SoftBank Books $7.1 Billion Loss as Gas Turbine Bet Sours

SoftBank posted a $7.1 billion quarterly loss on Tuesday, and more than half of it — over $3 billion — came from writing down deposits on natural gas turbines it no longer considers worth what it paid.

The write-down caps a bold procurement play. Starting roughly two years ago, SoftBank put down nonrefundable deposits on more than 13 gigawatts of turbines, nearly 50% more than it needed, according to reports. At the time, turbine order queues were already stretching into the early 2030s, and the deposits locked in supply and accelerated project timelines.

“SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away,” Peter Gardett, CEO of Noreva, told TechCrunch via email.

Why did SoftBank over-order in the first place?

The strategy fits SoftBank's house style. The firm staked $30 billion on OpenAI in early 2025, and in February it announced that subsidiary SB Energy would spend $33 billion building a 9.2-gigawatt natural gas power plant. If you can't be first, be biggest.

The energy world was caught flat-footed by AI's power demands. Queues for turbines now extend into the early 2030s, and natural gas had shaped up as the default answer for data center developers facing years-long grid interconnection delays. Masayoshi Son's wager was straightforward: deploy SoftBank's war chest to corner a slice of that market, and if the AI world wanted power for its data centers, it would have to come to him.

That bet fizzled. On Tuesday, SoftBank declared much of its nonrefundable turbine deposits worthless, attributing over half of the $7.1 billion quarterly loss to power-generating equipment.

Gardett framed the charge as an expensive lesson in market pricing. Rather than overpay and potentially still fall short of turbines, SoftBank overpaid and over-ordered.

“This charge is just that price discovery finally happening,” Gardett said.

Is natural gas still the answer for AI data centers?

The magnitude of the misplaced bet reflects the electricity mania that has overtaken the AI world. Faced with years-long waits for grid connections, many hyperscalers and developers reached for natural gas as an energy security blanket — proven, familiar, dispatchable.

But operators are learning the risks as AI data centers spend more time in real operation. Winter curtailments, when supplies dip and demand spikes, remain a genuine hazard, Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, told the Equity podcast on Wednesday.

Nelson's counterintuitive take: the grid might still be the best bet. Most data centers would be better served by a grid connection — just a smaller one than they'd normally request. Large batteries could cover the shortfall cost-effectively, shaving peak loads and riding out utility requests to curtail electricity use.

Data center operators may be reaching the same conclusion. That shift helps explain why the natural gas market is developing more slowly than SoftBank anticipated when it locked in its supersized turbine orders.

Will the loss trigger cancellations across the sector?

Probably not. SoftBank's gas-related losses are unlikely to cascade into mass cancellations of gas power plants, because AI's power demand is simply too large for the industry to walk away from the fuel. For now, natural gas is riding high.

SoftBank, though, may have found its ceiling — and paid more than $3 billion to learn where it sits.