⚛️ Most quantum computers live inside a refrigerator colder than deep space. This one sits in an ordinary room in Tsukuba, at ordinary pressure, and it was switched on last month. Its name is MoQuren, and the people who built it are betting that the quantum computer that wins will be the one you can actually put in a data centre.

No refrigerator, no vacuum chamber

On July 21, 2026, Japan's National Institute of Advanced Industrial Science and Technology (AIST) and a Tokyo startup called OptQC announced that they had finished building a photonic quantum computer and started running it. The machine, MoQuren, sits at AIST's Global Research and Development Center for Business by Quantum-AI Technology, known as G-QuAT, on the institute's campus in Tsukuba, about an hour north of Tokyo.

Both organisations call it a world first, and the claim is worth reading carefully. What they say is new is not photonic quantum computing itself, but that a domestically built machine intended for commercial service has been installed in a working research and development environment for the first time.

Set against the rest of the field, that is a narrow claim. Photonic machines have already moved into European research centres. Quandela of France delivered a 12-qubit system called Lucy to the CEA's TGCC computing centre in October 2025, where it is being coupled to a classical supercomputer, and QuiX Quantum of the Netherlands says it has been selling photonic computers to the German Aerospace Center since 2022.

Superconducting machines, the kind Google and IBM build, need dilution refrigerators that hold their chips a hair above absolute zero, which is why press photographs always show that gold chandelier hanging in a cylinder. Trapped-ion machines need ultra-high vacuum. Photonic machines encode information in light, so they run at room temperature and ordinary atmospheric pressure. Nothing has to be isolated from the world in quite the same way.

A second advantage matters more commercially than it sounds. Because the machine works with light in optical fibre, it speaks the same physical language as the telecoms industry, which has spent forty years and enormous sums making optical components cheap and reliable. AIST and OptQC frame the payoff in energy terms: as AI pushes data-centre electricity consumption up, a computer that needs no refrigeration and can be made small starts to look less like a physics experiment and more like equipment.

What the machine can actually do

According to OptQC's published development roadmap, this first unit is the company's "maintenance-free model", built for stable continuous operation rather than raw power. The listed specifications are 100 inputs of the analogue quantum kind and a 100 MHz operating clock, with optimisation solvers and neural networks named as the target applications. That is not a million qubits. It is not a hundred error-corrected logical qubits either. It is a machine that engineers can leave running and that software developers can write code against.

MoQuren is slated to become "System O" inside ABCI-Q, the hybrid quantum-classical computing platform AIST is assembling at G-QuAT. The other components are unusual company to keep: System H is a GPU supercomputer rated at 138 PFLOPS in double precision, System F is a Fujitsu superconducting machine with 64 physical qubits, and System Q is a QuEra neutral-atom machine with 260 physical qubits. Four different bets on the future of computing, inside one building.

The integration is not finished. AIST's own operational status page, in its update dated July 30, 2026, still lists System O as in preparation while H, F and Q are in service. OptQC says it is first developing a software development kit with a built-in simulator, so that developers can test algorithms on their own PCs before they ever touch the hardware. Japan's Science Portal reported on August 3 that cloud access for companies and research institutions is planned before the end of 2026.

From a Tokyo lab bench to a corporate roadmap

OptQC was incorporated on September 2, 2024, spun out of Akira Furusawa's laboratory in the University of Tokyo's Graduate School of Engineering. Furusawa is the physicist who demonstrated the world's first deterministic quantum teleportation at Caltech in 1998 and then spent two decades turning it into a computing architecture, a story we have told before.

The founding team came straight off that lab bench. CEO Kan Takase took his doctorate there and stayed on as an assistant professor before leaving to start the company. CTO Warit Asavanant arrived from Thailand in 2011 as an undergraduate, studied under Furusawa from his first degree through his doctorate, and now handles the large-scale entanglement that serves as the machine's processor. Furusawa is a director, as is Hidehiro Yonezawa, who leads RIKEN's quantum optical control research team. The squeezed light that machines like this one run on hit a new quality record this year, in work a research team including OptQC and NTT announced in March.

The company's own timeline lists a seed round of 650 million yen, about $4.1 million at the 157 yen to the dollar rate at which the currency closed on July 31, 2026. Public money is in the mix as well, through a Cabinet Office programme called BRIDGE.

Thirteen days after MoQuren started running, on August 3, 2026, OptQC announced a capital and business alliance with NTT. The two companies have signed a joint research contract running through the 2027 fiscal year, aimed at completing the architecture and core component design for a fault-tolerant photonic machine on the scale of a million physical qubits. Their stated goal is to realise it by fiscal 2030, with a 10,000-qubit-class machine by fiscal 2027 and customer proof-of-concept work on it in fiscal 2028. Those are announced targets, not results, and they sit a long way from a 100-input analogue machine.

Three countries, three different bets

The American giants have mostly bet on superconducting circuits. Google's Willow, announced in December 2024, is a 105-qubit chip that showed error rates dropping exponentially as the error-correcting code got bigger, the "below threshold" result the field had waited years for; Google has said its next target is a long-lived logical qubit. IBM has published a dated roadmap to Starling, a system it says will run 100 million gates on 200 logical qubits by 2029 at its site in Poughkeepsie, New York. Both live in refrigerators.

Photonics, though, is not a Japanese speciality, as those European machines already show. The heaviest bet on it is American. PsiQuantum, in Palo Alto, raised a $1 billion Series E in September 2025 at a $7 billion valuation and is breaking ground on utility-scale sites in Chicago and Brisbane. Put OptQC's 650 million yen seed round next to that and the asymmetry is not subtle.

China runs both tracks at once, and out of the same laboratory. The University of Science and Technology of China built Zuchongzhi, the country's superconducting line, whose third generation arrived in March 2025 carrying 105 qubits, the same count as Willow. USTC also builds the photonic Jiuzhang series. In May 2026, a team there led by Chao-Yang Lu and Jian-Wei Pan published Jiuzhang 4.0 in Nature: 1,024 high-efficiency squeezed light sources feeding an 8,176-mode circuit, producing detection events of up to 3,050 photons, an order of magnitude beyond the 2023 version. On raw physics, it is the most impressive photonic device on this list.

It is also built to do one thing. Gaussian boson sampling is a task chosen because classical computers find it punishing, not because a customer needs the answer, though the Nature paper does note that it can generate the bosonic codes fault tolerance will require. For now Jiuzhang is an instrument for settling a question about quantum advantage. MoQuren is a product being wired into a platform that already has a published price list.

That gap is the Japanese bet in miniature. Not the biggest machine and not the most photons, but a commercial one, built at home and slotted into a national platform where three rival modalities already run. Whether outside customers actually turn up is still an open question.

The energy argument

Ask why any of this matters outside a physics department and OptQC keeps returning to electricity. Take the refrigerator out of the power budget, and put the computer somewhere the existing optical network already reaches. That is the company's answer. Whether it holds is unsettled.

Photonics has its own unforgiving problem, which is that photons get lost, and every lost photon is a piece of the computation gone. Roadmaps in this field slip routinely. And the machine now humming in Tsukuba is, for the moment, a stable 100-input device aimed at optimisation problems, not a general-purpose quantum computer.

Still, something did change on July 21. For years the photonic case was made in papers. Now it is being made by a machine sitting in a room, next to a superconducting machine and a neutral-atom machine, waiting for someone to log in.

When quantum computing comes up where you live, which approach gets named first? And has anyone there started asking what it will cost to keep the thing running?

参照