GMI Cloud plans USD $12bn sovereign AI site in Japan

GMI Cloud has outlined a USD $12 billion plan for a large-scale sovereign AI infrastructure site in Kagoshima, Japan, designed to scale to 1 gigawatt of power capacity.

The project centres on an "AI Factory" that GMI Cloud plans to develop with manufacturing group Wistron. The broader development includes local and financial partners, including a joint venture led by CDIB Capital and Shinetsu Science Industry, as well as the Kagoshima Prefectural Government and Satsumasendai City.

Construction is expected to start in late 2026. The site is positioned as part of a broader push by governments to secure domestic control over AI computing resources and reduce exposure to foreign platforms, data-jurisdiction risks, and supply-chain disruption.

Japan focus

GMI Cloud calls the Kagoshima facility Japan's first domestically built AI Factory designed to support "Physical AI", which it links to robotics, autonomous vehicles, manufacturing, and industrial infrastructure.

Japan already has a strong base in industrial automation and advanced manufacturing. The initiative reflects a view among some technology suppliers and governments that AI compute has become a strategic asset, alongside energy and communications infrastructure.

Alex Yeh, Chief Executive of GMI Cloud, said the project is intended to keep domestic control over intelligent systems operating in the physical economy.

"Japan has built some of the world's most sophisticated industrial and manufacturing systems," said Alex Yeh, CEO of GMI Cloud. "The next frontier is ensuring those systems are powered by AI that Japan owns, controls, and can trust. That is exactly what we are here to build."

Partners named

Wistron, best known as a contract manufacturer across electronics and computing supply chains, is expected to support delivery of the facility. David Shen, Chief Technology Officer at Wistron, said the project's scale required a partner with experience deploying advanced hardware in production environments.

"Building a sovereign AI Factory at this scale requires a partner with a strong track record. GMI Cloud's demonstrated experience in deploying the most advanced hardware into production-grade AI infrastructure is exactly what this project demands. Wistron is designated to support that vision with the manufacturing excellence it requires," said David Shen, CTO, Wistron Corporation.

Data infrastructure supplier Vast Data is also involved and is positioning its "AI Operating System" as part of the deployment. The company said its technology supports data governance and orchestration for the planned facility.

"GMI Cloud is setting the standard for sovereign AI infrastructure through its Kagoshima AI Factory, designed to deliver national-scale performance and operational control. VAST's AI Operating System provides the unified data foundation that enables the Factory's persistent memory, governance, and data orchestration at scale, enabling sovereign Physical AI systems to operate with performance and trust engineered into the architecture," said Jeff Denworth, Co-Founder at VAST Data.

Energy and footprint

The project's stated maximum capacity of 1GW would put it among the larger single-site compute developments globally, if delivered at that level. No timeline was provided for reaching that peak, and there was no detailed breakdown of the expected compute build-out, including the types and quantities of chips to be used.

GMI Cloud said it would build the facility with a focus on sustainability and described it as a reference architecture for "environment-friendly sovereign AI infrastructure at scale". The announcement did not include details on power sourcing, grid connections, cooling design, or energy-efficiency targets.

Sovereign shift

Sovereign AI has become a prominent theme among cloud providers, chipmakers, and governments. It is generally tied to controlling where data is stored, who can access it, and who governs the computing infrastructure used to train and run AI models.

The Kagoshima plan follows a model that pairs large infrastructure investment with local partnerships and government involvement. The development entity for the location is Kai Shin Digital Infrastructure, described as a joint venture structured by CDIB Capital and Shinetsu Science Industry in collaboration with regional and city authorities.

GMI Cloud, based in Silicon Valley, describes itself as a provider of GPU-as-a-Service and "inference-first" AI infrastructure. It says it has deployed AI infrastructure across multiple continents and regulatory environments, but did not disclose current revenue, customer numbers, or the size of its existing data-centre footprint.

For Japan, the planned facility adds another data point in a region where governments and corporates have increased spending on compute infrastructure while seeking stronger assurances on data handling and operational control. Work is expected to begin in late 2026.