Nvidia will invest $1.5 billion in SoftBank backed SB Energy to lock up 8 gigawatts of AI compute at an Ohio campus being built for OpenAI.
Nvidia will invest $1.5 billion in SoftBank-backed SB Energy to lock up to eight gigawatts of AI compute at an Ohio campus being built for OpenAI. The deal, confirmed in releases from Nvidia's newsroom and SB Energy on Monday, puts the chipmaker directly into the power layer that consumes its own chips. The structure is the story.
Eight gigawatts is the average peak demand of roughly six million U.S. homes, or a shade under New York City's total summer load. Nvidia is buying the right to that much compute capacity, anchored initially at 4.25 gigawatts on a single site: the PORTS-Pike Technology Campus in Ohio, a facility outside Piketon that the agency is now positioning as the anchor of a regional grid buildout for AI. SB Energy is the developer on the project.
The $1.5 billion figure is the confirmed deal size, per Nvidia's release. That is roughly half the up-to-$3 billion figure Reuters and CNBC reported on Aug. 15, 2026, when the deal was still framed as talks. The gap between the rumoured ceiling and the announced floor is large enough to suggest either a narrower equity stake, a specific tranche, or a structure negotiated down as the parties moved from a wire story to signed releases. None of the company filings spell that out.
The reason the structure matters: Nvidia is now financing the infrastructure that will run its own chips. SB Energy, the developer, is building the campus. OpenAI, the intended anchor tenant, is the largest customer in Nvidia's order book. SoftBank backs both OpenAI and SB Energy, and per Reuters, sits on the cap table for each. Nvidia's $1.5 billion flows into a project that will eventually buy more of Nvidia's chips. Each layer of the stack (chips, power, data-center shell, anchor tenant) is a known Nvidia counterparty, and the capital follows the same loop.
Nvidia's release calls the SB Energy campus the guaranteed home of Nvidia AI compute. SB Energy's release says the company will develop at least ten gigawatts of new power generation in Ohio and invest $4.2 billion in grid infrastructure alongside SoftBank. The Department of Energy's Ohio fact sheet treats the AI power draw as a planning input for the regional grid. Each statement is consistent with the others, because they describe the same project from different seats.
Business Standard's re-report does exactly that. What the wire framing does not name is the mechanism: chipmaker capital financing the power layer that consumes the chipmaker's own product. That loop is the structural change, because the dollars, the power, and the customer are no longer sitting at different points in the AI supply chain. They are sitting in the same cap table.
The strongest falsifier for that framing would be a standard power-purchase offtake: a contract in which SB Energy sells electrons to a third party at a fixed price, with no equity or revenue link back to Nvidia. If the project clears the way it clears, with Nvidia holding capacity rights, an equity stake, and an offtake, the loop is structural rather than rhetorical. If the structure turns out to be a conventional lease or a conventional power offtake, the circularity language is just shorthand. The filings as written point to the first case.
A second-order effect to watch: credit. The same circular flow that helps Nvidia book demand for chips also concentrates the credit risk in the same place. A stress event at any node, OpenAI, SoftBank, SB Energy, or the Ohio grid, now reaches all the others through the same loop. None of the company releases disclose how that risk is allocated, and the Department of Energy fact sheet frames the policy upside without addressing concentration.
The deal closes the chapter on the Aug. 15 wire reports. The next chapter is the structure: whether the cap table is durable, whether the power-permitting path holds, and whether the $1.5 billion equity check is the only Nvidia money heading into the campus, or the first tranche of several.