SoftBank's $3.7 Billion Gas Turbine Loss and the Invisible Ceiling of AI Infrastructure

Masayoshi Son's boldest bet on the energy infrastructure powering artificial intelligence has ended in a US$ 3.7 billion write-down. SoftBank cancelled billions of dollars in gas turbine orders after over-reserving capacity, exposing an uncomfortable truth: even with trillionaires willing to bet fortunes, the energy market for data centers has limits nobody could accurately measure.

Son's gamble

Between 2024 and 2025, SoftBank through its subsidiary SB Energy reserved more than 13 gigawatts (GW) of gas turbine capacity with manufacturers including GE Vernova. The goal was to secure power supply for its US data centers — particularly a 9.2 GW Ohio project linked to SoftBank's US$ 64.6 billion investment in OpenAI.

The problem was straightforward: turbine delivery waitlists stretched into the early 2030s. To secure equipment, SoftBank needed to book far ahead of actual need. Peter Gardett, CEO of Noreva, told TechCrunch that "SB Energy had no choice but to reserve more turbines than it needed if it wanted to get equipment immediately". The company ended up reserving almost 50% more than required.

What started as defensive supply-chain strategy morphed into a market bet: Masayoshi Son saw the possibility that if the AI industry needed energy for data centers, it would have to come through him. SoftBank would deploy its massive capital — US$ 30 billion invested in OpenAI in February 2025, planning to spend US$ 33 billion on a 9.2 GW gas plant — to capture a significant share of the data center energy market.

What happened

In November 2026, SoftBank announced a quarterly loss of ¥1.12 trillion (US$ 7 billion), with more than half — US$ 3.7 billion — tied to gas turbines. The write-down consists of US$ 2.85 billion in non-refundable advance payments and US$ 0.88 billion in contract modifications and terminations. Over 18 months, subsidiary Silver Bands 7 LLC had paid ¥1.50 trillion (US$ 9.48 billion) in advances for equipment.

SoftBank denied any default after Elon Musk publicly stated on November 6 that "SoftBank is defaulting on their turbine orders". The company says the Ohio project continues and is conducting a "thorough reassessment," according to CFO Yoshimitsu Goto.

The impact was immediate: shares of GE Vernova and Bloom Energy fell following the announcement, and the data center gas turbine market entered a period of uncertainty about real demand.

What the loss reveals

SoftBank's US$ 3.7 billion write-down is a case study in the hidden risks of the AI infrastructure race. There are three main takeaways.

First: energy is the invisible bottleneck of AI. All eyes are on the chips — NVIDIA, TSMC, the TPU 8x — but the power needed to run those chips is an even more critical bottleneck. Gas turbines take years to deliver. Grid connections do too. SoftBank discovered firsthand that securing energy is as hard as securing computing hardware.

Second: the AI energy market has a price ceiling. SoftBank found that by reserving 50% more than needed, part of that reservation became unnecessary when project specifications were revised. This suggests other developers may be making the same calculation: book more than needed to secure equipment, then cancel the excess later. If real demand for gas turbines turns out to be lower than SoftBank expected, the market could contract more sharply than analysts predict.

Third: the alternative to dedicated energy may be arriving sooner than expected. Brent Nelson, senior managing director at Ascend Analytics, noted on the Equity podcast that for most data centers, connecting to the power grid — even at lower capacity — paired with large battery systems to cover peak demand, may prove more efficient than building dedicated gas plants. Cheaper solar and wind power, combined with storage, may offer a viable alternative that gas plant promoters overestimated.

What comes next

SoftBank remains one of the largest AI infrastructure investors in the world. Its US$ 64.6 billion stake in OpenAI represents 13% of the company. The 9.2 GW Ohio project continues. But the US$ 3.7 billion loss serves as a warning: even trillionaires can underestimate the complexity of building the infrastructure that powers the AI revolution.

The question is whether SoftBank will be the last to discover this ceiling, or whether larger players in the industry are still about to find the same limit.

Sources: TechCrunch, CNBC, Bloomberg

✓ Independent sources cross-checked and verified before publishing