For anyone building AI infrastructure in Japan, power availability has become the defining decision that shapes site choice, timeline, and customer commitments, as a new grid connection in Tokyo can take five to ten years to secure. A new arrangement between Japan’s largest power generator and two technology partners offers a new option.
On October 1, JERA, Japan’s largest power generator, signed a memorandum of understanding (MoU) with Dell Technologies and RHAELM Holdings, a London-based sovereign AI infrastructure developer, to plan an AI data center of up to 400 MW at its gas-fired power station in Chiba, near Tokyo. The facility would take electricity directly from the plant, sidestepping the years-long wait for a grid connection.
Apollo Global Management has signaled its intent to finance, with total capital deployment expected to exceed US$15 billion. Operations are targeted around 2028, under a power supply contract running 15 to 25 years.
The same day, Japan tightened its connection rules to clear out developers who were holding queue positions without serious intent to build. From October 1, any customer who does not pay the required contribution within three months of approval loses their place in line. The new requirement applies to extra-high-voltage customers, a category that covers large data centers.
For developers and tenants weighing whether to follow JERA’s model, getting power sooner is a real benefit. What is less clear is who carries the risks that come with depending on a single plant, and on what terms.
The Chiba Project: What Is Confirmed and What Is Not
The Chiba project would be the largest single-site AI infrastructure deployment in Japan, if built. So far, the partners have confirmed the location, a capacity of up to 400 MW, a target opening around 2028, and a power contract of 15 to 25 years.
What is yet to be confirmed: a final investment decision, committed financing, and a named anchor customer. For a project of this scale, all three would normally need to be in place before construction begins, though the involvement of JERA and Apollo gives the announcement more weight than a typical early-stage MoU.
A final investment decision is when partners formally commit to building, with construction contracts being signed and the site being prepared. Until it is made, every other timeline, including the 2028 target, is contingent. Confirmed financing means the capital has a committed source with agreed terms. Apollo has signaled intent to finance RHAELM’s share, but at US$15 billion across all phases, intent and commitment are different things. An anchor customer is the first major tenant who signs before the facility is built, committing to take defined capacity at a defined price. It is their contracted revenue that lenders use to underwrite the project’s economics.
Taken together, they describe a project with a credible foundation and significant details still to be resolved.
The Risks of Taking Power Straight From the Plant
For operators who need capacity by 2028, a behind-the-meter arrangement is not a marginal advantage. But taking power directly from a single plant raises three questions.
How is the electricity priced? The Chiba Thermal Power Station runs on LNG (liquefied natural gas), which Japan imports almost entirely from overseas and is priced in US dollars on global markets. That means the cost of generating electricity at Chiba moves with global gas prices and with the yen-dollar exchange rate. Japan has already seen what that exposure looks like in practice: when LNG spot prices spiked during the Hormuz crisis earlier this year, Tokyo Gas suspended all new high-voltage and extra-high-voltage contracts to prevent negative margins.
Whether that volatility lands on the tenant depends on how the power agreement is written. Under a pass-through arrangement, the tenant pays a price that moves with fuel costs, which means a tenant signing today cannot know what power will cost in, say, 2035. Under a fixed-price arrangement, JERA absorbs the fuel cost risk and charges a flat rate, giving the tenant predictability but likely at a premium. At present, the pricing structure for Chiba has not been disclosed. Before signing, a prospective tenant should ask for the pricing formula, request historical LNG price data to model downside scenarios, and confirm what happens if JERA’s costs exceed the agreed price.
What happens when the plant goes down? The more serious risk is not a routine maintenance window but a site-wide disruption—an LNG supply interruption or a fault affecting multiple generating units. None of the Chiba announcements describe the backup arrangement, and the JERA press release does not address what happens when the plant goes offline. A well-designed site keeps primary supply and backup supply distinct: either a grid connection kept on standby, or on-site battery storage sized to cover an interruption. Before committing, a prospective tenant should require written disclosure of the backup arrangement: what it covers, how long it can sustain operations, and who bears the cost when it is used.
How long is the lock-in, and on what terms? A 15-to-25-year power contract is longer than most enterprise technology cycles. The workloads running in a data center in, say, 2040 may look nothing like the workloads that justified the original build. Exiting or renegotiating a power agreement of this duration with a counterparty that controls the physical infrastructure on which the facility depends is not straightforward. Before signing, a tenant should ask for the exit provisions: under what conditions the agreement can be terminated, what penalties apply, and whether the power agreement transfers or terminates if the facility changes hands.
When This Makes Sense, and When the Grid Is Still Better
The core question for any tenant evaluating a behind-the-meter site is whether the facility can guarantee power continuity when the plant is interrupted, and whether the workload can tolerate an interruption if it cannot.
A site with confirmed backup supply—a standby grid connection, on-site battery storage, or both—can support workloads that need to stay on continuously. A site without a disclosed backup arrangement cannot reliably support any workload that cannot afford to go dark. The Chiba project has not yet publicly confirmed what its backup arrangement is. That is the question to resolve before any tenant commits.
For data center developers serving customers whose workloads run in scheduled batches and who cannot wait years for a grid connection, a behind-the-meter arrangement is worth evaluating seriously, provided the backup question has a satisfactory answer. For those running systems that need continuous availability without confirmed redundancy, or who have made public commitments to use clean energy that gas-fired power cannot meet, the grid queue is still the better option, even with the wait.
One more regulatory development is worth watching before making a final site decision. Japan’s Agency for Natural Resources and Energy has outlined a proposal, targeted for April 2027, that would impose stricter rules on large facilities drawing 30 MW or more from the grid. Under the proposal, a developer who delays its planned supply start date by more than a year would face a cost settlement, with the method still under consideration, and one that scales back its contracted capacity partway through a phased rollout could lose the unused portion of its grid slot. This is not yet confirmed policy, and the final threshold may change before it takes effect. For any operator currently holding a grid connection slot while weighing alternatives like Chiba, this proposal is the one to watch in early 2027.
What Chiba offers is concrete: power sooner, without the queue. But a 15-to-25-year power contract signed today will still be running when AI workloads, energy markets, and Japan’s grid have all changed significantly. The pricing structure, backup arrangements, and exit terms agreed at signing will determine how much room a tenant has to adapt to those changes, or whether the speed advantage at the start becomes a constraint that outlasts its usefulness.
More from Asia Tech Lens
India’s AI Data Center Boom Is Running Ahead of the Grid
India’s hyperscalers are committing billions to AI infrastructure, but electricity transmission constraints threaten project timelines.
Malaysia’s AI Boom Puts Clean-Power Buyers To The Test
Malaysia is opening new routes to renewable electricity procurement, but pricing uncertainties remain. Industrial buyers face a trade-off between securing supply early and committing before the full costs become clear.
Can Nuclear Power Fuel Southeast Asia’s AI Boom?
Nuclear energy offers the promise of continuous, low-carbon electricity for AI data centers. But regulatory readiness, financing and lengthy development timelines could keep projects from delivering power when it is needed.
Hot Chips, Cool Solutions: Asia’s Race to Reinvent Data Center Cooling
As AI workloads drive up electricity demand, Asian data centers are experimenting with liquid cooling and alternative designs. The challenge is achieving reliable energy savings without introducing new costs and operational risks.
Is Southeast Asia Ready for an AI Data Center Surge?
Malaysia and Indonesia are attracting significant AI data center investment, but grid limitations, water shortages and rising costs threaten expansion.