Nvidia isn’t just selling GPUs to OpenAI anymore. It’s underwriting the power lines, the concrete, and the decades-long lease on a decommissioned uranium enrichment site fifty miles south of Columbus, Ohio. The roughly $100 billion credit guarantee that Nvidia is finalizing this month for OpenAI’s massive SB Energy data center campus is the clearest signal yet that the bottleneck in artificial intelligence has shifted from silicon to financing and grid capacity. And that should make investors nervous.
This isn’t a straightforward chip supply deal. Nvidia is effectively acting as the backstop for OpenAI’s 20-year lease obligations on a facility planned to eventually draw up to 10 gigawatts of IT capacity. For context, that’s the power appetite of a mid-sized country. SB Energy, SoftBank’s infrastructure arm, will build and operate the site, but Nvidia’s balance sheet is what makes the project bankable. Without that guarantee, the financing likely doesn’t close.
The Real Product Is Credit, Not Compute
What’s striking is how far this stretches beyond semiconductor sales. I’ve been watching the chatter across trading floors and regional Ohio forums, and the consensus among skeptics is hard to ignore. OpenAI still isn’t profitable. The idea that it will reliably fill a campus consuming more power than most U.S. cities for two decades is, at best, an act of faith. If utilization falls short or lease payments falter, Nvidia absorbs the exposure. That’s not a sales cycle. That’s a credit position.
The deal already got haircuts. Earlier this summer, the guarantee was reportedly floating near $250 billion before investor pressure forced a reset to roughly $100–105 billion. That pullback wasn’t cosmetic. It was a warning shot from shareholders who realized Nvidia was about to turn itself into an infrastructure finance company with a semiconductor side hustle. We saw similar skepticism earlier this year when headlines were throwing around even larger figures for these same projects.
Nvidia is also in talks for a separate $3 billion investment in SB Energy itself, with roughly half at signing and the rest potentially tied to SB Energy’s IPO as early as September. That positions Nvidia not just as the exclusive chip supplier, but as an equity holder in the landlord. It’s a vertical integration play that would have seemed absurd three years ago.
Ohio’s Grid and the Leverage Problem
And then there’s the Ohio factor. Locals are already flagging the strain on regional grids and questioning whether the economic upside (tens of thousands of construction jobs, thousands permanent) justifies the infrastructure burden. The site itself sits on a former uranium-enrichment complex, which adds a layer of environmental and permitting complexity that standard data center builds simply don’t face. This isn’t a blank slate in Arizona or Texas. It’s a repurposed federal site with legacy contamination and a community that wants answers. The local pushback against data center sprawl we’ve been tracking isn’t theoretical here. It’s showing up in county meetings and Reddit threads.
Power is the real constraint. The first 800 megawatts might not come online until 2028, with billions in grid upgrades standing between today’s paperwork and actual electrons. That timeline creates a multi-year dependency risk. If GPU demand cools or OpenAI’s economics shift before the switches flip, Nvidia is still holding the bag on a project it doesn’t even own.
The underreported angle here is replicability. If this Ohio template works, Nvidia can replicate it globally, using its balance sheet to de-risk power and real estate projects in exchange for guaranteed GPU offtake. That’s a brilliant strategic moat. It’s also a dangerous concentration of risk. The company built its valuation on high-margin chip sales. Now it’s anchoring itself to multi-decade infrastructure economics, power grid timelines, and the financial health of a partner that burns cash faster than most nations.
I noticed something telling in the Reddit threads and trading floor discussions this week. The skepticism isn’t directed at the technology. Nobody doubts the demand for AI compute exists. The doubt centers on whether OpenAI can afford to pay for this much of it, and whether Nvidia shareholders signed up to become creditors to the power grid. It’s a sentiment that echoes the broader balance sheet experiments Nvidia has been running across the sector.
Nvidia had to scale this back from $250 billion because its own investors revolted. That should tell you everything. The market will tolerate Nvidia as a chip monopoly, but it’s not yet convinced the company should double as a utility financier. The Ohio campus will probably get built, and the first phase will likely come online by 2028. But the real story isn’t the concrete pouring south of Columbus. It’s that Nvidia couldn’t win the future just by designing better GPUs. It had to put its balance sheet on the line to guarantee anyone would buy them. And if OpenAI’s economics don’t hold up over that 20-year lease, the “AI boom” starts looking less like a technology cycle and more like a leveraged infrastructure bubble. That’s not a chip problem. That’s a credit problem.
