Japan’s largest power producer, JERA, has signed a non-binding memorandum of understanding with Dell Technologies and UK-based artificial intelligence infrastructure specialist RHAELM to build high-density AI data centers across the country. According to reporting by the Financial Times, the broader multi-site initiative could involve up to $140 billion (approximately ¥21 trillion) in cumulative capital expenditure, aiming to establish 3 to 4 gigawatts (GW) of dedicated computing capacity during the 2030s.
The initial deployment focuses on a $15 billion (¥2.3 trillion), 400-megawatt (MW) campus situated directly adjacent to JERA’s gas-fired Chiba Thermal Power Station, east of Tokyo. Backed by private equity giant Apollo Global Management as RHAELM’s strategic financing partner, the flagship Chiba project is scheduled to begin commercial operations around 2028. Once completed across all phases, it will rank among the largest AI computing facilities in Asia outside mainland China.
Key Takeaways
- Nationwide $140 Billion Scale: The partnership creates an industrialized framework to roll out standardized AI compute campuses across Japan, with industry estimates putting the multi-site ambition at up to $140 billion (3–4 GW of capacity).
- Flagship 400MW Chiba Facility: The first project will be constructed on JERA-owned land next to the Chiba Thermal Power Station, representing a multi-phase investment of more than $15 billion.
- Skipping the Grid Queue: By co-locating directly beside JERA’s generation facilities (“behind the meter”), the project bypasses Japan’s multi-year power grid interconnection backlogs to deliver capacity ahead of conventional timelines.
- Defined Division of Labor: JERA supplies industrial land and direct generation; Dell supplies pre-configured server racks and AI compute platforms; RHAELM leads engineering and facility management; and Apollo provides structured private equity financing.
- Sovereign AI Imperative: The initiative aligns with Tokyo’s strategic push to secure domestic computing autonomy, addressing concerns voiced by policymakers that Japan could risk becoming a digital dependency if it lacks domestic AI infrastructure.
- Operations Slated for 2028: The Chiba campus targets initial commercial availability around 2028, acting as a modular template for subsequent gigawatt-scale sites.
Central Question: Why Is Japan Pursuing a $140 Billion AI Data Centre Blueprint?
Direct Answer: Generative AI models and high-density GPU server clusters require vast amounts of steady, low-latency electric power that traditional national grids struggle to provide without lengthy interconnection queues. By partnering JERA (which generates roughly one-third of Japan’s total electricity) with Dell’s high-performance server supply chain and Apollo’s capital, Japan is creating an integrated power-to-compute model. Co-locating data centers directly beside thermal power plants allows developers to bypass grid transmission bottlenecks, build 400MW to gigawatt-scale campuses years faster, and establish sovereign AI capacity on Japanese soil.
THE INTEGRATED AI INFRASTRUCTURE MODEL
│
┌───────────────────────────────────┼───────────────────────────────────┐
▼ ▼ ▼
JERA (POWER & REAL ESTATE) DELL TECHNOLOGIES (COMPUTE) RHAELM & APOLLO (BUILD & CAPITAL)
• Supplies direct power off-grid • Provides turnkey AI server racks • Oversees engineering and modular build
• Industrial land at Chiba plant • High-density liquid cooling • Apollo backs project equity financing
• Eliminates multi-year grid delays • Enterprise hardware maintenance • Repeatable nationwide scaling blueprint
│ │ │
└───────────────────────────────────┼───────────────────────────────────┘
▼
THE CHIBA PILOT (400MW)
• $15B+ total capital commitment
• Operational target around 2028
• Template for 3–4 GW national pipeline
1. Project Architecture: The Chiba 400MW Flagship
The initial phase at the Chiba Thermal Power Station addresses the primary friction points of modern AI cluster deployment: power access, land zoning, and hardware integration.
+-----------------------------------------------------------------------------------+
| JERA-DELL-RHAELM CHIBA AI DATA CENTRE SPECIFICATIONS |
+-----------------------------------------------------------------------------------+
| Project Dimension | Operational Details |
+--------------------------------+---------------------------------------------------+
| **Location** | Chiba Thermal Power Station site (East of Tokyo) |
| **Power Capacity** | **Up to 400 Megawatts (MW)** |
| **Estimated Investment** | **>$15 Billion (¥2.3 Trillion)** across all phases|
| **Target Launch Window** | **Commercial operations starting ~2028** |
| **Power Sourcing Model** | Direct, behind-the-meter gas-fired generation |
| **Primary Compute Partner** | Dell Technologies (Pre-racked AI infrastructure) |
| **Lead Developer** | RHAELM (UK AI infrastructure developer) |
| **Financing Partner** | Apollo Global Management |
+--------------------------------+---------------------------------------------------+
1. Bypassing the Grid Interconnection Bottleneck
In major global data center hubs (such as Northern Virginia, Frankfurt, and Tokyo), utilities often quote waiting periods of five to seven years for new high-voltage grid connections.
- By drawing electricity directly from JERA’s operating generation station, the Chiba facility operates on a behind-the-meter architecture.
- Bradd Lewis, Chief Executive of RHAELM, noted: “We can deliver a 400MW facility years ahead of a conventional grid-connected timeline.”
2. Turnkey Compute Standardization
Dell Technologies is configuring pre-integrated, liquid-cooled server racks optimized for modern AI accelerator clusters. Standardizing rack layouts, power distribution units (PDUs), and thermal cooling manifolds across all 400 megawatts turns the facility into a modular computing factory, allowing hardware refreshes as new accelerator architectures debut toward the end of the decade.
2. National-Scale Ambition: Moving from 400MW to 4 Gigawatts
While JERA has officially confirmed the Chiba memorandum and its intent to pursue multi-gigawatt deployments in the 2030s, the broader $140 billion scope reported by the Financial Times outlines a multi-site expansion:
THE EXPANSION ROADWAY (2026–2030s)
│
Phase 1: Chiba Hub (2026–2028) ──► Phase 2: JERA Port Replicas ──► Phase 3: National Scale
• 400MW initial capacity • 1 to 2 GW additional load • 3 to 4 GW total footprint
• $15B capital allocation • Replicating standardized model • Up to $140B total investment
- Repeatable Infrastructure: JERA operates dozens of coastal thermal facilities, fuel import terminals, and industrial plots across Japan. If the Chiba template succeeds, the consortium plans to duplicate the design at other generation hubs, distributing compute load away from congested metropolitan Tokyo while keeping latency minimal.
- Industrializing AI Delivery: Instead of designing bespoke data center buildings for every customer, the consortium is shifting the model toward mass-produced, standardized campuses capable of hosting multinational hyperscalers, domestic research institutes, and enterprise AI workloads.
3. Geopolitical and Strategic Context: Japan’s Sovereign AI Race
The mega-project arrives amidst intense competition among global investors to secure Asian AI capacity:
+-----------------------------------------------------------------------------------+
| MAJOR AI INFRASTRUCTURE INVESTMENTS IN JAPAN |
+-----------------------------------------------------------------------------------+
| Investor / Consortium | Committed / Target Capital | Focus & Strategy |
+--------------------------------+----------------------------+-------------------------------------+
| **JERA / Dell / RHAELM** | **$15B – $140B** | 400MW to 4GW behind-the-meter sites |
| **Blackstone** | **$30 Billion** | Multi-year AI data centre expansion |
| **Microsoft** | **$2.9 Billion** | Hyperscale cloud & AI data centers |
| **Oracle** | **$8.0 Billion** | Sovereign cloud & data residency |
+--------------------------------+----------------------------+-------------------------------------+
- Preventing “Digital Colonization”: Japanese government ministers have repeatedly emphasized the risk of digital dependency if local companies rely entirely on foreign-hosted cloud infrastructure. Developing sovereign compute clusters ensures sensitive industrial, robotic, and financial data remains governed under domestic jurisdiction.
- Private Equity Capital Rotation: The involvement of Apollo Global Management reflects a broader shift by global alternative asset managers, who view power-secured AI data centers as essential digital infrastructure alongside pipelines, toll roads, and energy grids.
Frequently Asked Questions (FAQs)
What is the $140 billion Japan AI data centre project?
The $140 billion initiative refers to a long-term, multi-site plan reported by the Financial Times involving power producer JERA, Dell Technologies, and developer RHAELM (financed by Apollo) to build up to 3 to 4 gigawatts of AI data center capacity across Japan during the 2030s.
Where is the first data centre being built?
The first campus will be constructed adjacent to JERA’s gas-fired Chiba Thermal Power Station, located east of Tokyo. It will have a capacity of 400 megawatts and is estimated to cost more than $15 billion across all development phases.
Why is the data centre being built next to a power plant?
Building directly next to a power plant allows the facility to take electricity directly (“behind the meter”) without waiting years for public grid upgrades, shortening construction and deployment timelines.
When will the Chiba data centre begin operations?
Commercial operations for the initial phase of the 400MW Chiba facility are slated to begin around 2028.
What are the respective roles of Dell, JERA, and RHAELM?
JERA provides the land and direct power supply; Dell Technologies supplies the pre-configured server racks and computing hardware; RHAELM manages design and construction; and Apollo Global Management acts as the lead financing partner.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
