Capital allocated to bypass electrical grid bottlenecks carries immediate balance-sheet depreciation when hardware procurement runs ahead of actual deployment. SoftBank announced a $7.1 billion quarterly loss, with over half of that sum stemming from losses related to power-generating equipment. The write-down exposes the financial hazards of attempting to solve AI energy shortages through unilateral equipment accumulation.
Multi-billion-dollar losses on gas-fired generation show that aggressive independent hardware purchasing cannot immediately resolve the energy crunch facing artificial intelligence infrastructure. Starting about two years ago, SoftBank put down deposits on more than 13 gigawatts of turbines, booking almost 50% more than it needed. Queues for turbine deliveries now extend into the early 2030s, creating an environment where securing capacity required extreme advance commitments.
The Price of Cornering Turbine Supply
Facing multi-year interconnect queues, hyperscalers turned to natural gas as a familiar energy bridge, and SoftBank attempted to capture that commercial demand. In early 2025, the firm committed a $30 billion bet on OpenAI, followed by an announcement that its subsidiary, SB Energy, would spend $33 billion on a 9.2-gigawatt natural gas power plant. According to Bloomberg, the Japanese conglomerate also intends to secure a $10 billion margin loan backed by its stake in OpenAI, while partnering with the company on The Stargate Project, a $500 billion venture aimed at overhauling U.S. infrastructure for AI needs over the next four years.
To lock down the hardware for these ambitions, SB Energy contracted far beyond its baseline operational volume. Peter Gardett, CEO of Noreva, told TechCrunch via email that SB Energy had no alternatives under market pressure:
"SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away,"
Peter Gardett explained that the move was also a costly bit of price discovery, noting that the resulting charge is simply that price discovery finally taking effect on the balance sheet. Rather than risk entering the market short of hardware, SoftBank chose to overpay and over-order, resulting in nonrefundable deposits that the company now considers largely worthless. What is the real cost of attempting to bypass electrical shortages through speculative generation orders?
Structural Limits and Grid Realities
The financial damage from the misplaced bet exceeded $3 billion, reflecting how quickly supply-chain hedges can turn into stranded capital. The market for natural gas infrastructure has developed more slowly than SoftBank anticipated, largely because on-site thermal generation introduces operational problems of its own. Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, told the Equity podcast that natural gas carries risks including winter curtailments when fuel supplies dip and regional demand spikes.
Because of those supply disruptions, data center developers are evaluating alternative architectures rather than relying entirely on captive fossil plants. Brent Nelson explained that most facilities would be best served by standard utility interconnections, even if forced to accept smaller capacities than originally requested:
"To make up for the shortfall, large batteries are a cost-effective choice, Nelson said. They’d help shave peak loads or ride out requests from the utility to curtail electricity use."
Capital requirements for batteries and reduced utility interconnections are competing directly with standalone thermal generation plants. SoftBank’s write-down will not cause a domino effect or mass cancellations of gas power plants, because aggregate AI electricity demand remains too massive for a single cancellation cycle to upend. Yet hoarding generation equipment cannot insulate data center developers from physical supply bottlenecks, turning speculative infrastructure procurement into an expensive balance-sheet liability.
Why it matters
The financial penalty for stockpiling hardware ahead of actual deployments is mounting, as hoarding generating capacity fails to insulate AI buildouts from grid bottlenecks. SoftBank’s gambit to beat vendor queues by reserving thirteen gigawatts of gas turbines proved that securing machinery without operational clearance turns prepayments into deadweight capital. While the group committed to surplus volumes to anchor massive initiatives like The Stargate Project, seasonal fuel constraints and practical site limits quickly outpaced the balance-sheet hedge. Throwing billions at equipment queues cannot substitute for the slow reality of plugging power directly into the grid.