Nvidia is in formal talks to back OpenAI's planned datacenter in southern Ohio, a $250 billion project that would mark the chipmaker's largest move into physical infrastructure capital. The arrangement centers on a 10-gigawatt facility being developed by SoftBank's energy subsidiary, with Nvidia providing capital to help OpenAI secure the lease. No binding agreements exist. Yahoo Finance reported the discussions Wednesday morning.
The Ohio site sits in a corridor where American Electric Power already holds transmission capacity and where Duke Energy has filed permits for grid extensions. SoftBank's energy arm acquired land options in the region in late 2024, betting on compute density requirements that exceed what hyperscalers can build alone. OpenAI's board approved datacenter expansion plans in December but left financing structures open. Nvidia's participation would convert customer dependency into balance-sheet alignment—OpenAI uses Nvidia GPUs almost exclusively, and a lease guarantee ties hardware orders to fixed timelines. The 10-gigawatt scale exceeds the combined power draw of all AWS us-east-1 availability zones.
This matters because Nvidia is inserting itself into the capital structure of AI training, not just the supply chain. If Nvidia guarantees OpenAI's lease, it effectively underwrites the revenue visibility that SoftBank needs to finance construction, and OpenAI gets datacenter capacity without front-loading equity dilution. The move pulls forward chip demand—long-term lease commitments let OpenAI lock in multi-year GPU orders, smoothing Nvidia's revenue curve past the volatile hyperscaler cycle. It also answers a structural question the market has circled for eighteen months: whether Nvidia would tolerate customer concentration risk or engineer capital structures that bind customers to hardware roadmaps. This is the latter. Worth noting that SoftBank holds a $5.5 billion position in Nvidia common stock, disclosed in November filings, so the energy subsidiary's economics tie back to equity upside in both directions.
Operators should watch three near-term events. First, whether American Electric Power files interconnection studies for additional southern Ohio capacity—those studies take 90 to 120 days and signal whether the grid can absorb 10 gigawatts without new transmission builds. Second, whether OpenAI closes its announced $40 billion equity round in Q2, which would clarify how much datacenter capital it can self-fund versus what it needs underwritten. Third, whether Nvidia's April earnings call mentions customer financing structures—management has avoided the phrase in prior quarters, but a $250 billion lease backstop would require board approval and likely disclosure under vendor-financing rules. A former OpenAI employee's hedge fund disclosed large Nvidia options positions this week, though the connection to this transaction is unclear.
The Ohio project is not yet financed, not yet permitted, and not yet contracted. But Nvidia's willingness to talk signals that it views datacenter capital as a margin-preservation tool, not a distraction. SoftBank's energy unit gets a creditworthy anchor tenant. OpenAI gets compute capacity scaled to its training roadmap. The structure is rational if the partners believe AI training demand will absorb 10 gigawatts before Ohio's grid needs expensive upgrades. Grid studies will tell us if the grid believes it too.