SoftBank Group Corp. and its SB Neo subsidiary are launching a neocloud business in the U.S. to rent out GPU compute resources [1, 2].
This move targets the critical shortage of hardware needed for artificial intelligence development. By providing scalable infrastructure, SoftBank aims to support American efforts in AI training, and large language model deployment [1].
The initiative leverages a 10 GW U.S. server farm [1]. This infrastructure will be managed via the Infrinia AI Cloud OS, a software stack developed by SoftBank's internal team [1].
SoftBank said the Infrinia AI Cloud OS supports Kubernetes-as-a-Service (KaaS) within a multi-tenant environment [1]. The system also provides Inference-as-a-Service (InfaaS), which allows users to access large language model inference capabilities through APIs [1].
Japanese technology conglomerate Softbank Group Corp. and its telecommunications subsidiary SoftBank Corp. are planning to launch the business to address growing demand for compute resources [2]. The entry into the "rent-a-GPU" market places the firm in direct competition with other specialized cloud providers seeking to fill the gap between traditional cloud giants and local hardware clusters [1].
SoftBank said the Infrinia AI Cloud OS is designed to provide a streamlined path for developers to scale their AI operations without the need for massive upfront capital expenditures on hardware [1].
“SoftBank is entering the AI compute market by launching a neocloud business in the US.”
SoftBank's entry into the U.S. neocloud market signals a shift toward specialized, high-density AI infrastructure. By integrating a massive power capacity of 10 GW with a proprietary OS, the company is attempting to vertically integrate the hardware and software layers of AI training. This reduces the reliance of AI startups on the three major hyperscalers and increases the availability of raw compute power for American AI development.
