The future OpenAI data center in Ohio rests on a 20-year lease at the PORTS-Pike campus, an 8-gigawatt site built on a former uranium enrichment plant. Nvidia will guarantee up to $105 billion in lease and power obligations, an unprecedented financing structure disclosed in an SEC filing. The build promises 35,000 construction jobs and 2,500 permanent positions by 2032.
Key Takeaways
- A 20-year lease covering 8 gigawatts of computing capacity, the largest infrastructure commitment OpenAI has ever signed.
- Nvidia guarantees up to $105 billion in conditional obligations and becomes the site’s exclusive chip supplier.
- 35,000 construction jobs over six years, 2,500 permanent roles, and a first 800-megawatt phase targeted for 2028.
A 20-Year Lease on a Former Nuclear Site
The site chosen for the OpenAI data center is anything but ordinary. The PORTS-Pike campus sits on the grounds of the former Portsmouth Gaseous Diffusion Plant in Piketon, Pike County, which enriched uranium for the American nuclear program from 1954 to 2001. Part of the land is federal, which explains the Department of Energy’s seat in the deal.
OpenAI signed a 20-year lease there, announced Monday August 17, securing 8 gigawatts of IT capacity powered by 10 gigawatts of new energy generation. The company walked through the partners and local commitments in its official announcement of the PORTS-Pike project.
The scale dwarfs everything Ohio has hosted so far. The state had already become a compute stronghold with the tent data centers Meta assembles there to outrun its own building permits. PORTS-Pike plays in another league, with a first 800-megawatt phase targeted for 2028.
Construction and operations fall to SB Energy, the energy subsidiary of SoftBank Group. The schedule spreads the build over six years, through 2032, for a site designed as a full campus rather than a single building. Picking a rehabilitated nuclear site brings land, historic grid connections and a regulatory environment already mapped out.
The symbolism is lost on nobody in the region. Pike County lived half a century on the rhythm of the atom, with the paychecks and closures that came with it, before two decades of decommissioning. Watching a giant employer return to the same ground replays an industrial story the Ohio valley knows by heart, with AI cast in the role uranium once held.
The lease format matters as much as the size. OpenAI pays its own energy and infrastructure costs, including the transmission upgrades the site requires, rather than folding them into a cloud contract. That structure keeps the capacity fully dedicated, off any shared platform, and priced against long-term power agreements instead of spot cloud rates that fluctuate with every capacity crunch.
The closest precedent in size remains the giant lease signed on Anthropic’s side. The Riot Platforms deal with Anthropic had already shown that frontier labs now prefer renting turnkey gigawatts over building alone. OpenAI pushes the logic one step further by adding a guarantor.
The Nvidia Guarantee That Unlocks the Financing
The financial heart of the OpenAI data center sits in one number disclosed in an SEC filing. Nvidia commits to guaranteeing up to $105 billion in conditional lease and power payment obligations owed to SB Energy. If OpenAI fails to pay, the chipmaker pays in its place.
The guarantee covers a first phase of 4.25 gigawatts of computing capacity, with an option on 3.75 additional gigawatts. In parallel, Nvidia invests $1.5 billion directly into SB Energy and becomes the site’s exclusive chip supplier. The same player finances, equips and insures the project.
The structure answers a very concrete problem. No bank comfortably lends tens of billions to a client whose commitments grow faster than its revenue, and the traditional bond market is already saturating. The $20 billion bond sale Nvidia launched had already signaled that the chipmaker was taking on the role of the sector’s banker.
The method recalls the project finance of the telecom era, with one major difference. The guarantor is also the vendor of the hardware that will fill the site, which loops the economics back onto itself. Nvidia locks in years of GPU orders in exchange for the credit risk, turning its balance sheet into a commercial weapon.
The credit mechanics also reshuffle who carries the sector’s risk. Guarantees of this size sit off the beaten path of bank syndicates and rating agencies, concentrated instead on one chipmaker’s balance sheet. If AI demand ever stalls, the losses would surface not at a lender but at the very company whose GPUs power the boom, tying the industry’s fate into a single knot.
For competitors, the bill turns dizzying. Google is already mobilizing its balance sheet and its partners’, as Alphabet’s $80 billion infrastructure push shows, while Anthropic rents at scale. Players without an industrial patron or a giant balance sheet are now out of the gigawatt race.
More articles on Horizon
- Anthropic Revenue Hits a $65 Billion Annual Pace
- Dario Amodei Calls the AI Backlash a Trust Crisis
- Anthropic Left Its Bioweapon Filter Off for 11 Months
35,000 Construction Jobs and a Record Energy Bill
The employment package attached to the OpenAI data center sets the scale of the build. The project promises 35,000 construction-related jobs over the six-year build and 2,500 permanent operating positions once the campus runs. That ratio captures the economics of data centers, capital-intensive and light on durable employment.
OpenAI wrapped the announcement in a calibrated local package. A $40 million community fund, financed with SB Energy, will support priorities defined by residents, and $84 million in Codex credits will open the tool to every college student in Ohio. The sequence targets local officials as much as it targets national public opinion.
The political context explains the caution. Data centers have become a campaign issue in the American midterms, driven by worries over electricity bills and water use. The site claims closed-loop, air-cooled systems with no evaporation towers, water consumption comparable to an office building, and public reporting once operations start.
The carbon question remains wide open. The 10 gigawatts of new generation will lean overwhelmingly on natural gas, a choice dictated by grid timelines rather than climate. The emissions record set by Amazon’s Texas data center showed what these trade-offs produce once the turbines spin.
Comparing orders of magnitude frames the ambition. Eight gigawatts amounts to the output of several nuclear reactors dedicated to a single customer, on a single site. No cloud operator has ever concentrated that much capacity in one place, and the staged rollout through 2032 leaves room for successive chip generations inside the same halls.
For developers and API customers, this site is the 2028 roadmap in concrete form. A dedicated 8-gigawatt capacity changes the ceiling on trainable models and the availability of production agents, both throttled today by compute scarcity across every provider on the market. Current queues and quotas are the direct consequence of infrastructure that does not exist yet.
The sheer size of the structure carries its own fragility. A single guarantor concentrated on one sector, a tenant whose growth is slowing and a six-year build add up to a colossal bet on AI demand through 2032. Nobody knows what the market will look like by then, but the concrete will be poured, and the rent runs until 2046.
Follow the story on Horizon.
