On August 17, OpenAI, Nvidia and SB Energy — the SoftBank-backed energy and infrastructure developer — announced one of the largest AI compute projects ever planned: an up-to-8GW campus at the PORTS-Pike Technology Campus in Pike County, Ohio. Read past the "another giant data center" headline and the deal structure is genuinely different. Nvidia is investing $1.5 billion in SB Energy, becoming its exclusive AI compute infrastructure provider for roughly 8GW of capacity, and — the number that matters most — signing residual value guarantees with a total payment obligation capped at $105 billion. That is the largest financial commitment a chipmaker has ever put behind a single AI project.
What makes this deal worth studying is not "who built the biggest AI factory." It is the first large-scale execution of the compute-asset narrative we analyzed last month: when Nvidia promised a 25% residual-value backstop for AI projects, the industry entered the financialization phase of AI capex. Ohio is that logic applied to a real contract. Nvidia is no longer just selling chips to a customer. It is guaranteeing the lease of the building the customer rents, owning a stake in the landlord, and supplying every piece of compute inside — at the same time.
A $1.5 Billion Investment Behind a $105 Billion Guarantee
Break the structure down. The site is the former Portsmouth Gaseous Diffusion Plant — a Cold War uranium-enrichment complex that powered the US nuclear weapons program from 1954 until it shut down in 2001, with decontamination still ongoing since 2011. The US Department of Energy put it on a list of 16 federal sites designated for rapid data-center construction in 2025, and announced a public-private redevelopment in March 2026.
The three players split the work like this:
- OpenAI is the tenant. It signed a 20-year lease with SB Energy for approximately 8GW (IT load) of capacity, which will become a core source of compute for frontier training and product inference once fully delivered. OpenAI pays only as completed capacity becomes available, and added $40 million to SB Energy's existing $40 million community benefits fund.
- SB Energy is the developer and operator. The SoftBank-backed company (founded 2019; SoftBank and OpenAI are both shareholders) will build, own and operate the campus and lease it to OpenAI under the 20-year agreement. It is also planning at least 10GW of new power generation and $4.2 billion in Ohio grid infrastructure investment with SoftBank.
- Nvidia is exclusive compute provider, investor and guarantor. It is investing $1.5 billion in SB Energy equity, providing credit support for the land, power and shell buildout of the initial 4.25GW, holding an option on the remaining 3.75GW — and the campus will exclusively run Nvidia's full-stack DSX architecture (GPUs, CPUs, networking, software).
The financial layer is where the deal gets serious. Nvidia signed multiple residual value guarantees with SB Energy with an aggregate payment obligation capped at $105 billion, subject to conditions including the lessor meeting ready-for-service requirements (expected from 2028). Reports from WSJ and others say Nvidia had discussed guarantees as high as $250 billion, plus up to $350 billion in financing for OpenAI chip purchases; the $105 billion cap is the converged number after negotiation.
In Nvidia's own framing, each generational rollout at PORTS-Pike involves roughly 1.5 million GPUs (about $150-200 billion), and the pipeline could represent about $600 billion of Nvidia compute revenue by 2030.
From Chip Seller to Lease Guarantor: Nvidia's Role Shift
To see why this matters, rewind one month. On August 11, Nvidia joined Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create an independent financing platform targeting $500 billion+ in third-party capital for AI infrastructure. Four days later, Jensen Huang explained the 25% residual-value backstop: the core argument was that GPUs are long-lived, revenue-producing assets — not consumables.
Ohio is the first real test of that argument. Compare the two positions: the 25% backstop was a mechanism framework — Nvidia commits to absorbing part of the residual-value risk on individual projects, with financial institutions independently evaluating each one. Ohio is a project execution — Nvidia takes on up to $105 billion in guarantee obligations for a single campus, roughly a fifth of the project's estimated $500 billion+ total cost.
The role change matters more than the dollar amount. In the traditional model, Nvidia is an upstream equipment vendor: customers (clouds, Neoclouds) finance chip purchases on their own balance sheets. In the Ohio model, Nvidia appears in four layers of the same deal — earning hardware revenue as the equipment maker, sharing long-term value as an equity holder in SB Energy, absorbing residual-value risk as guarantor, and lowering the project's cost of capital as the credit provider.
Nvidia's business model is migrating from "selling shovels" to "shovels, mine, insurance and bank in one." It can no longer just care about how many GPUs ship; it now cares whether the customer can get cheap power, whether the project can raise financing, and who absorbs the loss when assets age out — because every layer of risk eventually points at the same balance sheet.
Circular Financing: When the Money Circles Back to the Chip Seller
The most contested part of this structure is circular financing: the company Nvidia invests in and guarantees turns around and buys Nvidia's chips. Nvidia's credit backing lowers financing costs, SB Energy builds faster, the campus comes online sooner, GPU purchases happen earlier — and Nvidia's money flows back to Nvidia as revenue.
The logic is commercially rational. Nvidia carries roughly $3 trillion in market cap and investment-grade credit; using it to lower the cost of capital across the AI infrastructure chain pulls the demand curve forward. Analyst Rohan Paul put it precisely: "If Nvidia's backing lowers SB Energy's financing cost, capacity can be built sooner and GPU purchases can arrive sooner too — but some risk that normally belongs to the borrower or tenant migrates onto Nvidia through the guarantee."
Trace the risk paths:
- Customer credit risk. If AI demand disappoints, OpenAI's ability to pay lease obligations weakens, guarantees could be triggered, and Nvidia would have to pay SB Energy.
- Asset obsolescence risk. GPUs cycle in 3-5 years; the lease runs 20. If a new architecture devalues the installed GPU fleet faster, residual-value guarantee payouts grow as the assets depreciate.
- Systemic risk. If project bonds backed by Nvidia guarantees get priced by markets as quasi-safe assets, risk can propagate through the financial system — the MBS concern of 2008 replayed on AI infrastructure.
Nvidia's position: guarantees are evaluated project by project, capped, and conditional on ready-for-service — not an open-ended backstop. The critics' sharper point stands: independent capital is not independent demand. Financial engineering can extend the construction cycle, but it cannot create downstream demand. When the strongest player in an industry needs its own balance sheet to sustain the expansion narrative, the capital cycle has moved from the technology-dividend phase to the financial-engineering phase.
A Cold War Uranium Plant Becomes an AI Powerhouse: Energy Is the Real Bottleneck
The deal also exposes a fact the AI arms race keeps understating: the real constraint is no longer chips. It is power.
PORTS-Pike sits on the former Portsmouth Gaseous Diffusion Plant — one of America's three Cold War uranium-enrichment sites, which produced enriched uranium for nuclear weapons and Navy submarines. The location is not an accident: DOE controls the land, grid connections exist, regional capacity is ample, and the federal government designated 16 sites for rapid data-center construction in 2025. SB Energy and SoftBank plan at least 10GW of new generation (including 9.2GW of natural gas, backed by roughly $33 billion in Japanese financing) plus $4.2 billion in grid investment.
Line it up and the constraint hierarchy is clear: chips can be solved by fab capacity, land can be solved by federal sites, but power needs years of generation construction and billions in grid investment. Whoever locks down "land with electricity" holds the entry ticket to the next round. This is why pure-tech companies like Nvidia and OpenAI are moving deep into energy — after Ohio, energy assets are effectively on the strategic balance sheets of AI giants, the supply-side twin of the $2 trillion in off-balance-sheet AI debt we mapped earlier.
What This Means for the Industry
Whether this ends as a great financial innovation or a sophisticated Ponzi structure, the deal is already reshaping the ecosystem:
- For AI clouds and Neoclouds: the financing paradigm changed. They no longer have to borrow on their own credit ratings; top players can now lock in long-term capacity at lower cost through a standardized tenant-developer-vendor structure with Nvidia backing. SB Energy is targeting an IPO as soon as September 2026 (seeking at least $5 billion) — if it succeeds, it validates the "compute landlord" as a standalone, market-priced asset class.
- For institutional investors: compute is entering asset allocation for the first time as guaranteed long-term lease cash flows. The $105 billion cap means AI infrastructure debt quality is now tied to a chipmaker's balance sheet — more investable, but also more concentrated risk.
- For the industry structure: after Ohio, compute supply is shifting from "clouds building their own data centers" to a triangular structure of specialized developers building campuses, model companies signing long leases, and chipmakers guaranteeing the risk. Whoever holds chips, capital and power simultaneously will have pricing power on the 2030 compute map.
For anyone tracking AI infrastructure, the signals to watch are not GPU shipments. Watch three things: the size and structure of Nvidia's guarantee portfolio as projects are disclosed; whether OpenAI's lease actually ramps (the first 800MW is targeted for 2028); and whether the energy buildout — 10GW of generation and the grid investment — clears financing and permitting on schedule. Those three signals tell you more about where the AI capital cycle stands than any model leaderboard.
