Jera, Dell, and Rhaelm Plan $15B Behind-the-Meter AI Data Center
Hyperscale 400 MW facility at Chiba Thermal Power Station aims to bypass Japan's power grid delays
Japanese energy giant Jera, Dell Technologies, and AI provider Rhaelm have signed a Memorandum of Understanding to build national-scale AI infrastructure across Japan. The initial project is a 400 MW behind-the-meter AI data center co-located at Jera's Chiba Thermal Power Station, backed by Apollo Global Management with an estimated $15 billion total investment.
Key details
Under the agreement, Jera, Dell Technologies, and Rhaelm are establishing a standardized framework spanning electrical infrastructure, cooling systems, and rack-scale compute to accelerate AI data center deployment. The flagship site at Chiba Thermal Power Station—a liquefied natural gas facility with 4.38 GW of total capacity in Chiba Prefecture—will host up to 400 MW of direct-connected AI compute capacity.
The project is structured on a behind-the-meter basis, drawing power directly from the thermal power plant to avoid conventional electric grid interconnection backlogs. Across all development phases, including land acquisition, power infrastructure, building construction, and Dell AI Factory compute hardware, the project represents an estimated $15 billion investment. Operations at the Chiba campus are targeted to begin in or around 2028.
Why this matters
Connecting 400 MW of high-density compute directly to an existing 4.38 GW LNG power plant allows developers to bypass multi-year transmission queue backlogs that currently bottleneck hyperscale AI buildouts. However, generating power via dedicated fossil fuel infrastructure highlights the environmental tradeoffs of rapid AI scaling, as behind-the-meter natural gas consumption directly increases operational emissions to power AI training and inference workloads.
Context
Japan consumes significant energy, with Jera supplying roughly one-third of the nation's total electricity and nearly all the power for the Tokyo metropolitan region. As public utility grids face mounting pressure from AI energy demands, hyperscalers and sovereign AI developers are increasingly turning to behind-the-meter configurations at thermal generation plants. Jera and Rhaelm plan to replicate this deployment model across more than 20 Jera sites, targeting a multi-gigawatt sovereign AI footprint across Japan in the 2030s.
Risks and open questions
Key concerns involve the long-term carbon footprint of powering AI infrastructure with liquefied natural gas rather than renewable sources. Additionally, securing continuous fuel supply and managing high-density liquid cooling requirements at thermal power sites present technical and operational challenges as power densities per rack continue to rise.
What happens next
The consortium will finalize site engineering and capital allocation with Apollo Global Management while preparing the Chiba facility for construction. Industry observers will monitor whether this behind-the-meter framework is successfully expanded across Jera's remaining 20 power station locations across Japan.
Source: Data Center Dynamics Published on AI Usage Global, author: AUG Bot
