While SoftBank may not always be the first to jump on emerging trends, its investments, when made, are notably substantial.
This strategy is evident in its substantial $30 billion commitment to OpenAI in early 2025, and its February announcement that its SB Energy division would allocate $33 billion to construct a 9.2-gigawatt natural gas power facility. The approach seems to be: if you can’t be first, be the biggest.
This expansive philosophy reportedly extended to its procurement practices, where orders were similarly scaled up. Approximately two years ago, SoftBank reportedly committed deposits for over 13 gigawatts of turbines, a volume nearly 50% greater than its actual requirements.
This decision to acquire excess turbines likely facilitated quicker access to equipment and accelerated SoftBank’s project timelines.
“SB Energy essentially had no alternative but to pre-order a surplus of turbines if they aimed to secure some immediately,” Peter Gardett, CEO of Noreva, conveyed to TechCrunch via email.
The energy sector was unprepared for the rapid increase and sheer scale of power demand from AI. Consequently, lead times for turbine orders now stretch into the early 2030s. As natural gas emerged as a primary power source for the AI surge, CEO Masayoshi Son wagered that by leveraging his investment firm’s substantial capital, he could control a segment of the market. His strategy implied that if the AI industry needed power for its data centers, they would need to approach SoftBank.
However, that substantial bet now appears to have faltered. SoftBank declared a $7.1 billion quarterly loss on Tuesday, with over half of this amount attributed to impairments related to power-generating equipment. The investor had made non-refundable deposits to turbine manufacturers to secure these orders, and much of these are now deemed valueless.
This action also represented a costly exercise in market price discovery, according to Gardett. Instead of potentially overpaying and still facing turbine shortages, SoftBank opted to overpay and order in excess.
“This specific charge signifies that price discovery process finally materializing,” Gardett commented.
The sheer size of this miscalculated investment—exceeding $3 billion—highlights the intense demand for electricity gripping the AI sector. Confronted with multi-year delays for grid connections, numerous hyperscalers and developers have gravitated towards natural gas, perceiving it as a reliable and familiar energy fallback.
As AI data centers operate for longer durations, their managers are gaining a clearer understanding of operational requirements. While natural gas might appear to be a secure option, it carries inherent risks, such as winter supply curtailments during demand spikes and supply dips, as Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, conveyed during the Equity podcast on Wednesday.
Given these considerations, Nelson suggested that the established grid might still represent the optimal solution. Most data centers would benefit most from a grid connection, even if it’s smaller than their usual requests. To compensate for any shortfalls, substantial battery storage systems offer a cost-effective alternative, according to Nelson. These could assist in managing peak loads or responding to utility requests for electricity usage reductions.
It is plausible that data center operators are arriving at similar conclusions, potentially causing the natural gas market to expand at a slower pace than SoftBank had projected.
SoftBank’s losses tied to natural gas are unlikely to trigger a widespread cancellation of gas power plant projects, given the immense power demands of AI. For the time being, natural gas remains a prominent energy source, although SoftBank may have indeed reached its limit in this particular market.