SoftBank has announced it will enter the U.S. neocloud market with a new company called SB Neo, which will rent GPU compute to hyperscalers for AI training. The service is planned to begin in fiscal 2027.
Background
The capacity will make use of a roughly 10‑gigawatt server farm that SoftBank is building in the United States. Masayoshi Son, CEO of SoftBank Group, described the project as "world‑class AI infrastructure" during the announcement.
The timing of the move also reflects recent financing developments at SoftBank Group: the company returned to lenders seeking a $10 billion loan secured against its stake in OpenAI. Banks raised concerns about the collateral, and Masayoshi Son ultimately provided a personal guarantee for the loan.
Market context and competitors
SoftBank's decision to monetize GPU capacity follows similar actions by other large technology firms. SpaceX has offered spare compute for sale, Meta is opening some of its clusters, and incumbent neocloud providers such as CoreWeave already maintain large pools of rentable GPUs. Together, these developments indicate several companies that overbuilt capacity for AI are now looking to sell excess compute rather than leave it idle.
Why this matters
A broad industry effort to lease out surplus compute suggests that persistent scarcity of GPU capacity may be less acute than earlier investment narratives assumed. If many major players are attempting to offload compute, the market may be moving from a phase of perceived scarcity to one of growing supply, with implications for pricing and future capital expenditure in AI infrastructure.
Conclusion
SB Neo lets SoftBank repurpose its U.S. server investment to generate revenue, but the launch is also intertwined with the group's recent financing decisions. The $10 billion loan and the personal guarantee from Masayoshi Son underscore the liquidity pressures and risk management considerations behind the strategy, while the wider industry trend of renting excess compute raises questions about the long‑term balance of supply and demand in GPU infrastructure.
