SoftBank’s $7.1 Billion AI Power Bet Hits a Ceiling

By Daniel Okafor

SoftBank announced a $7.1 billion quarterly loss on Tuesday, with more than half tied to losses on power-generating equipment.

More than $3 billion of that charge came from a turbine procurement strategy meant to reserve scarce equipment for AI data centers. SoftBank had paid turbine manufacturers nonrefundable deposits, then came to view much of that deposited value as worthless.

Why did SoftBank order more turbines than it needed?

AI’s electricity demand grew faster than the supply chain could deliver. TechCrunch reported that queues for turbine orders now extend into the early 2030s. Natural gas was becoming a go-to power source for AI data centers, while Masayoshi Son tried to use SoftBank’s capital to corner a large part of the market.

The strategy assumed AI developers would have to source power through SoftBank. Its spending commitments reflected that approach:

• In early 2025, SoftBank made a $30 billion bet on OpenAI.
• In February, subsidiary SB Energy said it would spend $33 billion on a 9.2-gigawatt natural gas power plant.
• Starting about two years ago, SoftBank placed deposits for more than 13 gigawatts of turbines, almost 50% more than it needed.

Noreva CEO Peter Gardett told TechCrunch by email that the surplus gave SoftBank an advantage in a constrained market. “SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away,” he said.

The mechanism was straightforward. SoftBank used nonrefundable deposits to reserve delivery slots and accelerate its projects. The excess capacity could prove valuable if turbines remained scarce; it became costly if demand weakened or equipment values fell. TechCrunch said the overbooking helped SoftBank secure turbines and speed its projects, but much of the deposit value is now considered worthless.

What did the $7.1 billion loss reveal?

SoftBank’s equipment losses exceeded $3 billion. Gardett said SoftBank chose to overpay and overorder rather than risk paying more later and still coming up short. The scale of the error also reflected the electricity buying frenzy surrounding AI, according to TechCrunch.

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

The charge exposed the cost of buying capacity far ahead of actual need. A company with a large balance sheet can reserve years of turbine supply, but it also absorbs the loss when its expected pipeline cannot support that reservation. SoftBank’s procurement premium became a balance-sheet charge.

That makes the loss larger than a single turbine order. It put a price on SoftBank’s ambition to become a major supplier for AI power and showed how a shortage can reward large buyers while exposing them to overcapacity.

Can AI data centers still rely on natural gas?

The broader demand case remains intact. Years-long waits for grid connections pushed many hyperscalers and developers toward natural gas as an energy-security option. Gas offered a proven and familiar source while data centers waited for grid access.

Longer operating experience still complicates that choice. Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, told the Equity podcast Wednesday that natural gas carries risks, including winter curtailments when supply falls and demand rises. As AI data centers spend more time operating, their operators are learning what they need to keep those facilities running.

Nelson said the grid might still be the best option for most data centers, even if they request less capacity than usual. Large batteries can cover the shortfall at a reasonable cost. They can shave peak loads or carry operations through utility requests to curtail electricity use.

This points to a slower gas market than SoftBank expected, rather than a collapse in AI power demand. Operators may choose smaller grid connections and add battery capacity instead of ordering enough turbines to support SoftBank’s scale.

How far can the losses spread?

TechCrunch assessed that SoftBank’s natural gas-related losses are unlikely to trigger mass cancellations of gas plants. AI’s power demand appears too large for one company’s procurement mistake to reverse. For now, natural gas remains strong, while SoftBank may have found its ceiling.

The sharper implication sits with equipment suppliers and project developers. Turbine queues already reach into the early 2030s, but SoftBank’s $3 billion-plus write-down shows that scarcity does not remove the risk of overpaying for capacity customers may not need.

The next test is whether other data-center developers follow SoftBank toward smaller grid connections and batteries. That would slow natural gas growth without changing the scale of AI’s electricity demand.