The numbers don’t all mean the same thing. That distinction matters.
On March 20, 2026, a groundbreaking ceremony took place in Piketon, Ohio. SoftBank CEO Masayoshi Son and U.S. Secretary of Commerce Howard Lutnick stood on the grounds of a former uranium enrichment facility, now called PORTS Technology Campus, to announce what the Department of Energy’s official fact sheet describes as a 10-gigawatt AI data center campus developed through a federal public-private partnership. The DOE is providing federal land and permitting support. SoftBank Group is the lead private investor. SB Energy, a SoftBank subsidiary, is building the energy infrastructure. AEP Ohio is the grid transmission partner.
Here’s what’s confirmed. The DOE fact sheet directly states that SB Energy and AEP Ohio have committed $4.2 billion to build new electrical transmission infrastructure and grid upgrades in Ohio. The campus will include a 10-gigawatt data center and up to 10 gigawatts of new power generation capacity, with approximately 9.2 gigawatts coming from natural gas turbines deployed by SB Energy. The groundbreaking occurred. The federal partnership is official.
Here’s what’s reported but not yet T1-confirmed. The first phase of the data center is expected to include approximately 800 megawatts of capacity at a cost of $30–40 billion, with completion targeted for early 2028, according to multiple published reports. Reportedly $33.3 billion in Japanese investment is earmarked for the natural gas generation component, per cross-reference sources without T1 confirmation. Son’s headline number, “The investment starts at $500 billion and could eventually reach $1 trillion” – is his stated aspiration for the full campus buildout, not a committed or contractual figure.
The editorial distinction our audience needs to hold: $4.2 billion is real, confirmed, and infrastructure-specific. The $500 billion is a CEO’s visionary framing of a long-term campus buildout. Treating them as equivalent misreads the deal structure.
The site context adds meaning. Portsmouth’s Gaseous Diffusion Plant enriched uranium for the U.S. nuclear weapons program and later commercial reactors before closing in 2001. The federal government has owned this land ever since. Repurposing legacy federal industrial sites for AI compute is a genuinely new direction in AI infrastructure policy, the DOE isn’t just subsidizing a private project from a distance. It’s contributing the land and the permitting infrastructure that makes the scale possible.
The energy commitment is the detail that deserves attention from the technology and policy community. Nine-point-two gigawatts of natural gas generation, dedicated to a single AI campus, is a concrete statement about what large-scale AI compute actually costs in energy terms. It also arrives at a moment when AI’s electricity consumption is an increasingly active policy and regulatory conversation.
What to watch: whether DOE permitting for the natural gas infrastructure encounters regulatory friction; the pace of the first-phase 800 MW buildout toward the early 2028 target; and whether other federal agencies follow DOE’s model of contributing legacy land to AI infrastructure development. Local reporting from the Scioto Post will be worth tracking for ground-level construction milestones.
The committed figure is $4.2 billion. The aspiration is $500 billion. The precedent is federal land becoming AI infrastructure. All three of those facts belong in the same sentence, with their differences clearly labeled.