SoftBank has established SB Neo, a new $100B+ AI cloud computing service for the U.S. market; OpenAI is reportedly being considered as an early anchor customer.
On July 2, SoftBank Group and its telecommunications subsidiary SoftBank Corp. announced the establishment of SB Neo Inc., a Delaware-registered joint venture, formally entering the US "neocloud" market this month. The company will be 51% owned by SoftBank Corp. and 49% by SoftBank Group.
SB Neo will leverage the 10-gigawatt-class energy and AI infrastructure currently being developed by SoftBank Group, with plans to launch neocloud services in fiscal year 2027 (ending March 2028). The company will provide large-scale computing resources required for AI model training and inference to major US enterprises. Drawing on experience from its AI GPU cloud service testing program in Japan, which began in May 2026, SoftBank will apply these learnings directly to the US market.
Junichi Miyakawa, head of SoftBank Corp., characterized this US expansion as the company's "second founding." According to sources familiar with the matter, if the US neocloud business expands smoothly, the annual operating profit of SoftBank's telecom subsidiary is expected to grow 3 to 4 times from current levels, reaching 3 trillion to 4 trillion yen (approximately $18.5 billion to $25 billion).
SB Neo aims to increase its data center supply capacity to 10 gigawatts by around 2030. This target is underpinned by the Ohio data center project announced in March 2026, which features an investment of up to $500 billion for a single campus—among the largest in the world—powered by a $33 billion natural gas power plant. For context, the electricity required to run a 1 GW data center can power approximately 750,000 homes simultaneously.
SB Energy, SoftBank's energy arm, has secured the natural gas turbine supplier for the Ohio project, with all equipment scheduled to be operational by the end of the decade. The first phase of the data center, featuring around 800 megawatts of power supply, is expected to be completed in early 2028 at a cost of $30 billion to $40 billion.
Junichi Miyakawa views power procurement as SoftBank's key competitive differentiator. In the current computing power race, a stable and sufficient power supply has become one of the core bottlenecks constraining data center expansion, and SoftBank aims to leverage this advantage as a competitive barrier for its US market entry.
OpenAI may become one of SB Neo's first customers. SoftBank Group has committed to investing approximately $65 billion cumulatively in OpenAI by October 2026. This deep partnership provides potential foundational demand support for the new cloud business.
Competition in this space is already intense. Specialized neocloud providers like CoreWeave and Nebius have already captured significant market share; Amazon AWS, Microsoft Azure, and Google Cloud also offer AI compute leasing services. Meta Platforms, announced on the same day, is also planning a cloud computing business, with plans to sell surplus AI compute capacity. Major technology companies are increasingly monetizing their idle compute capacity, reshaping the supply and demand dynamics of the AI computing power market.
SoftBank timed the SB Neo launch to coincide with the completion of its $10 billion additional investment in OpenAI, as well as restarted negotiations for a $10 billion loan backed by OpenAI equity. Together, these three developments illustrate SoftBank's strategic logic: simultaneously betting on "investing in AI companies" and "building AI infrastructure," attempting to position itself across the AI value chain from both capital and asset perspectives.
The challenges, however, are equally apparent. The neocloud sector is rapidly becoming crowded, with power, land, and equipment all scarce resources. SB Neo is not scheduled to launch until fiscal year 2027, while competitors are already operating. Whether SoftBank's 10-gigawatt blueprint can be delivered on schedule will be a critical test of this Japanese giant's transformation in the coming years.