SoftBank has created SB Neo, a U.S.-based AI cloud company that will supply AI chips and infrastructure services to model developers and large enterprises as the company tries to turn its AI infrastructure spending into a standalone compute business.
The company announced SB Neo on July 2, saying the new business will begin providing services in SoftBank’s next fiscal year and will be anchored in the United States. SoftBank Corp. will own 100% of the shares. The announcement also ties SB Neo to a much larger target: building AI infrastructure with a combined power capacity of 10 gigawatts by 2030.
That number is the real signal. A 10-gigawatt target puts SoftBank’s plan in the same conversation as the biggest AI data-center and cloud-capacity bets now being made by hyperscalers, neoclouds, and model companies. The market is no longer only about who can buy the most GPUs. It is about who can secure land, power, cooling, networking, chips, financing, and enough customers to keep enormous clusters busy after the initial shortage cycle passes.
SB Neo is more than a GPU rental shop
SoftBank describes SB Neo as a provider of “AI chips and cloud services,” but the announcement makes clear that the company wants to operate deeper in the stack than simple capacity resale. SB Neo is expected to use Infrinia AI Cloud OS, SoftBank’s software layer for AI data centers. The platform is meant to manage GPU clusters and automate operations for large-scale training and inference workloads.
SoftBank began beta testing Infrinia in May 2026 and has been positioning the system as a way to improve utilization and operations inside AI data centers. In practice, that means the SB Neo story is not only about racks of accelerators. It is also about the scheduling, monitoring, orchestration, and operations software needed to make those accelerators usable for customers that may not want to build an entire AI infrastructure team themselves.
That distinction matters for buyers. Model labs and enterprise AI teams do not only need raw chips; they need predictable cluster availability, high-speed networking, storage, security controls, deployment tooling, and support for long-running training and inference jobs. A provider that can combine capacity with software operations may have a stronger pitch than a smaller GPU cloud that mainly sells access by the hour.
The neocloud market is getting crowded
SB Neo arrives as specialized AI cloud providers are trying to prove they can hold their place beside AWS, Microsoft Azure, Google Cloud, and Oracle. CoreWeave, Nebius, Lambda, Crusoe, and other AI infrastructure companies have all benefited from the gap between model demand and available GPU capacity. At the same time, the largest platform companies are spending aggressively on their own data-center buildouts, and Meta is reportedly exploring a cloud infrastructure business of its own.
SoftBank’s advantage is different from a typical startup neocloud. It has capital-market reach, a long history of technology investing, telecommunications operations through SoftBank Corp., chip exposure through Arm, and a central role in Stargate-related AI infrastructure plans. Those connections do not guarantee execution, but they make SB Neo a more serious entrant than a narrow capacity broker.
The challenge is timing. AI compute has been defined by scarcity, but scarcity does not last evenly across every chip generation, region, workload, and customer type. If too many companies build capacity for the same expected demand curve, the market can move from shortage to pricing pressure quickly. That is why SB Neo’s software and operations layer may matter as much as the 10-gigawatt headline: durable margins will depend on utilization and service quality, not only construction scale.
What enterprise customers should watch
For enterprises, SB Neo’s launch is a sign that AI infrastructure buying is becoming more strategic. Companies that rely on external GPU clouds should not evaluate providers only on headline capacity or chip names. They should ask where clusters will be located, which accelerators will be available, how networking is designed, how data is isolated, what compliance controls are offered, and whether the provider can support production inference as well as large training runs.
Customers should also watch priority rules. If a provider is tied closely to large strategic customers, frontier model labs, or a parent company’s own AI projects, smaller enterprise buyers need clarity on whether their workloads can be delayed, preempted, or repriced during peak demand. Contract terms around reserved capacity, service-level guarantees, export-control compliance, incident reporting, and data handling will matter more as AI clusters become a core part of business operations.
SB Neo is still early. SoftBank has announced the corporate structure, the U.S. focus, the expected service timing, and the 2030 infrastructure ambition, but it has not yet published detailed pricing, regions, hardware configurations, customer commitments, or technical architecture. Those details will determine whether SB Neo becomes a meaningful AI cloud option or mainly a strategic wrapper around SoftBank’s broader infrastructure ambitions.
Even so, the direction is clear. AI cloud competition is moving beyond access to scarce GPUs and toward companies that can finance, power, operate, and sell large-scale compute as a managed platform. SB Neo is SoftBank’s attempt to claim a place in that next phase before the market settles around a smaller set of durable infrastructure providers.
