Natural gas was supposed to be the energy security blanket for AI power-hungry data centers. Instead, it just became the biggest procurement mistake in the industry's history.
SoftBank Group announced a $7.1 billion quarterly loss on November 10 — its first loss in nearly two years — and more than half of it traces back to a single, audacious bet: that by deploying its vast war chest, the Japanese investment conglomerate could corner a significant share of the U.S. gas turbine market and become the dominant supplier of power infrastructure for AI data centers. The bet, it turns out, was too big even for SoftBank.
The numbers tell the story. Between late 2025 and early 2026, SoftBank placed reservations for more than 13 gigawatts of gas turbines, primarily from GE Vernova, according to the Wall Street Journal and the company's own financial disclosures. That amount is roughly 50 percent more than the company ultimately needed — enough to generate the equivalent power of six Hoover Dams, as one report put it. The total advance payments reached approximately $9.5 billion, and when SoftBank realized its procurement had grown unwieldy, it cancelled the excess reservations and unwound some contracts further along in the pipeline. The result: a $3.7 billion charge recorded in its second-quarter fiscal 2026 results, composed of $2.85 billion in non-refundable advances written off and $880 million in derecognized advances.
The context matters. When SoftBank began its turbine orders, the AI infrastructure boom was hitting a physical bottleneck that had nothing to do with chips or models: electricity. Grid connections were taking years to secure. Hyperscalers and data center developers needed power fast, and natural gas — with turbines that could be deployed on-site and scaled incrementally — looked like the perfect bridge. CEO Masayoshi Son envisioned SoftBank not just buying turbines for its own projects, but becoming a major supplier to the entire industry. If the AI world wanted power for its data centers, it would have to come through SoftBank.
That ambition collided with reality. As data centers started running, operators got a clearer picture of their actual power requirements. The picture showed that a mix of grid connections (even if undersized) combined with battery storage was often more cost-effective and reliable than the massive gas-fired plants SoftBank had bet on. Brent Nelson, a senior managing director at Ascend Analytics, told TechCrunch's Equity podcast that most data centers would likely end up choosing grid connections at reduced capacity plus battery backup over standalone gas plants.
The market reaction was immediate. GE Vernova shares slid more than 6 percent on November 10, the day SoftBank's results were announced. Elon Musk took to X to declare that SoftBank had defaulted on its turbine orders — a characterization that was disputed by SoftBank itself, which clarified that it had not defaulted on any contracts and was not downsizing its PORTS-Pike project in Ohio.
The PORTS-Pike project, a $33 billion, 9.2-gigawatt natural gas plant on a former uranium enrichment site in Pike County, Ohio, remains on track. SoftBank's CFO, Yoshimitsu Goto, confirmed during the earnings call that the company was not discussing downsizing that project. The Milam County, Texas data center — already under construction with 753 megawatts of IT capacity and a 15-year contract with OpenAI — is also unaffected. The $3.7 billion loss came from Silver Bands 7, a subsidiary backed by SoftBank and Vision Fund II-2 that is separate from SB Energy.
What does this mean for the broader AI infrastructure buildout? The consensus among analysts is cautious: SoftBank's procurement missteps are unlikely to trigger mass cancellations of gas-fired power plants. The underlying demand for AI data center power remains enormous — the Brookings Institution estimates spending will hit $10.3 trillion between 2025 and 2032. No single buyer's mistake can reverse that trajectory.
But there is a subtler signal here, one that matters more than the headline loss. SoftBank was the most valuation-tolerant equity buyer in the AI infrastructure space — a balance-sheet financier willing to write checks at marks that public-market investors would not. When CFO Yoshimitsu Goto said at the results call that it was only natural to wait and see how things play out for a little while, he was signaling that the marginal buyer of AI equity was turning price-sensitive. The era of the price-insensitive capital backer is pausing, at least temporarily.
That shift changes the funding calculus for everyone else. If SoftBank slows its pace, the next tranche of AI infrastructure capital has to come from operating cash flow (Microsoft, Google, Amazon, Meta), from Nvidia's vendor-financing capacity, from sovereign wealth funds like MGX and the PIF, or from private credit markets. All of those sources are more selective and demand higher returns, which raises the hurdle rate on marginal AI capacity — particularly for companies like OpenAI that leaned heavily on SoftBank as their largest non-strategic backer.
The lesson for SoftBank is not that AI demand is slowing — it's that the fine print on industrial procurement has real teeth. Prepaying for scarce physical inputs in a fast-moving market is not a strategy; it's a gamble. And when the market shifts beneath you, those prepayments don't come back. For the rest of the AI industry, the takeaway is simpler: the power bottleneck is real, but the path to solving it may look different from what anyone expected in 2025. Grid and batteries. Not gas and turbines. The question is whether the companies that pivoted early — before writing their non-refundable checks — just saved billions, or whether the AI power appetite is so voracious that gas will always be the fallback, regardless of the economics.
Sources: TechCrunch, Surf AI, Unite.AI
✓ Independent sources cross-checked and verified before publishing
