SoftBank Plans to Build World's Largest AI Data Center in US: 10GW, Investment Exceeds $66 Billion

How Will AI’s 10GW Power Demand Reshape Local Energy Systems?

IT Home, March 21 — Bloomberg published a blog yesterday (March 20), reporting that SoftBank Group plans to invest over $66 billion in Ohio, USA (IT Home note: approximately 455.595 billion RMB at current exchange rates), to build the world’s largest AI data center with a total capacity of 10 gigawatts (GW).

This massive 10 GW power demand will significantly impact the local energy system. Ohio’s total power generation capacity in 2024 is about 30 GW, meaning this single data center will account for one-third of the state’s total power output.

According to the disclosed plan, SoftBank will invest between $30 billion and $40 billion to develop computing infrastructure, and an additional $33 billion to build a large natural gas power plant. The energy capacity of this plant is comparable to nine nuclear reactors and will become one of the largest gas turbine units in the world.

The data center will be located in Piketon, Ohio, on a 3,700-acre integrated campus. During the Cold War, this site was used for producing weapons-grade uranium, later converted to civilian nuclear fuel production, and finally closed in 2001.

To meet the enormous electricity demand, SB Energy, supported by SoftBank, is deploying approximately 9.2 GW of gas turbine units and plans to add an extra 800 MW of backup capacity.

These turbines will not be concentrated in one location but distributed across the region. Meanwhile, the local utility company will invest $4.2 billion to upgrade transmission and grid infrastructure.

SoftBank has not yet announced what hardware the data center will use, but industry experts expect it will deploy multiple generations of AI chips in phases. The first batch of gas turbines is expected to be installed within a year, likely running NVIDIA’s Rubin architecture or AMD’s Instinct MI455X accelerators in the initial phase.

Source: NVIDIA

The project is expected to be fully deployed by 2030, at which point the facility may utilize NVIDIA’s Feynman (or post-Feynman era) hardware and AMD’s MI600 or even MI700 series chips.

It is worth noting that this super project, valued at over $66 billion, is not related to the previously highly publicized “Stargate” project but is part of a recent bilateral investment plan between the US and Japan totaling $550 billion.