JERA puts a $15 billion-plus AI campus behind the meter at its Chiba gas plant

Spread over 400 megawatts and every phase of the build, the $15 billion-plus budget comes to at least $37.5 million per megawatt ^[3][4][1]. RHAELM will develop, build, operate and finance the data center, and Dell supplies the AI hardware ^[6]. JERA contributes the land and the power ^[7]. So a generator that produces roughly a third of Japan's electricity ^[8] is acting as landlord and supplier. The halls, and the capital tied up in them, sit with RHAELM and its backer ^[6][2].

JERA's prize is the contract. In Cryptopolitan's account, the 15-to-25-year supply deal fixes the largest single running cost for the operator and hands JERA a guaranteed buyer for years ^[5][9]. If the site ran at the full 400 megawatts for all 8,760 hours of a year, it would draw about 3.5 terawatt-hours ^[2], all of it sold to one customer next to the plant. Yukio Kani, JERA's global CEO and chair, pointed to the company's reach across the LNG value chain ^[18]. "JERA is uniquely positioned to power Japan's AI ambitions," he said ^[14].

Location follows from the grid. Cryptopolitan says the site choice comes down to electricity and how long it takes to get ^[12]. AWS, Microsoft and Oracle have already spread capacity across several regions to find power the Tokyo grid cannot supply ^[11]. A campus wired directly to an existing station skips the connection queue entirely ^[10].

The financing is less settled than the headline suggests. Apollo intends to act as RHAELM's strategic investment and financing partner on the first site, and the whole arrangement is a memorandum of understanding ^[2][1]. The source does not say how Apollo's role splits between debt and equity, or how much of the $15 billion Apollo would put in. Blackstone President Jonathan Gray said in June that his firm plans to invest $30 billion in Japanese AI data centers over three to five years ^[13]. One Chiba campus costs at least half that sum ^[3].

Replication raises the bill. If the multi-gigawatt ambition ^[15] means as little as 2 gigawatts, Chiba's cost per megawatt puts the program at $75 billion or more ^[4]. That is two and a half times Blackstone's entire Japan plan ^[5].

I think the evidence supports the location argument, and the claim that private credit is paying for this has not been shown yet. The location case would fail if a hyperscaler got several hundred megawatts of new Tokyo grid capacity within two or three years. The private-credit case would fail if Apollo's package turns out to be mostly equity, or if the debt rests on a JERA guarantee.