced preliminary claims. Neither has been published for independent review.

In the first project, researchers used RIKEN's ROQUO GPU supercomputer to train an AI model on quantum data generated by Reimei to simulate magnetism. The release states the model predicts better than one trained only on classical simulations and is more precise than the quantum computer alone. The stated targets include ultra-high-field MRI and smart power grids. These are claims from the partners, not peer-reviewed findings.

The second project combines Reimei, Fugaku and an Nvidia evolutionary AI agent running on ROQUO to simulate a photoactive biomolecule. The partners link this work to more efficient solar cells. Again, no published results. Again, the promise outpaces the evidence by a comfortable margin.

Both executives framed the deal in expansive terms. "The newly signed Memorandum of Understanding will pave the way for expanding our research" into quantum computing for AI for Science, said Mitsuhisa Sato, who leads RIKEN's Quantum-HPC Hybrid Platform Division. Quantinuum CEO Rajeeb Hazra said Japan's ambitions "require advanced capabilities today and a sustained commitment to developing what comes next."

Sustained commitment is exactly what the release does not quantify.

Why is SoftBank involved?

The SoftBank Corp. agreement adds a commercial dimension. The three parties will assess whether Japanese infrastructure is ready to connect quantum computers directly to AI data centers, and how linked systems could deliver commercial quantum-AI hybrid services. "Our responsibility is to turn these achievements into industrial value," said Takenori Kobayashi, head of product R&D at SoftBank Corp.

That is an assessment, not a deployment. No data centers are named. No connection dates appear.

The deal does extend Quantinuum's commercial momentum. Last week the company raised its 2026 revenue outlook to $31-34 million after signing its largest cloud deal. That figure frames the scale honestly: this is a company with tens of millions in guided revenue, pairing with a national lab and a telecom operator to build demand that does not yet exist.

What comes next?

The commercial test is whether the NEDO-funded hybrid platform converts preliminary quantum-AI results into published, reproducible science, and whether SoftBank's infrastructure study yields a concrete data center integration. Until then, the partnership rests on a 56-qubit machine, two MoUs and revenue guidance under $35 million — a real base, but a narrow one for promises of fault tolerance.