SoftBank just posted a $7.1 billion quarterly loss, and more than half of it came from one thing: gas turbines it ordered and no longer needs. TechCrunch AI reports that the company put nonrefundable deposits on more than 13 gigawatts of turbines over roughly the past two years. That’s almost 50% more than its projects required. It announced the loss on Tuesday, and it’s the clearest sign yet that the rush to power AI with natural gas may have a ceiling.
⚡ The Short Version
- What happened: SoftBank wrote off much of the deposits it paid turbine makers. The power-equipment charge came to more than $3 billion.
- Why it happened: It over-ordered on purpose, betting it could corner part of the gas-power market that AI data centers rely on.
- Why it matters: Data center operators are learning what they actually need, and gas demand is growing more slowly than SoftBank expected.
- What it doesn’t mean: No one expects a wave of gas plant cancellations. AI’s power appetite is still too large for that.
🎯 How SoftBank Got Here
SoftBank tends to show up late to a trend and then go bigger than everyone else. It put $30 billion into OpenAI in early 2025. In February, its SB Energy subsidiary announced a $33 billion, 9.2-gigawatt natural gas power plant.
That same supersized approach showed up in its turbine orders. The energy industry didn’t see AI’s power demand coming this fast, and turbine order queues now stretch into the early 2030s. Masayoshi Son’s plan was simple. Use SoftBank’s huge war chest to lock up turbine supply, so anyone building AI data centers would have to come to him for power.
In fairness, over-ordering wasn’t crazy. “SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away,” Peter Gardett, CEO of Noreva, told TechCrunch. Over-ordering got SoftBank its turbines and sped up its projects.
💸 An Expensive Way to Learn the Price
Gardett calls the write-off a costly round of price discovery. SoftBank could have paid too much and still ended up short on turbines. It chose to pay too much and end up long. “This charge is just that price discovery finally happening,” he said.
What stands out to me is the size of the miss. A $3 billion-plus write-off on equipment deposits alone shows how much panic has driven AI power buying. Facing years-long waits for grid connections, hyperscalers and developers grabbed natural gas as a security blanket because it’s proven and familiar.
🔋 The Grid Might Win After All
Now that AI data centers have been running for a while, operators have a better idea of what keeping them online really takes. And gas isn’t as safe as it looks.
Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, said on TechCrunch’s Equity podcast that gas comes with real risks. One example is winter curtailment, when supplies drop just as demand spikes. His suggested setup for most data centers:
- A grid connection, but smaller than operators usually ask for
- Large batteries to make up the gap, which he called a cost-effective choice
- Peak shaving and curtailment response, with batteries covering demand spikes and utility requests to cut usage
If more operators reach that same conclusion, it explains why the gas market is developing more slowly than SoftBank bet it would.
🔭 What to Watch
For practitioners and anyone planning AI infrastructure, a few takeaways:
- Turbine scarcity may be easing at the margins. SoftBank’s write-off suggests some of those queue slots weren’t real demand.
- Hybrid power is gaining ground. Grid plus batteries looks more and more like the sensible default, not the compromise.
- Gas isn’t going away. It’s still a major source of AI power. The fantasy of unlimited demand is what’s fading.
My take: SoftBank didn’t misread whether AI needs power. It misread what kind of power and how much of it. Natural gas is still doing well for now, but the market just found out where its limits might be, and it took a $3 billion write-off to show it. TechCrunch AI has the full report.