SoftBank began placing turbine orders about two years ago, as AI’s surging electricity needs collided with long waits for grid connections. The company reportedly secured more than 13 gigawatts of turbines—nearly 50% more than it needed—betting that an enormous order would guarantee equipment for its data-center projects.¹

That urgency reflected a wider scramble: turbine queues now stretch into the early 2030s, and natural gas looked like a familiar way for data-center developers to secure power while waiting for the grid.¹ Peter Gardett, CEO of Noreva, said SB Energy “did not have a choice except to overbook turbines in the future if they wanted some turbines right away.”¹

But the bet became costly. On Tuesday, SoftBank reported a $7.1 billion quarterly loss, with more than half tied to power-generating equipment. The company had paid nonrefundable deposits to reserve turbines and now considers much of that money lost. Gardett described the resulting charge as “that price discovery finally happening.”¹

The reversal does not mean AI’s appetite for electricity has disappeared. Rather, operators may be learning that gas is not a risk-free answer: supplies can tighten during winter peaks, while a smaller grid connection paired with batteries may offer a more practical balance. Brent Nelson of Ascend Analytics said the grid “might still be their best bet.”¹

SoftBank’s losses are not expected to trigger a wave of gas-plant cancellations, given the scale of projected AI power demand. But the company’s oversized order points to a shift: securing energy quickly mattered first; working out how much gas generation is actually needed comes next.¹