When tech giants announce massive infrastructure projects, the public conversation typically fixates on job numbers and investment totals. The SoftBank deal announced in Piketon, Ohio, this week offers a valuable lesson in what happens when we look past the headlines: the arrangement reveals real tensions between private gain and public cost in the race to build AI infrastructure.
SoftBank Group's SB Energy is planning to build 10GW of new power generation, including 9.2GW of natural gas generation, that will connect to the local grid and provide power to a new 10GW data centre development at the Portsmouth Site in Pike County, Ohio. The facility will occupy a 3,700-acre complex that once produced weapons-grade uranium during the Cold War.
Here is where the fiscal responsibility question matters. SB Energy is investing $4.2 billion with AEP Ohio to upgrade and build new transmission lines in Southern Ohio. On paper, that looks like private investment shouldering the cost of grid expansion, which is precisely what advocates for cost-shifting argue should happen. The project is framed as consistent with President Trump's Ratepayer Protection Pledge, which aims to shield consumers from price increases driven by surging AI infrastructure demand.
Yet the broader context complicates this picture. Electricity demand is rising faster than the US power grid, much of it built decades ago, was designed to handle, making power a central operational and strategic constraint shaping where data centres are built. When one company's infrastructure demand begins reshaping the entire electrical system, the costs inevitably spread across the system itself. AEP Ohio has paused all new data centre interconnections due to insufficient power infrastructure, meaning the SoftBank project has effectively locked in capacity that smaller, less politically connected companies cannot access.
The scale is undeniably enormous. According to US Energy Secretary Chris Wright, if the facility were completed today it would be larger than all other data centres in the world combined. Project leaders estimate 35,000 construction jobs and 2,500 permanent jobs once the data centre is complete.
There are legitimate counterarguments here. Foreign investment in critical infrastructure carries strategic value. Japan has committed to invest $550 billion across America, with this $33 billion power project in Portsmouth, Ohio as one of the largest components. SoftBank's willingness to fund its own grid upgrades is genuinely preferable to utilities absorbing those costs. The project is expected to support thousands of jobs, expand domestic manufacturing, and contribute excess generation and transmission capacity back to the grid to help lower electricity prices, while also including environmental remediation efforts at the Portsmouth site and community development initiatives.
The challenge is one of timing and concentration. Global power demand from data centres is forecast to increase 50 per cent by 2027 and by as much as 165 per cent by 2030 compared with 2023, according to Goldman Sachs research. If every company building at scale funds its own infrastructure, we may avoid direct cost-shifting to ratepayers in the short term. But the combined $320 billion-plus in data centre capital expenditure from just five companies in a single year represents an unprecedented concentration of infrastructure investment, with the technology sector now outspending the entire US electric utility industry on energy-adjacent infrastructure by a factor of two.
The Portsmouth project represents rational behaviour within existing rules. SoftBank is not breaking any pledge or violating any agreement. What it reveals, however, is a question policymakers must now confront: Are today's cost-allocation rules sufficient when demand of this magnitude reshapes entire regional systems? Or will individual ratepayers, even with such projects theoretically shielded from infrastructure costs, ultimately subsidise private computing through higher baseline electricity prices across the grid?
The answer matters far more than the job projections.