A non-binding MoU between Japan’s largest power generator, Dell Technologies, and RHAELM Holdings targets a behind-the-meter AI data centre at Chiba Thermal Power Station — and the mechanism is more significant than the headline number.
This is a NON-BINDING Memorandum of Understanding. Nothing below reports the project as agreed, funded or certain to proceed.Every headline number is hedged in the release itself: capital expected to EXCEED $15 billion, power UP TO 400 MW, operations TARGETED around 2028. The per-megawatt figure is this publication’s division of a floor by a ceiling. Nothing here is investment advice.
On October 1, 2026, JERA — Japan’s largest power generator — announced a non-binding Memorandum of Understanding with Dell Technologies and RHAELM Holdings Ltd. The agreement targets an AI data centre at JERA’s Chiba Thermal Power Station. Total capital deployment is expected to exceed US $15 billion, or 2.3 trillion yen.
The site would draw up to 400 MW of power behind the meter — meaning directly from JERA’s operating generation asset, before electricity reaches the public grid. Operations are targeted to begin around 2028. The release states this structure enables capacity to be delivered years ahead of a conventional grid-connected development schedule.
The division of labour is stated explicitly. JERA provides 24/7 power. Dell Technologies standardises the compute layer through what the release calls the Dell AI Factory. RHAELM Holdings Ltd., described in the release as a sovereign AI infrastructure developer, leads project delivery and integrates the capabilities.
The key insight: The behind-the-meter structure does not shorten the grid interconnection queue — it removes Chiba from that queue entirely. That is a different category of solution, not a faster version of the same one.
One input is a floor and the other is a ceiling, so the quotient is an order of magnitude rather than a unit price. No comparison to other projects is offered, because none is sourced here.
The Structural Read
Grid connection has become the pacing item for large AI sites. No amount of capital shortens a queue you are standing in. Every developer waiting for a conventional interconnection approval is competing for the same scarce resource: a slot on the transmission network.
Siting behind the meter changes which queue you are in. It also changes who the gatekeeper is. The utility interconnection process stops being the bottleneck. The power company becomes the counterparty instead.
That structural shift is why the energy partner is named first in this release. JERA is not supplying power to a data centre developer. JERA is hosting one — on its own operating asset, under its own control. The generator becomes the landlord, not the utility bill.
Three layers with three named owners is unusually legible for an announcement at this stage. It signals that the parties have defined their roles, even if the capital, phasing, and approvals are not yet settled.
Yukio Kani — JERA Global CEO and Chair
“JERA is uniquely positioned to power Japan’s AI ambitions.”
What the Release Does Not Settle
Every load-bearing number in this announcement is hedged. Capital is expected to exceed $15 billion rather than equal it. Power is up to 400 MW. Operations are targeted to begin around 2028. Four conditionals carrying one headline figure.
That is not a criticism. An MoU between a national power generator, a listed hardware vendor, and a named developer is a real step. The parties are real. But it sets a firm limit on what a reader can conclude.
Nothing in this announcement establishes that the project is funded, approved, or certain to proceed. For scale only: dividing the expected floor of $15 billion by the ceiling of 400 MW produces roughly $37.5 million per megawatt. One input is a floor and the other a ceiling, so treat that figure as an order of magnitude, not a unit price.
Not established and therefore absent from this analysis: funding sources and split, committed financing, chip vendor, customers, phasing beyond the 2028 target, land status and permits, and whether a binding agreement is expected to follow.
Three Implications
Power Companies Become Infrastructure Landlords
When grid interconnection is the binding constraint, operators of running generation assets hold a structural advantage. Chiba proposes a model where the generator hosts rather than supplies. Whether other power companies follow is not something this publication is forecasting, and the MoU establishes nothing beyond this one site.
Dell AI Factory Becomes a Repeatable Module
The release frames Chiba as a site that can generate practical experience for a scalable model and broader deployment. That framing positions Dell AI Factory as a standardised compute layer that drops into behind-the-meter sites. The compute stack becomes portable; the location constraint does the differentiation.
Japan’s AI Infrastructure Race Has a New Pacing Variable
JERA’s stated ambition is to help shape Japan’s AI infrastructure. Behind-the-meter siting, if it progresses, could compress the timeline for large-scale AI capacity in Japan relative to conventional grid-connected routes. The relevant variable is no longer just capital availability — it is access to operating generation assets.
Business Engineer Framework
Harness Theory — Who Controls the Scarce Layer Wins
Harness Theory maps where structural advantage accumulates when a new capability — here, delivered AI compute — depends on a resource others cannot easily replicate. At Chiba, that resource is a running power asset with no queue in front of it. The Map of AI shows how this layer dynamic plays out across the full AI stack.
A non-binding MoU is not a built data centre — and this one carries four explicit hedges around its biggest numbers. What it does establish is a structural logic: when the grid queue is the constraint, the answer is not a faster application but a different location. JERA sitting behind its own meter at Chiba is a real demonstration of that logic, with real parties attached. Everything that follows — funding, permits, phasing, customers — is still to be determined.
Every detail above comes from JERA’s own news release of 1 October 2026, read directly. Nothing has been independently verified and none of the three parties has been contacted.The agreement is a non-binding Memorandum of Understanding. Nothing above should be read as reporting that the project is funded, approved, under construction or certain to proceed, and the release does not say whether a binding agreement is expected to follow.
The headline figures carry the release’s own hedges and those hedges matter. Capital deployment is expected to EXCEED US $15 billion, which is a floor rather than a total. Power capacity is stated as UP TO 400 MW, which is a ceiling. Operations are TARGETED to begin AROUND 2028.The figure of roughly $37.5 million per megawatt is this publication’s division of the capital floor by the capacity ceiling.
Because of that, it is an order of magnitude rather than a unit price, and no comparison with any other project is offered because none is sourced here.The release does not address the emissions profile of powering AI compute from a thermal generation asset, and nothing above characterises it in either direction. RHAELM Holdings is described only as the release describes it, a sovereign AI infrastructure developer.
Also absent: who funds the capital and in what proportions, whether financing is committed, the chip vendor, customers, phasing, land status and permits. Dell Technologies is a listed company. Nothing above predicts anything, and nothing here is investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
