SoftBank’s loss exposes the limits of betting on gas for AI

SoftBank’s loss shows the limits of betting on gas power for AI

Betting on extra turbines

SoftBank is not always the first to invest in popular sectors, but it often makes large bets. In early 2025, the company invested $30 billion in OpenAI. In February, its subsidiary SB Energy announced plans to spend $33 billion on a 9.2 GW gas-fired power plant.

About two years ago, SoftBank began making deposits on turbines with a combined capacity of more than 13 GW — nearly 50% more than it needed. Ordering extra capacity helped reserve equipment and move projects forward faster. Noreva CEO Peter Gardett said SB Energy effectively had no other way to secure some of the turbines quickly: the company had to order them in advance and in excess.

Demand for electricity from AI grew faster and more sharply than expected. Queues for gas turbines now stretch into the early 2030s. SoftBank CEO Masayoshi Son expected the company’s large investments to give it a prominent position in the gas-power market: data center developers in need of energy would have to turn to SoftBank.

The cost of miscalculation

On Tuesday, SoftBank reported a quarterly loss of $7.1 billion. More than half of that amount came from losses tied to power-generation equipment. The company made nonrefundable deposits with turbine manufacturers to secure orders, but now considers a significant portion of that spending useless.

Gardett called the losses an expensive way to find out what the equipment really costs. SoftBank chose to overpay and order extra turbines so it would not face a shortage, even at a higher price. In Gardett’s view, the write-down reflects the point at which that price test became visible in the financial results.

Losses of more than $3 billion show how strongly the AI market has focused on securing future electricity supply. With grid connections taking years, many large cloud companies and developers have come to see gas as a familiar, proven way to obtain reliable power.

Gas, the grid and batteries

As data centers spend longer operating under normal conditions, their operators are getting a better sense of how much energy they need. Gas may seem like a reliable option, but it carries risks. In winter, for example, supplies can be constrained just as demand rises, Brent Nelson, Ascend Analytics’ senior managing director for markets and strategy, told the Equity podcast on Wednesday.

Nelson said connecting to the grid may still be the preferred option for most data centers, even if they receive less power than they typically request. Large batteries can help make up the shortfall: they can smooth peak demand and let operators comply with utilities’ requests to cut consumption.

Data center operators may be reaching similar conclusions, which could explain why the gas-power market is developing more slowly than SoftBank expected.

The limits of the bet

SoftBank’s losses are unlikely to lead to widespread cancellations of gas power plants: demand for energy from AI is still too high. Gas generation remains in demand, but SoftBank’s experience has made the limits of a strategy built on placing large turbine orders in advance more apparent.

Dataist’s take: The episode shows the difference between demand for electricity and the market’s willingness to pay for equipment far in advance. With grid connections taking years, gas remains one option, but supply constraints and the ability to combine grid power with batteries make it harder to rely on gas as a universal solution.