But wait... there's more! In January, SoftBank's SB Energy, the developer building the massive Ohio and Texas data center campuses, granted warrants to OpenAI as a "lease incentive" for signing a data center lease in Milam County, Texas. That's right: the landlord paid the tenant, in equity, to rent the building. Some of those warrants were forfeited in August when OpenAI signed the 20-year Ohio lease, but ¥542 billion ($3.4 billion) of them still sit on SoftBank's balance sheet as a derivative liability. And here is the punchline: because the warrants get more valuable as SB Energy does, every time the data center business is marked higher, SoftBank books a loss, ¥204 billion in the first half, plus another ¥136 billion of share-based compensation for the unit's staff for the same reason. A business whose success shows up as a loss is a neat trick even by AI-bubble standards. But the circularity doesn't end there: Nvidia, the exclusive chip supplier for the Ohio campus, has wired $1.5 billion to SB Energy under a share issuance agreement and is providing credit support for OpenAI on the first 4.25 gigawatts. So the chip vendor (NVDA) funds the landlord (Softbank) and backstops the tenant (OpenAI), who then buys the chips (NVDA). SoftBank Corp, the listed Japanese telecom arm, sold its SB Energy stake to Vision Fund 2 and booked an ¥80 billion gain. One SoftBank pocket selling to another (for a "profit"): the gain vanishes on consolidation, but it looks great at the telco, and a whole bunch of clueless Japanese momentum-chasers buy something that moves higher. The Vision Funds, meanwhile, marked down their stakes in SoftBank's own subsidiaries, mainly SB7 (the turbine buyer) and PayPay, by ¥364 billion. That one vanishes on consolidation too. And SB Energy's IPO, which was supposed to crystallize a $50 billion-plus valuation, has been frozen since September due to regulatory issues well before the midterms.