SoftBank unveils plans to enter the US neocloud business with SB Neo

Japanese technology conglomerate Softbank Group Corp. and its telecommunications subsidiary SoftBank Corp. are planning to launch a new neocloud business in the U.S. that will rent access to artificial intelligence chips to hyperscalers and other businesses in need of compute resources.

The company said it’s going to establish a new entity called SB Neo Inc., which will operate as a neocloud business when it begins operations in fiscal 2027 next March, according to a press release. Ownership of the necloud business will be split, with SoftBank Corp. having a 51% stake, while SoftBank Group Corp. will get 49%. SB Neo will then operate as a consolidated subsidiary of SoftBank Corp., which itself is a subsidiary of SoftBank Group Corp.

Neoclouds are a relatively new and fast-growing segment within the cloud computing infrastructure business, which focus on renting access to graphics processing units and other AI accelerators. They sprang up to take advantage of the massive demand for GPUs. They’re essentially niche cloud platforms that are laser-focused on providing AI services.

However, there are questions about the financial stability of neoclouds. A report last year by McKinsey & Co. warned that the business model of most neoclouds is extremely fragile because of its commoditized nature, with operators only able to offer limited differentiation based on the specific hardware they can offer. Whether or not SoftBank knows something that McKinsey doesn’t isn’t clear.

SoftBank Corp. says it has been running a beta operation of its GPU cloud service in Japan since April. It’s powered by a software stack called Infrinia AI Cloud OS, which is a software stack for AI data centers. The company says it wants to apply the expertise it has gained from running this business on a much bigger scale in the U.S.

Infrinia AI Cloud OS was developed by SoftBank Corp.’s Infrinia development team and supports Kubernetes-as-a-service in multitenant environments and inference-as-a-service capabilities for large language models. In a prepared remark, SoftBank Chief Executive Masayoshi Son said “The SoftBank Group will work together to deploy world-class AI infrastructure and drive the AI revolution.”

A report in the Japan Times says SB Neo will eventually supply around 10 gigawatts of AI compute capacity to U.S. customers by 2030 for scaling AI model training and inference workloads. In doing this, SoftBank Group hopes to triple or even quadruple its annual operating income to between $18.5 billion and $25 billion, according to people familiar with its plans.

It shouldn’t come as a surprise that Son wants to go after the neocloud market, said Holger Mueller, an analyst with Constellation Research. But he’s chosen to enter the race at an interesting time, as many neocloud providers are retooling their infrastructure offerings to focus on AI inference rather than training. “There’s an opportunity for SB Neo to redefine what an inference-focused neocloud for enterprises actually looks like,” the analyst said. “The key will be having access to enterprise data in as close to real-time as possible, or else having extremely fast data replication access to relevant enterprise data.”

SB Neo may have a ready-made customer in the AI giant OpenAI Group Corp., which has benefited massively from SoftBank’s largesse in recent years. The technology group has invested a staggering $65 billion into the ChatGPT maker, and that’s likely to ensure it has enough influence to sway OpenAI CEO Sam Altman into buying into its neocloud vision.

However, SB Neo will still face a lot of competition from neocloud rivals such as CoreWeave Inc., as well as traditional public cloud operators Amazon Web Services Inc., Microsoft Corp. and Google Cloud, which also sell access to GPUs and their own specialized AI chips. Meta Platforms Inc. also has plans to enter the neocloud business. However, Junichi Miyakawa, the head of SoftBank Corp., told the Japan Times that SB Neo has an advantage thanks to its ability to secure source of energy from gas-fired plants.

SoftBank also revealed that it’s planning to push ahead with its plans to construct a number of new gigawatt-scale data centers in Japan later in the year. One thing it did not talk about, though, was how it intends to finance all of these plans. Bloomberg reported separately that SoftBank has reengaged with lenders in an effort to secure a $10 billion loan that’s backed by its stake in OpenAI. Lenders were initially less than enthusiastic about such a transaction at first, but the company is said to be offering to guarantee repayment of the loan, which could give them some recourse if the shares in OpenAI pledged as collateral lose their value.

Son has been one of the most vocal proponents of AI, and recently told shareholders that talk of the industry being in a bubble is an “insult to AI.” He added that he thinks it’s “blasphemy against AI if you say it’s a bubble.” His comments came after Altman last year stated that he does think the AI industry is in a bubble.

Photo: SoftBank