View / The AI race is getting monstrously expensive. Can Oracle afford it?

Larry Ellison brought a knife to a gun fight.

Oracle’s announcement that it will raise $50 billion in debt and equity to finance its massive data center commitments has laid bare the brutal economics of the AI buildout: Some players can afford it, and some can’t.

Oracle has been the bellwether for AI worries for a while now. It doesn’t have the cash-cow businesses that allowed Amazon, Alphabet, Meta, and Microsoft to pour billions into their AI dreams without (mostly) anyone batting an eye. Its basic tension is that it’s behaving like a hyperscaler while carrying the balance sheet and investor expectations of an old-school software company.

Creditors haven’t exactly called its bluff, lining up in droves for Oracle’s $25 billion bond issuance yesterday. But the cost to insure its debt has skyrocketed, and some bondholders sued last month, saying the company misled them about future fundraising plans that triggered a “swift and bracing” selloff. And it has $248 billion of lease commitments that aren’t debt in the strict accounting sense, but are debt in the sense that failing to make those scheduled payments would be very, very bad.

Oracle’s overwrought response to reporting that Nvidia may renege on the $100 billion it promised to OpenAI — Oracle’s biggest customer — is unnervingly phrased. Oracle said it was confident OpenAI can raise money, not make money. That distinction captures the precarious foundation of today’s AI economy, which runs on venture-capital FOMO, not paying customers.

Every emerging technology depends on investors willing to carry it until revenue materializes. But AI’s price tag is unprecedented and for Oracle — without the deep pockets of its hyperscaler rivals — the game may already be too expensive to play.