Analysis
Nvidia is backing as much as $105 billion in financing for the first phase of an OpenAI data center in Ohio, according to CNBC, which cites people familiar with the arrangement. The site, in Pike City, Ohio, will be built and operated by SB Energy, a subsidiary of SoftBank Group, with OpenAI signing on as tenant through a 20-year lease. Bloomberg reports the financing covers the project's initial 4.25 gigawatts of computing capacity, with an option to add another 3.75 gigawatts -- enough combined capacity to rival a mid-sized regional power grid. Pulse previously covered SoftBank's escalating infrastructure commitments alongside OpenAI, of which SB Energy's role here is the latest and largest installment.
This is not a check Nvidia is writing from its balance sheet in the way a venture round works. It is a financing guarantee: Nvidia is putting its credit behind the debt that gets the concrete poured, the transformers installed and the GPUs racked, which lets OpenAI and SB Energy borrow the capital needed to build at this scale without OpenAI carrying the debt on its own books. Nvidia separately agreed to invest $1.5 billion directly into SB Energy, layered on top of a $1 billion infrastructure commitment OpenAI and SoftBank made together earlier this year.
Why Ohio, why SoftBank
SB Energy is the same SoftBank unit that has been racing to stand up gigawatt-scale campuses across the U.S. as part of the Stargate buildout, and Ohio has become a magnet for this kind of project because of cheap land, available power interconnects and state incentive packages that other states have been slower to match. OpenAI framed the jobs number -- 35,000 construction roles through 2032, 2,500 permanent positions after -- as the kind of economic pitch that state governments have started demanding in exchange for tax abatements on projects this large.
The structure matters more than the headline number. Vendor financing arrangements like this one let Nvidia lock in a buyer for years of future GPU shipments while OpenAI avoids putting $105 billion of new debt directly on its own balance sheet ahead of a widely expected IPO. It is the same playbook Nvidia has now run with CoreWeave, Lambda and a widening set of neoclouds: back the capital structure, guarantee the debt, and collect the margin on every chip that fills the building. The company's SpaceX equity position, disclosed separately this week at roughly $21 billion, is a reminder that Nvidia's balance sheet has become as central to the AI buildout as its chip roadmap.
The math skeptics will run
What gets lost in the topline figure is that "up to $105 billion" is a ceiling on a financing guarantee, not a wire transfer. The number assumes both phases get built, power gets delivered on schedule to a rural Ohio grid that has never hosted a load this size, and OpenAI's revenue keeps compounding enough to justify leasing capacity this large through 2046. OpenAI's own revenue run rate is reportedly near $40 billion annualized -- a strong number, but one that makes a 20-year lease on a $105 billion facility a bet on sustained, multi-decade demand for inference and training compute rather than a near-term utilization problem.
The counterweight is regulatory and physical, not financial: Ohio's grid operator, PJM Interconnection, has already flagged capacity constraints from AI data center demand across its territory, and a single 4.25-gigawatt load is large enough to force rate conversations with residential customers in the region. If local utilities push back on interconnection timelines the way they have in Virginia and Georgia, the 2028 target is the first thing that slips.
What to track next: how much of the $105 billion actually gets drawn once construction financing closes, and whether Nvidia extends similar backstops to the other hyperscale sites OpenAI has floated for Texas and New Mexico. If this becomes the template, Nvidia's balance sheet -- not OpenAI's revenue -- is the real constraint on how fast the AI buildout can scale.
