SB Energy is paying OpenAI $5.5bn in warrants to keep it as a tenant

Say you are budgeting capacity for next year and the most you can honestly promise is three years with an out. The reason that lands badly on the other side of the table has less to do with your volume than with what the developer needs to show a lender. Oracle's $16.3bn data-centre package required PIMCO to anchor $10bn after US banks stepped back ^[11], roughly 61% of the package from one non-bank ^[5], and SoftBank's $40bn March bridge facility is still waiting for something permanent to replace it ^[12]. In that market the scarce input is not megawatts. It is a tenant whose name survives underwriting, and SB Energy's warrants are the price of one ^{[1] [2]}.

The arithmetic is worth doing slowly. Granted in January at $3.6bn ^[3], marked at $5.5bn by the end of June ^[4], the position gained about 53% in five months on the developer's own equity rather than on anything the tenant did ^{[1] [4]}. No cash left SB Energy to make that happen, because a warrant only pays out if the issuer's stock appreciates, and the dilution shows up for whoever buys the IPO ^[13]. Set $5.5bn against the $5bn to $7bn the listing targets and the signing incentive equals 79% to 110% of the proceeds ^{[7] [2]}. Spread it across the capacity committed, ten gigawatts in Ohio and 1.2 gigawatts in Texas, and it works out near $491m a gigawatt ^{[5] [6] [3]}. Treat that as a rate of exchange rather than a contract term: the reporting places the warrants in the same period as the Texas lease and the $500m each that OpenAI and SoftBank put into SB Energy ^[6], not in a per-megawatt formula. The warrants are now worth eleven times OpenAI's own $500m cheque ^[4].

The pitch everyone else hears is that capacity is scarce and early commitment is prudent. This transaction says the developer was the party that needed to lock something in, and paid for the privilege. OpenAI ends up with the warrants, a single-digit stake after the listing ^[7], and the upside of the value its signature created ^[14]. Nvidia is weighing about $3bn into SB Energy on the same Ohio project, according to The Information ^[8], and has separately been in talks to guarantee $250bn of OpenAI data-centre debt ^[10]. SoftBank, which owns the landlord, is past $64bn of commitments across the model layer and the physical layer ^[9]. Neither OpenAI nor SB Energy has commented on the warrant terms, and the account rests on unnamed sources ^[15].

For anyone smaller, the useful grid has two axes: the term you can credibly hold through a bad year, and whether a third party will lend against your name. Long term plus underwritable makes you an anchor, and the ask is equity, not discount, because equity is what got paid here ^[14]. Long term without credit gets you a prepayment schedule. Short term with credit, where most enterprise buyers sit, means taking your value as exit rights and delivery dates rather than dollars per kilowatt. Short term without credit means buying leftovers on someone else's calendar, and a 2028 first-operations date tells you when those leftovers appear ^[5]. Of the two axes, term is the one you can argue about inside your own company, and in this market it is worth more than another round of price benchmarking.