SoftBank Leans on Energy Subsidiary to Power Its AI Infrastructure Pivot

A Pivot from Venture Bets to Power Plants

In August, residents of Piketon, Ohio gathered to learn about a proposal that would convert the site of a decommissioned uranium enrichment facility into a natural gas power complex. The project, spearheaded by SB Energy, aims to supply electricity directly to what is planned as one of the largest data centre campuses globally. For a village of 2,000, the scale is staggering.

The Piketon initiative represents a marked departure for SoftBank Group, the Japanese conglomerate known for high-profile venture investments in ride-hailing apps, co-working spaces, and e-commerce platforms. Under chief executive Masayoshi Son, SoftBank is now channelling resources into infrastructure that is far more capital-intensive and slower to return profits than the software-driven bets that defined the Vision Fund era.

At Opentechwire, we've tracked SoftBank's portfolio evolution across Asia and beyond for years. This latest move signals something deeper than diversification: it is a recognition that artificial intelligence workloads demand not just algorithms and chips, but also reliable, abundant electrical power and the physical facilities to house compute clusters at scale.

Why Energy Became a Bottleneck

The explosion in demand for AI training and inference has exposed a fundamental constraint: data centres capable of running thousands of graphics processing units require electrical capacity that many existing grids cannot deliver. Hyperscalers and cloud providers have been forced to queue for grid connections, sometimes waiting years for utilities to expand transmission infrastructure.

SB Energy's strategy is to bypass the queue entirely. By constructing gas-fired generation assets adjacent to data centre sites, the subsidiary can guarantee power availability from day one. This co-location model eliminates transmission losses and reduces reliance on overburdened regional grids.

The Piketon site is emblematic. The former uranium plant already possesses grid interconnection infrastructure and sits on land zoned for industrial use, lowering permitting friction. SB Energy plans to build a natural gas facility with capacity sufficient to power a data centre campus designed to host tens of thousands of GPUs.

March saw a groundbreaking ceremony attended by Masayoshi Son and US Commerce Secretary Howard Lutnick, underscoring the political dimension of the project. The Biden administration has prioritised domestic AI infrastructure, and SB Energy's capital commitment aligns neatly with that agenda.

The Economics of Owning the Stack

For SoftBank, vertical integration offers both cost control and strategic optionality. By owning generation, transmission, and compute infrastructure through SB Energy, the group can offer hyperscalers and AI laboratories turnkey solutions: guaranteed power, pre-configured data halls, and flexible lease terms.

This model contrasts with traditional data centre REITs, which lease space but rely on third-party utilities for electricity. When power availability becomes the binding constraint, owning the electrons confers pricing power.

Yet the capital requirements are formidable. Gas-fired power plants cost hundreds of millions of dollars per gigawatt of capacity, and data centre construction at this scale demands further hundreds of millions. SB Energy has reportedly filed for a US initial public offering to fund expansion, though valuation concerns have delayed the timeline according to market observers.

The financing challenge is compounded by rising interest rates, which make infrastructure debt more expensive. Some data centre developers have turned to novel collateral structures, including securitising GPU hardware itself, to access capital markets. SB Energy's approach appears more conventional: equity raise via IPO, supplemented by project finance for individual assets.

Reception and Risk

The pivot has met mixed reactions. In Piketon, local officials have welcomed the employment and tax revenue that construction and operations will generate. A rural community that lost its industrial anchor when the enrichment plant closed sees the data centre project as a lifeline.

Environmental groups and some residents have raised concerns about the reliance on natural gas. While cleaner than coal, gas-fired generation still produces carbon emissions at a time when many technology companies have committed to net-zero targets. SB Energy has not publicly detailed plans for carbon offsets or renewable integration at the Piketon site.

Operationally, the model carries execution risk. Building power plants and data centres simultaneously demands expertise in two distinct domains. Delays in either can cascade: a data centre without power sits idle, while a power plant without an anchor customer bleeds cash.

SoftBank has also announced plans to develop AI infrastructure in Malaysia's Sarawak region, in partnership with Grab, the Southeast Asian super-app. That project similarly envisions co-located generation, though details on capacity and timeline remain sparse.

Implications for the AI Infrastructure Race

SB Energy's strategy points to a broader trend: the infrastructure layer of artificial intelligence is becoming as contested and capital-intensive as the model layer. Nvidia-backed data centre provider Firmus recently scrapped a five-billion-dollar Australian IPO, citing market conditions, illustrating the financing headwinds facing the sector.

At the same time, demand shows no sign of abating. Supply chain participants report continued strong orders for server components, networking gear, and cooling systems, even as some observers question whether the pace of AI investment is sustainable.

For SoftBank, the infrastructure pivot is both a hedge and a bet. If AI workloads continue to scale, owning the physical substrate becomes a toll-road business with durable margins. If the AI boom moderates, the group will be left holding expensive, illiquid assets in a sector notorious for cyclicality.

The Piketon groundbreaking in March was symbolic: Masayoshi Son, who once championed the asset-light model of software investing, literally broke ground on a capital-intensive power plant. Whether this marks a permanent strategic shift or a tactical response to the AI moment will become clear over the next several years as projects come online and financing costs crystallise.

What is already evident is that the race to build AI infrastructure has moved beyond chips and cloud APIs. It now encompasses electrons, substations, and the political economy of rural America.