Nvidia just cut more than $130 billion of potential OpenAI exposure without giving up the part it really wants: the chips.
On Monday, August 17, Nvidia (NVDA) agreed to invest $1.5 billion in SoftBank-backed SB Energy and secure up to 8 gigawatts of AI computing capacity at the PORTS-Pike campus in Ohio. OpenAI signed a 20-year lease for the broader 10-gigawatt site, while Nvidia will be the exclusive chip supplier for roughly the first 5 gigawatts.
That is a much better deal for Nvidia shareholders than the version investors were staring at three weeks ago.
Nvidia found a way to finance demand without owning the risk
The original plan under discussion had Nvidia potentially guaranteeing about $250 billion so OpenAI could lease the project. By Friday, that proposed initial guarantee had been cut to less than $120 billion. The signed structure now has Nvidia backstopping part of the value of completed data centers rather than guaranteeing OpenAI's operating lease.
Nvidia is still helping finance its own customer, but it has moved a large piece of the risk away from its balance sheet.
That distinction matters because the biggest question hanging over the AI boom is no longer whether Nvidia can sell chips. It is who can finance the trillions of dollars of infrastructure needed to keep buying them.
The Ohio campus shows the new model. Nvidia contributes capital, gets exclusive chip status, and helps make the project financeable. OpenAI gets compute. SB Energy gets a 20-year tenant. Lenders get a stronger asset backstop.
The circular-financing concern has not disappeared. Nvidia is still putting money into the ecosystem that buys Nvidia hardware. But Monday's structure is closer to using Nvidia's balance sheet as a bridge than using it as a warehouse for OpenAI's credit risk.
The quieter trade is 9.2 gigawatts of gas
The part I would not ignore is the power bill.
The Ohio plan calls for 10 gigawatts of new generation, including at least 9.2 gigawatts of natural gas, plus $4.2 billion of new transmission investment. AEP Ohio, owned by American Electric Power (AEP), is the utility partner, and it expects power to begin flowing to the site in 2029.
This is not just an AI chip project. It is a $37.5 billion power-and-grid project attached to one data-center campus.
That makes GE Vernova (GEV), Vertiv (VRT), Quanta Services (PWR), Caterpillar (CAT), Baker Hughes (BKR), and AEP worth watching alongside NVDA. None should be treated as an automatic winner from this specific campus without a disclosed contract, but the demand backdrop is already showing up in company numbers.
GE Vernova ended the second quarter with a $176 billion backlog. Its data-center orders topped $5 billion through June, already more than double its total for all of 2025. Gas Power equipment backlog and slot reservations reached 116 gigawatts.
Vertiv raised its 2026 outlook on July 29 to about $14 billion of sales at the midpoint, with 31% organic growth and roughly 60% growth in adjusted EPS at the midpoint.
Those numbers tell me the AI trade is broadening from chips into the physical infrastructure required to make the chips useful. Power generation, transmission, cooling, and electrical equipment are becoming the limit on how fast new compute can come online.
August 26 is the next test
Nvidia reports fiscal second-quarter results on Wednesday, August 26 at 5:00 p.m. Eastern.
The number to watch is not just revenue. Watch what Jensen Huang says about financing commitments, customer credit support, and how much future chip demand now depends on Nvidia helping fund the infrastructure itself.
If Nvidia can keep locking in multi-gigawatt deployments while reducing direct guarantees, the market gets a cleaner version of the AI spending boom. If financing support keeps expanding faster than ordinary customer demand, investors will have a reason to put a higher discount on every new "gigawatt" headline.
