What SoftBank actually announced
SoftBank formed SB Neo, Inc. on July 2, 2026, a Delaware-incorporated subsidiary owned 51% by SoftBank Corp. and 49% by SoftBank Group Corp. Its business is renting Nvidia GPU compute for AI training and inference to US customers, with service scheduled to begin in fiscal year 2027, the year ending March 31, 2028. SoftBank Group chairman Masayoshi Son framed the move as working to "deploy world-class AI infrastructure and drive the AI revolution," tying SB Neo directly to the group's broader US AI infrastructure ambitions rather than treating it as a standalone side bet unconnected to the rest of the company's strategy.
SB Neo is not a cold start, and that distinction matters for how seriously buyers should take the timeline. It builds directly on SoftBank Corp.'s beta GPU cloud, which has been running in Japan since May 2026 on the company's own Infrinia AI Cloud OS stack, already handling real workloads rather than existing only as a roadmap item. That gives SB Neo working software and genuine operational experience before it takes a single US customer, a meaningfully different starting position than most neoclouds that end up building their stack while onboarding early tenants at the same time.
The scale SoftBank is targeting
SoftBank is scaling SB Neo in phases toward 10 gigawatts of AI data center capacity, a figure that places it in the same conversation as the largest dedicated AI infrastructure buildouts announced anywhere this year, not a boutique GPU reseller testing the waters of a new market. SoftBank Corp. is separately planning gigawatt-scale AI data centers inside Japan once site preparations are complete, which signals the US launch is one piece of a coordinated global capacity strategy rather than an isolated, opportunistic bet on a single geography.
That scale matters for buyers because capacity commitments at gigawatt scale typically come with the balance sheet to back multi-year contracts without resorting to the single-tenant debt structures some neoclouds have used to fund their GPU purchases in the current financing environment. SoftBank's history of large, patient infrastructure bets, from telecom networks to its Arm investment, gives SB Neo a meaningfully different credibility profile than a venture-backed neocloud still working through its first institutional debt round to fund expansion.
Who SB Neo is actually competing with
SB Neo enters a GPU cloud market already populated by CoreWeave, Crusoe, Lambda, Nebius, and Vast.ai, providers that built their businesses specifically around GPU availability, flexible pricing, and bare-metal access rather than the bundled proprietary services hyperscalers sell alongside compute. The GPU-as-a-Service market itself was valued at roughly $3.2 billion in 2023 and is projected to approach $50 billion by 2032, a growth rate that keeps attracting new entrants with deep pockets.
What differentiates SB Neo from most of that field is that it arrives with an already-operating cloud OS stack and a named path to hyperscaler-adjacent customers, not just AI startups. SoftBank explicitly lists hyperscalers as a target customer segment, which suggests SB Neo intends to sell overflow or specialized GPU capacity to the same companies that operate AWS, Azure, and Google Cloud, a positioning few existing neoclouds have stated as directly. That ambition also puts SB Neo in an unusual position of being both a potential supplier to, and a competitor against, the very hyperscalers it hopes to count as early customers.
What this means for capacity planning now, not in 2028
Fiscal 2027 service start is roughly a year and a half out from today, which means SB Neo is not a near-term option for anyone actively capacity planning for 2026 or early 2027 workloads that need compute booked now. The more immediate value for CIOs is as a signal rather than a procurement option: a well-capitalized conglomerate with its own operational AI cloud stack judged the US neocloud market attractive enough to enter from scratch rather than acquire an existing player, which says something concrete about how much headroom SoftBank sees in GPU rental demand relative to current hyperscaler supply.
For infrastructure teams building a multi-year GPU sourcing strategy, SB Neo is worth tracking starting now, well before it has a sales team making calls, precisely because entering the evaluation pipeline early gives procurement real leverage once service actually launches. Vendors courting their first enterprise cohort typically negotiate harder on price and contract terms than they will once reference customers are locked in and the provider has pricing power it did not have at launch.
The broader signal for enterprise cloud strategy
SB Neo's launch adds to a pattern worth naming directly for anyone setting cloud sourcing strategy this year: the GPU cloud market is attracting entrants with fundamentally different capital structures, from venture-and-debt-funded neoclouds to conglomerate-backed subsidiaries with decades of infrastructure financing experience behind them. That diversity is good news for buyers worried about vendor concentration risk in AI compute, since it widens the pool of credible alternatives to the three hyperscalers well beyond a handful of similarly structured, similarly financed startups competing on the same narrow terms.
The practical move for enterprise architects is to keep a running shortlist of qualified alternative GPU providers, refreshed on a quarterly cadence, rather than treating neocloud selection as a one-time vendor bake-off conducted once and never revisited. A market this dynamic, with new entrants carrying genuinely different risk profiles arriving every few months, rewards buyers who already know who the new credible entrants are well before a capacity crunch forces a rushed, poorly informed decision under time pressure. SB Neo belongs on that list the moment it opens a sales conversation, not a year after competitors have already locked in reference pricing.
