SoftBank has booked a $7.1 billion quarterly loss, and more than half of it stems from a failed attempt to corner the market for the gas turbines that power AI data centers. Why it matters: it is the clearest sign yet that the scramble for AI electricity has a ceiling — even a buyer with SoftBank’s war chest can overpay for power capacity that the market does not need as fast as expected.
November 13, 2026
What happened
Starting about two years ago, SoftBank reportedly put down deposits on more than 13 gigawatts of natural gas turbines — almost 50% more than it needed. The logic was scarcity: AI’s rising power demand caught the energy world off guard, turbine order queues now stretch into the early 2030s, and chief executive Masayoshi Son bet that by controlling a slice of that supply, SoftBank could make itself a gatekeeper for data-center power. The deposits were nonrefundable.
On Tuesday, SoftBank announced a $7.1 billion quarterly loss, with over half of it stemming from losses related to power-generating equipment. The company now considers much of the deposit money worthless; the misplaced bet totals over $3 billion. Reporting by Heatmap put the deposit loss at about $3.7 billion and said roughly 4 gigawatts of SoftBank’s reservations are expected to return to manufacturers, with about 9 gigawatts already earmarked for an OpenAI data-center project in Pike County, Ohio.
Why a giant over-ordered on purpose
The over-ordering was not simply a mistake — it was the price of speed. “SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away,” Peter Gardett, chief executive of energy firm Noreva, told TechCrunch. Rather than overpay and still risk being short, SoftBank overpaid and over-ordered, and the charge now landing in its results is, in Gardett’s words, the cost of that price discovery finally happening.
The bet sat alongside SoftBank’s other giant AI wagers: a $30 billion investment in OpenAI in early 2025, and a February announcement that its SB Energy subsidiary would spend $33 billion on a 9.2-gigawatt natural gas power plant. What changed is the demand picture. As data centers spend more time in operation, operators have a better feel for what they actually need — and analysts note natural gas carries its own risks, including winter curtailments when supplies dip. A smaller grid connection paired with large batteries may serve most data centers better than a giant dedicated gas plant.
Why it matters
SoftBank’s loss is unlikely to trigger mass cancellations of gas power plants — AI’s power demand is still enormous, and natural gas remains a go-to source for developers who cannot wait years for a grid connection. But the write-down marks a limit: the first phase of the AI power rush rewarded whoever grabbed capacity fastest, and the second phase is rewarding whoever sized it correctly. For readers watching the AI buildout, the practical takeaway is that the power bottleneck is real but not infinite — and the companies financing it are starting to find out exactly where it ends, one nonrefundable deposit at a time.
Frequently asked questions
How much did SoftBank lose on turbines?
SoftBank announced a $7.1 billion quarterly loss, with over half stemming from power-generating equipment. The turbine-deposit bet itself totals over $3 billion; Heatmap reported a loss of about $3.7 billion on the deposits.
Why did SoftBank order more turbines than it needed?
Turbine queues stretch into the early 2030s, so over-ordering was the fastest way to secure equipment and a position in the AI power market. About 13 gigawatts were reserved — nearly 50% more than SoftBank required.
Does this mean AI data centers are cancelling gas plants?
No. Analysts expect AI power demand to remain large enough that widespread cancellations are unlikely. The loss mainly caps SoftBank’s own natural gas ambitions.
Sources: TechCrunch, Wall Street Journal reporting via TechCrunch, Heatmap, Bloomberg