SoftBank’s AI Power Bet Burns Cash

Its attempt to get ahead of the AI power scramble has turned into an expensive lesson in what happens when urgency meets nonrefundable deposits. TechCrunch reports that SoftBank booked a $7.1 billion quarterly loss, with over half tied to losses related to power-generating equipment. The hole appears to sit largely in an effort to secure gas turbines for AI-linked electricity projects: the company had reportedly put down deposits on more than 13 gigawatts of turbines, roughly 50% more than it needed for a 9.2-gigawatt natural-gas plant plan announced by SB Energy in February.

That is not just a bad quarter. It is a visible cost of one of the louder assumptions around AI infrastructure: that electricity demand would be so immediate, and grid access so slow, that over-ordering gas equipment was the sensible way to buy time.

In SoftBank’s case, buying time seems to have meant buying too much of it. Peter Gardett, chief executive of Noreva, told TechCrunch that developers wanting turbines quickly had little choice but to overbook future supply. In a market where turbine queues reportedly stretch into the early 2030s, the logic is easy enough to follow. Scarcity can make bad procurement look strategic right up until the write-down arrives.

The reported hit of more than $3 billion on turbine-related deposits also says something broader about the AI power buildout. Faced with long waits to connect to the grid, hyperscalers and developers have treated natural gas as an energy-security blanket for data centers: familiar, dispatchable and, at least in theory, a way around the queue. But “familiar” is not the same thing as frictionless. As Brent Nelson of Ascend Analytics told TechCrunch’s Equity podcast, natural gas still carries risks, including winter curtailments when supplies dip and demand spikes.

TechCrunch also cites Nelson arguing that many data centers may be better served by smaller grid connections backed by large batteries rather than oversized gas commitments. If operators are reaching the same conclusion, SoftBank’s loss looks less like an isolated accounting bruise and more like a costly bit of market research.

Natural gas is still central to plenty of AI-era power plans. SoftBank just paid billions to find out that cornering the turbine market is not the same as cornering demand.