Infrastructuresoftbankneocloudrent a gpudata centers

SoftBank launches SB Neo neocloud service in US

||By LDS Team
6.8
Relevance Score
SoftBank launches SB Neo neocloud service in US
Photo: image.theregister.com · rights & takedowns

SoftBank Corp. and SoftBank Group Corp. announced on July 2, 2026 they will jointly establish SB Neo, Inc., a new US neocloud subsidiary that will begin offering GPU cloud services to American enterprises and hyperscalers in fiscal year 2027 (ending March 31, 2028). SB Neo will be 51% owned by SoftBank Corp. and 49% by SoftBank Group Corp., and plans to scale toward 10 gigawatts of AI data-center capacity, drawing on the SoftBank Group's broader US infrastructure buildout. The company will apply lessons from SoftBank Corp.'s beta GPU cloud service in Japan, which has run on its Infrinia AI Cloud OS software stack since May 2026. Chairman Masayoshi Son said the group will "work together to deploy world-class AI infrastructure and drive the AI revolution."

For practitioners planning AI training or inference capacity, a new entrant building toward 10-gigawatt scale signals that US GPU rental supply is set to expand meaningfully over the next two years, at a moment when hyperscalers are also racing to add proprietary capacity.

What happened

SoftBank Corp. and SoftBank Group Corp. announced on July 2, 2026 that they will establish SB Neo, Inc., a Delaware-incorporated subsidiary that will operate a neocloud (rent-a-GPU) business in the United States, according to the companies' joint press release. SB Neo will be 51% owned by SoftBank Corp. and 49% owned by SoftBank Group Corp., and will launch neocloud services in fiscal year 2027 (ending March 31, 2028), providing GPU compute for AI model training and inference to major US enterprises and hyperscalers. The company plans to scale capacity in phases toward a 10-gigawatt target, drawing on SoftBank Group Corp.'s broader US AI infrastructure buildout.

Background

SoftBank Corp. has run a beta GPU cloud service in Japan since May 2026, built on its Infrinia AI Cloud OS software stack, and will apply that operational experience to the US launch, per the companies' press release. SoftBank Group Chairman and CEO Masayoshi Son said, "The SoftBank Group will work together to deploy world-class AI infrastructure and drive the AI revolution." SoftBank Corp. President and CEO Junichi Miyakawa added that the company plans to proceed with gigawatt-scale AI data centers in Japan once preparations are complete.

For practitioners

A new large-scale neocloud entrant expands the pool of GPU rental capacity available to teams that cannot or do not want to build proprietary data centers, but the rent-a-GPU segment has grown increasingly price-competitive as more capacity comes online. Teams evaluating SB Neo or similar providers should weigh API-level features, SoftBank's Infrinia stack offers Kubernetes-as-a-Service and Inference-as-a-Service, alongside raw GPU pricing and availability.

What to watch

SB Neo's pricing and initial hyperscaler or enterprise customers once services launch in fiscal 2027; further disclosures on the pace of SoftBank's 10-gigawatt buildout; and whether other Japanese or Asian conglomerates follow SoftBank into the US neocloud market.

Key Points

  • 1SoftBank Corp. and SoftBank Group Corp. will jointly launch SB Neo, a US neocloud subsidiary offering GPU compute to enterprises starting fiscal year 2027.
  • 2SB Neo plans to scale toward 10 gigawatts of AI data-center capacity, building on SoftBank's existing Japan-based Infrinia GPU cloud service.
  • 3New large-scale neocloud entrants expand available GPU rental supply but compete in an increasingly commoditized, price-competitive rent-a-GPU market segment.

Scoring Rationale

A major, well-capitalized conglomerate committing to 10-gigawatt-scale US AI cloud infrastructure is a significant capacity signal for the rent-a-GPU market, confirmed via SoftBank's own press release and corroborated by Bloomberg. Kept in the notable tier since it is a capacity/business announcement rather than a new model or technical breakthrough.

Sources

Public references used for this report.

4 sources

Practice with real Banking data

90 SQL & Python problems · 15 industry datasets

250 free problems · No credit card

See all Banking problems