AivexaNewsSearch
AI news for builders and product teamsChecked every hour

SoftBank’s billion-dollar loss may have put a cap on AI’s natural gas ambitions

Collected Nov 12, 2026

SoftBank reported a $7.1 billion quarterly loss on Tuesday, and more than half of that figure came from write-downs on power-generating equipment. The Japanese investment group had placed nonrefundable deposits with turbine manufacturers to lock in orders, and it now considers much of that commitment worthless. The gas-related damage alone exceeds $3 billion, per the company's disclosure as reported.

The backstory is one of scale. SoftBank's SB Energy subsidiary said in February it would spend $33 billion on a 9.2-gigawatt natural gas power plant. Roughly two years ago, SoftBank reportedly began putting down deposits on more than 13 gigawatts of turbines — close to 50% more capacity than its projects required. The intent was straightforward: turbine queues now stretch into the early 2030s, so over-ordering was the surest way to get hardware sooner rather than later. Peter Gardett, CEO of Noreva, put it to TechCrunch that SB Energy had no real choice but to overbook turbines if it wanted any delivered promptly.

The broader context is the power crunch created by AI. Grid interconnection queues in many markets take years, and hyperscalers and data center developers have increasingly looked to on-site or dedicated natural gas generation as a dependable, familiar alternative to waiting. CEO Masayoshi Son's wager, in effect, was that SoftBank's balance sheet could corner a slice of that turbine supply and make the AI buildout come to him. The write-down suggests the demand behind that thesis did not materialize on his timeline.

Gardett characterized the charge as price discovery — SoftBank overpaid and over-ordered rather than risk coming up short on turbines, and the loss is the market settling that bill.

The buyer side may also be rethinking gas. Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, told the Equity podcast that natural gas carries its own risks, including winter curtailments when supply tightens and demand spikes. His view is that most data centers are still best served by a grid connection, even a smaller one than they might normally request, with large batteries filling the gap — batteries can shave peak loads or absorb utility curtailment requests at lower cost than self-generating.

If operators are converging on that conclusion, the dedicated gas market would develop more slowly than SoftBank's order book assumed. That is a plausible reading of the write-down: the deposits were a bet on a procurement race that cooled before the turbines arrived.

The loss is unlikely to trigger a wave of cancellations across gas plant projects. AI's electricity demand is simply too large for that, and gas remains in high demand overall. The more likely takeaway is that SoftBank found the ceiling of what one investor can profitably pre-book in this market.

For developers and product teams, the practical lesson is about infrastructure lead times. Power availability now shapes where and how fast compute can be stood up, and the hedging strategies around that constraint carry real financial risk, as a multi-billion-dollar charge demonstrates. Teams planning large training or inference footprints may find that grid capacity plus storage is a more flexible and cheaper path than contracting dedicated generation, an inference from Nelson's remarks rather than a stated conclusion. The turbine queues extending into the 2030s also mean gas-backed capacity cannot be treated as a near-term lever; procurement timelines should be modeled in years, not quarters. Finally, the episode is a reminder that supply-chain bets in AI infrastructure are being repriced as operators learn their actual load profiles — the early assumptions about how much power AI would need, and how urgently, are still being tested.

Why it matters: Data center operators and the vendors serving them are watching whether gas remains the default answer to AI's power gap or whether grid connections and batteries take share, since that choice drives siting, cost and timelines for new compute. SoftBank's write-down shows the financial exposure of locking in turbine capacity ahead of demand, which may make investors and operators more conservative about pre-ordering equipment. For developers, power constraints remain a first-order planning variable rather than a procurement detail.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

SoftBank lost over $3 billion preordering natural gas turbines to power AI data centers. Here's what the loss means to power markets.