Japan’s planned 400-megawatt Chiba AI campus starts from a different premise than a conventional grid connection: the power station is part of the data-centre design. JERA, Dell Technologies and RHAELM say the first project will sit beside JERA’s operating Chiba Thermal Power Station, use behind-the-meter power and target operations around 2028.[1]
The companies expect capital deployment above $15 billion across land, power infrastructure, facilities and AI compute. The integration is intended to remove repeated project-by-project design work by standardising the connection between fuel, generation, electrical systems, cooling and rack-scale computing. That may shorten the route to power, but it also concentrates energy, technology and financing assumptions inside one model.[1]
The Chiba project as announced
| Measure | Value | Context |
|---|---|---|
| Power capacity | Up to 400 MW | Behind-the-meter capacity adjacent to an operating JERA generation asset.Source 1 |
| Capital deployment | More than $15 billion | Across all phases, including land, power, construction and compute.Source 1 |
| Targeted operations | Around 2028 | A target, not a confirmed commercial-operation date.Source 1 |
| Energy foundation | Gas-fired generation and LNG value chain | JERA describes fuel procurement, shipping, import and generation as part of the model.Source 1 |
| Scale ambition | Multi-gigawatt in the 2030s | The partners intend to explore the model at other JERA sites.Source 1 |
A power rating becomes an annual system when utilisation is specified
If a 400-megawatt campus averaged 90% of its maximum load for a year, it would consume about 3.15 terawatt-hours. That is an illustrative calculation: 400 MW multiplied by 8,760 hours and by 0.90. The actual result will depend on build phases, hardware occupancy, cooling, maintenance and the shape of workloads.
The calculation matters because “400 MW” can describe an electrical ceiling while annual fuel use, emissions and grid effects depend on energy over time. A project can advertise the same peak capacity while producing very different annual outcomes. Public reporting should therefore pair maximum megawatts with average load, annual megawatt-hours and the source of that electricity.
Integrated build: what may improve and what still needs proof
| Design choice | Potential advantage | Evidence still needed |
|---|---|---|
| Behind-the-meter siting | Shorter route to power than a conventional new grid connection | Operating limits, backup arrangements and any grid exchangesSource 1 |
| Standardised compute layer | Faster repeat deployment and fewer bespoke interfaces | Measured commissioning time and reliability across sitesSource 1 |
| Integrated LNG and generation | Fuel and power organised by one major energy company | Fuel-price exposure, emissions performance and transition planSource 1 |
| Flexible computing software | Ability to adjust lower-priority work around grid conditions | Verified response time, duration, rebound and effect on useful workSource 2 |
Behind the meter is not outside the energy system
Locating compute beside a generator can reduce dependence on a new transmission path. It does not make the project independent of fuel markets, maintenance, local environmental conditions or the wider grid. JERA’s announcement explicitly roots the model in reliable gas generation and the LNG chain. The energy-security benefit and the carbon consequence must therefore be assessed together.[1]
The IEA’s East Asia work treats data-centre demand as a grid-planning issue, not only a technology-sector procurement decision. Japan’s constraints include the pace of generation, transmission and telecommunications development. Chiba’s “watt-bit” approach is an attempt to plan those layers together; its evidence will come from delivery, not from the elegance of the concept.[3]
Chiba is important because it turns the energy plant from a supplier into a component of the AI campus. That can accelerate construction, but it also means the quality of the AI infrastructure cannot be judged without the fuel contract, generator, electrical architecture and operating rules. The data centre begins at the power station boundary, not at the first rack.
Sources & further reading
- JERA, Dell Technologies and RHAELM Sign MoU to Accelerate National-Scale AI Infrastructure Development in Japan JERA · 1 Oct 2026 · Accessed 8 Oct 2026
- JERA Invests in U.S.-Based Emerald AI to Advance Energy Solutions for AI Data Centers JERA · 26 Aug 2026 · Accessed 8 Oct 2026
- Energy and AI in East Asia International Energy Agency · 27 Mar 2026 · Accessed 8 Oct 2026
- Energy demand from AI International Energy Agency · Accessed 8 Oct 2026