Published August 18, 2026 — San Francisco. OpenAI locked a 20-year lease on August 17, 2026, for a 10-gigawatt data center on federal land in Ohio, with NVIDIA providing a 20-year backstop guarantee and SoftBank's SB Energy providing project financing.
Data last verified September 2026 from the Wall Street Journal and OpenAI press releases.
At a glance
- OpenAI locked a 20-year lease on August 17 2026 for a 10-gigawatt data center on federal land in Ohio.
- The project is part of the 500 billion dollar Stargate program; full 10GW capacity is targeted for 2030.
- NVIDIA is providing a 20-year backstop guarantee and supplying the next-generation Vera Rubin GPU platform.
- SoftBank's SB Energy is providing project financing in exchange for long-term power purchase agreements.
- The deal is expected to create 10,000 construction jobs and 1,000 permanent operations jobs in Ohio.
OpenAI's August 17 announcement of a 10-gigawatt data center on federal land in Ohio is the largest single AI infrastructure commitment of 2026, larger than Microsoft's 80 billion, Meta's 60 to 65 billion, or Amazon AWS's 90 billion dollar AI infrastructure spend for the year. The deal combines three unusual ingredients: a 20-year NVIDIA backstop guarantee, SoftBank's SB Energy as project financier, and federal land allocated through the Department of Energy.
This guide covers the project details, why 10GW is unprecedented, NVIDIA's role and the backstop mechanics, SoftBank SB Energy's financing logic, the economic and environmental impact on Ohio, and how the deal fits the broader AI infrastructure race.
OpenAI Ohio data center project details
| Field | Detail |
|---|---|
| Capacity | 10 gigawatts (10 GW) |
| Location | Federal land in Ohio (Department of Energy) |
| Total cost | Approximately 500 billion US dollars (20-year lease term) |
| Financing | OpenAI equity, NVIDIA backstop, SoftBank SB Energy debt |
| GPU platform | NVIDIA Vera Rubin (next-generation) |
| Power equivalent | 10 GW equals the power for 7.5 million US homes |
| Groundbreaking | Q4 2026 (expected) |
| Initial operations | Q4 2027 (first 1 GW) |
| Full capacity | 2030 (full 10 GW) |
| Jobs | 10,000 construction, 1,000 operations |
Source: Wall Street Journal, OpenAI, NVIDIA, and SoftBank press releases of August 17 2026.
Why 10 gigawatts is unprecedented
10 gigawatts is approximately the total power consumption of Manhattan, more than the entire state of Ohio consumes during peak hours, and equivalent to the power for 7.5 million US homes. The OpenAI Ohio site is five to ten times the size of the largest existing AI data center, whether Microsoft's 800MW Quincy WA campus or Meta's 1.5GW Eagle Mountain site. The 10GW of GPU compute represents approximately 10 million H100-equivalent GPUs running at full load, per OpenAI's 2026 infrastructure disclosure.
For comparison, the 10GW project is comparable in scale to the Apollo program in total energy. The capacity is enough to train a frontier model of the next generation in a single training run, and to serve hundreds of millions of inference requests per day, per the OpenAI infrastructure team.
The compute math behind the 10GW figure is what makes the deal strategic. A single NVIDIA Vera Rubin rack draws roughly 50 to 60 kilowatts at full training load. At 10GW, the Ohio site can host approximately 170,000 to 200,000 GPU racks, or 8 to 10 million GPU units if the racks are sized closer to the lower density bound. That is enough compute to train a next-generation foundation model end-to-end, run a multi-trillion-parameter mixture-of-experts in production, and serve inference at a population scale, all from a single physical site.
The cooling and power-delivery engineering required at this scale is itself a constraint. 10GW of continuous draw cannot be served by air cooling; the site requires liquid cooling loops, substation-level grid interconnects, and on-site natural gas turbines for peaking power. The total water consumption for direct liquid cooling at this density is estimated at 1.5 to 2 billion gallons per year, per the Lawrence Berkeley National Laboratory 2026 data center report. OpenAI's commitment to 100 percent renewable energy offsets the operating emissions but does not address the water and land footprint.
What the Stargate program is and where the Ohio site fits
Stargate is OpenAI's 500 billion dollar infrastructure program announced in January 2025 in partnership with SoftBank, Oracle, and MGX. The program aims to build 10 gigawatts of AI data center capacity over four years. The Ohio project is the largest Stargate site to date, and other Stargate projects are in Texas, New Mexico, and the UAE. OpenAI needs this capacity to train and serve next-generation AI models, and the geographic spread reduces regulatory and grid-risk concentration in any single US state.
The Stargate structure is unique in that the financial backstops and equity sponsors are split across multiple counterparties. OpenAI provides the demand and the operational expertise, SoftBank provides the project-level debt and renewable energy infrastructure through SB Energy, NVIDIA provides the GPU supply and a 20-year backstop on the lease payments, and Oracle and MGX provide additional capital and operational support. The structure is designed to keep OpenAI's balance sheet from absorbing the full capex of a project that exceeds the GDP of most US states.
NVIDIA role and the backstop guarantee
NVIDIA is providing a 20-year backstop guarantee for the OpenAI Ohio data center. The backstop means NVIDIA guarantees OpenAI's lease payments to SoftBank and the project financiers. If OpenAI cannot make the payments, NVIDIA steps in. The structure is needed because the 500 billion dollar project is too large for OpenAI to finance on its own, and SoftBank's SB Energy requires a creditworthy counterparty for the long-term power purchase agreements.
NVIDIA's stock fell 5 percent on the initial Wall Street Journal report in August 2026, then recovered when the deal was finalized on August 17. The motivation for NVIDIA is to lock in massive GPU demand for the Vera Rubin platform. The deal also signals that NVIDIA sees the AI infrastructure cycle as a multi-decade capital commitment, not a quarterly GPU sales story.
Why SoftBank SB Energy is financing the deal
SoftBank's SB Energy is a renewable energy infrastructure company providing project financing for the Ohio data center in exchange for long-term power purchase agreements. SB Energy will also build the on-site solar and storage to power the data center. SoftBank has been a key partner in OpenAI's Stargate program since January 2025. The deal creates a massive anchor customer for SoftBank's renewable energy business and ties the financing to a 20-year revenue stream backed by OpenAI and NVIDIA.
Economic impact on Ohio
The Ohio data center is expected to create 10,000 construction jobs during the four-year buildout and 1,000 permanent operations jobs, per the Ohio Department of Development. The project is also expected to generate 2 to 3 billion US dollars in annual economic impact, new tax revenue for the state, and new infrastructure improvements for roads, water, and power. Ohio offered tax incentives worth an estimated 1.5 billion dollars to secure the project. Other states, including Texas, Virginia, and Arizona, competed aggressively for the project.
Environmental and community concerns
The 10GW data center has raised five environmental and community concerns. First, power grid impact: 10GW is more than the entire state of Ohio consumes during peak hours, requiring substantial grid upgrades. Second, water use: data center cooling requires millions of gallons of water, straining local supplies. Third, land use: more than 1,000 acres of federal land will be developed. Fourth, noise: data center fans and cooling towers create constant low-frequency noise. Fifth, light pollution: 24/7 operations create persistent light spillover.
OpenAI has committed to 100 percent renewable energy for the project and a water-efficient cooling design, per OpenAI's 2026 sustainability disclosure. The company is also working with the Department of Energy on grid upgrade coordination and with local utilities on-site solar and battery storage to reduce transmission load.
How the deal fits the broader AI infrastructure race
- OpenAI Stargate: 500 billion dollars, 10GW across multiple sites.
- NVIDIA: 20-year backstop on OpenAI, 100 billion dollar investment in OpenAI.
- Microsoft: 80 billion dollar AI infrastructure spend in 2025, owns the OpenAI Azure relationship.
- Meta: 60 to 65 billion dollar AI infrastructure spend in 2025, building 1.5GW+ data centers.
- Amazon AWS: 90 billion dollar AI infrastructure spend, Trainium and NVIDIA chips.
- Google: 75 billion dollar AI infrastructure spend, TPU and NVIDIA chips.
- SoftBank: 500 billion dollar Stargate investment, SB Energy infrastructure financing.
What this means for the AI industry in 2026
The OpenAI Ohio data center signals continued massive AI infrastructure investment despite recent market uncertainty. The scale of the project, 500 billion dollars over 20 years, is unprecedented in the technology industry. The deal also signals OpenAI's commitment to remaining independent of any single cloud provider, NVIDIA's strategic investment in OpenAI to lock in long-term GPU demand, the continuing arms race between AI labs for compute, the increasing role of sovereign and corporate financing for AI infrastructure, and the growing geopolitical importance of AI compute as a strategic resource.
Resources and next steps
- OpenAI press: openai.com/news
- OpenAI infrastructure deep-dive: openai.com/index/openai-infrastructure
- NVIDIA investor relations: nvidianews.nvidia.com
- SoftBank SB Energy: softbank.jp/en/sbenergy
- Wall Street Journal AI infrastructure coverage: wsj.com
Read next
The safety framework the AI labs are coordinating on is OpenAI Anthropic Google Tri-Lab Safety Talks 2026 Outcomes. The chip-supply side this infrastructure will sit on top of is AI chip war 2026 NVIDIA AMD TPU compared. The grid-side implications are covered in AI data center electricity demand 2026, and the investment-cycle framing is in AI factories token economy Jensen Huang 2026.






