Nvidia will backstop up to $105 billion in financing for OpenAI's planned data center at the PORTS-Pike Technology Campus in Pike County, Ohio, disclosed in an August 17 SEC filing. SB Energy, the SoftBank-owned developer, will build, own, and operate the site under a 20-year lease to OpenAI, with Nvidia as the exclusive compute provider. The campus is planned at roughly 8 gigawatts, and the first phases come online in 2028. The number is enormous. The structure — a chip vendor guaranteeing the debt that buys its own chips — is what operators should study.
Key takeaways
- A guarantee, not a check. Nvidia's $105 billion backstop is tied to the residual value of the leased infrastructure, not a direct cash investment.
- Circular capital. The exclusive compute vendor is financing its biggest customer's buildout — and separately investing $1.5 billion in SB Energy.
- Capacity lands in 2028, not now. Ohio's first phases and OpenAI's Georgia campus both start delivering in 2028. Nothing here eases 2026–2027 constraints.
- Announced gigawatts are inflated. Wood Mackenzie projects roughly two-thirds of US data-center power requests will never be drawn from the grid.
- Plan accordingly. Assume constrained regional capacity and occasional rate limits near term. Weight signed utility contracts over press releases.
What was actually announced
The initial credit supports 4.25 GW of compute capacity, with an option on the remaining 3.75 GW of the planned 8 GW campus. SB Energy and SoftBank will build power sources supporting 10 GW and invest at least $4.2 billion in regional grid infrastructure. Nvidia is putting a separate $1.5 billion into SB Energy itself. OpenAI says the project will create 35,000 construction jobs over a six-year buildout and 2,500 long-term operating positions.
Ohio is the second multi-gigawatt OpenAI campus announced this summer. On July 22, the company unveiled a 1,400-acre campus in Effingham County, Georgia, with at least $20 billion committed initially and a total cost expected to exceed $30 billion. OpenAI secured 3.2 gigawatts from Georgia Power for that site, with phases running from 2028 through 2032 and several hundred megawatts expected available beginning in 2028. Add the two together and the buildout is real. The timing is the catch: none of it serves workloads before 2028.
Compute is becoming a financed asset
The most important detail is how the guarantee is structured. Nvidia is not writing a $105 billion check. It is standing behind the residual value of leased infrastructure — the way aircraft and rail fleets get financed. That is a maturity signal: lenders will now fund GPUs and data-center shells at scale if a credible party backs the residual.
But the credible party here is the chip supplier. Nvidia sells the chips, backstops the financing that buys them, and invests in the energy developer building the site. Each leg is defensible on its own. Together, they make the demand signal hard to read — some portion of this buildout is underwritten by the vendor whose revenue depends on it.
For an operator, that translates into counterparty risk. If you are making long-term commitments to a model provider, part of what you are relying on is a capacity pipeline financed by its own supplier. That is not a reason to walk away. It is a reason to keep workloads portable and contract terms short enough to adjust.
When the chip vendor guarantees the loans that buy its own chips, the demand signal you are reading is partly a mirror.
The phantom-load caution
Five days before the Ohio announcement, a Wood Mackenzie analysis — first reported by Bloomberg on August 12 — put the broader buildout in perspective. Developers have been submitting duplicative power requests to multiple utilities at once, with no meaningful upfront financial commitment, and planners have treated those inflated interconnection queues as real load.
The Ohio project looks more real than most: a disclosed financing structure, a named builder on a 20-year lease, and committed grid investment. But the base rate matters. When two-thirds of announced load never materializes, headline gigawatts are a poor signal of where capacity will actually exist. Discount every announcement until it comes with signed utility contracts and interconnection agreements.
What this changes for your planning
The operator read is straightforward: the capacity wave is coming, it is financed in a circle, and it arrives in 2028 at the earliest. Four moves follow.
- Assume constraint through 2027. Design systems that degrade gracefully under rate limits: request queueing, response caching, and fallbacks across providers or model tiers.
- Discount press-release pipeline. When a provider's roadmap leans on announced capacity — or when siting your own infrastructure — weight signed interconnection agreements and utility contracts, not project announcements.
- Price concentration risk. Vendor-financed megaprojects add counterparty and ecosystem-leverage risk to long-term provider commitments. Keep workloads portable so a financing hiccup does not strand your roadmap.
- Watch for 2028 leverage. If the gigawatts land, supply loosens and pricing leverage shifts toward buyers. Keep contract terms short enough to renegotiate when it does.
This is a summer of compute-market shifts: AMD and Anthropic's 2 GW deal widened the supplier field in July, and inference prices hit a 2026 low this month even amid constrained capacity. If you want help pressure-testing your vendor strategy against these moves, our AI strategy advisory practice does exactly that — book a consult.