Nvidia is in advanced talks to provide a roughly $250 billion financial guarantee that would let OpenAI lease a 10-gigawatt data center campus in southern Ohio — and the chipmaker is separately discussing financing up to $350 billion of OpenAI's Nvidia chip purchases for the same site. If it closes, total project cost — the facility plus the GPUs inside it — would top $500 billion, making it the largest single data center project ever announced. The deal is not final and could fall apart, but the structure reveals a new financing playbook that directly affects anyone building on top of OpenAI, Nvidia, or hyperscale compute over the next four years.
TL;DR — Last verified: 2026-07-30
- $250B lease/debt backstop from Nvidia -> helps OpenAI lease the SB Energy (SoftBank) campus on DOE land in Piketon, Ohio.
- Up to $350B in chip financing — separately negotiated; not part of the $250B guarantee.
- 10 GW total IT load, first phase ~800 MW online in 2028; power from a $33B Japan-funded natural gas plant controlled by the U.S. government.
- Why Nvidia is doing it: lock in multi-year chip demand without spending its own capital upfront; counter chatter it was scaling back OpenAI support.
- Why OpenAI is doing it: first directly owned/leased infrastructure, reducing reliance on Microsoft, AWS, and Oracle rented compute.
- Risk flag: OpenAI is valued at ~$500B, still unprofitable, with projected compute spend of ~$750B through 2030 — the guarantees only pay off if that revenue materializes.
Why is Nvidia guaranteeing $250 billion for OpenAI?
Nvidia's guarantee is effectively a credit wrapper — a financial structure in which an investment-grade company (Nvidia, ~$5 trillion market cap) lends its balance-sheet credibility to a riskier borrower (OpenAI, unprofitable, no investment-grade rating) so that lenders extend debt at more favorable terms. Nvidia does not hand over $250 billion in cash; it commits to step in if the financing vehicles default, which is what lets SoftBank's SB Energy raise the construction and lease debt cheaply enough for the project to pencil out.
This matters because it explains who actually bears the risk and why the deal can be this large. OpenAI cannot borrow tens of billions on its own signature — it has no public credit rating and burns more cash than it earns. Nvidia, by contrast, can support the financing without the same upfront capital outlay, while securing something it cannot buy on the open market: guaranteed, years-long demand for its most expensive chips. Reuters and the Wall Street Journal both report the $250 billion figure covers the data center lease and debt financing specifically, not the chips themselves (WSJ via Reuters; WSJ original).
What is the Ohio data center, and who is building it?
The campus is a 10-gigawatt AI data center on federal land at the former Portsmouth Gaseous Diffusion Plant in Piketon, Ohio (Pike County) — a decommissioned Cold War uranium-enrichment site about 50 miles south of Columbus. The developer is SB Energy, a SoftBank Group subsidiary, leasing the DOE land under a public-private partnership announced in March 2026 with the U.S. Department of Energy, the Department of Commerce, and AEP Ohio (DOE/AP News; Data Center Dynamics).
For context, 10 GW is roughly the output of ten commercial nuclear reactors and enough electricity to power several million homes. Building on federal land is a deliberate workaround: it sidesteps the local permitting battles and ratepayer-subsidy fights that have snarled data center approvals across the country. SB Energy is the same company already developing OpenAI's 1.2 GW Stargate facility in Milam County, Texas, so there is an existing operating relationship (DCD).
How is the timeline and first phase structured?
- Groundbreaking: March 2026, with Commerce Secretary Howard Lutnick, Energy Secretary Chris Wright, and SoftBank CEO Masayoshi Son on site (Statehouse News Bureau).
- Phase 1: ~800 MW, targeted for 2028.
- Full build-out: 10 GW across multiple phases, several years to complete.
- Power source: a 9.2 GW natural-gas generation complex built and operated by SB Energy on the federal land.
The first phase is the part builders should actually track. 800 MW in 2028 is a meaningful but incremental addition to U.S. AI compute capacity — not a step-change. The full 10 GW is the headline number, but it depends on financing, equipment delivery (Nvidia GB200/GB300-class systems and successors), and the gas plant coming online in step. Treat the 2028 first-phase date as a planning anchor, not a guarantee.
Where does the power come from, and who controls it?
This is the most unusual structural feature of the deal. The electricity is not a standard utility interconnection. It is controlled by the U.S. government and funded separately through a U.S.–Japan trade arrangement: Japan pledged roughly $33 billion to build the 9.2 GW natural gas plant, in exchange for lower Japanese import tariffs (Scioto Post; WSJ via Live Mint).
U.S. Commerce Secretary Howard Lutnick is personally involved in deciding who gets access to that power. According to the WSJ reporting, OpenAI has shown the strongest interest, but Anthropic, Microsoft, and Google have all spoken with Lutnick in recent weeks about the site. That means power allocation is a political and commercial bottleneck, not a market transaction — and it means the compute capacity from this site isn't guaranteed to flow to OpenAI until the lease closes.
The gas plant itself will be owned by the U.S. government; SB Energy receives an operating fee, and Japan and the U.S. split power sales revenue until Japan recoups its $33 billion, after which the U.S. government takes roughly 90% of revenue (WSJ via Live Mint). This is a sovereign-backed infrastructure project dressed up as a private data center.
What is the separate $350 billion chip-financing deal?
Beyond the $250 billion lease/debt backstop, Nvidia and OpenAI are separately discussing Nvidia financing OpenAI's chip purchases for the site, in a deal reported to be worth up to $350 billion on its own (NYT; TradingView/WSJ reporting). Nvidia previously invested $30 billion directly in OpenAI, so this would deepen an already large financial entanglement.
This is the part that should make builders pay attention, because it is the circular-funding dynamic the WSJ flagged as a systemic risk: Nvidia is financing the building, financing the chips going into the building, and is itself a shareholder of the company consuming both. If OpenAI's revenue growth stalls, Nvidia is exposed on at least three layers of the same bet.
How does this fit OpenAI's $750 billion compute-spending plan?
OpenAI has raised its projected compute spending to roughly $750 billion through 2030, up from ~$600 billion projected earlier in 2026 (WSJ; Crypto Briefing). The Ohio deal isn't necessarily a clean addition to that number — it overlaps with OpenAI's broader shift from renting cloud capacity (Microsoft Azure, AWS, Oracle, CoreWeave) toward owning and leasing its own facilities.
Put alongside OpenAI's separately announced Project Camellia — a ~3.2 GW data center in Effingham County, Georgia, with a $20 billion initial commitment and a Georgia Power contract running 2028–2032 — the strategy is clear: OpenAI wants to control the infrastructure, not just rent it. The Ohio campus would be the flagship of that shift if it closes.
For builders, the practical implication is that OpenAI's capacity roadmap is now a mix of rented cloud and directly controlled data centers, with the后者 coming online mostly from 2028 onward. Short-term capacity pricing and availability still depends on the hyperscalers; the long-term picture depends on whether these megaprojects land on schedule.
OpenAI valuation and the risk question: is $250B sustainable?
OpenAI was most recently valued at roughly $500 billion as of October 2025, after a secondary share sale that let employees sell ~$6.6 billion in equity — making it the world's most valuable private company, ahead of SpaceX (Reuters; NYT; Fortune).
That valuation is the load-bearing fact for the whole Ohio deal. Financing guarantees of this scale only work if OpenAI's future revenue grows into them. The company remains unprofitable, and per reporting summarized by financial analysts, OpenAI's annual cash burn is projected to climb steeply — from roughly $17 billion in 2026 toward an estimated $47 billion by 2028 (Wiss, summarizing The Information). If those numbers hold, the Ohio bet is a growth story that needs roughly $50–80 billion in fresh capital over the next few years simply to keep the operating model intact, before any of the megaproject capacity comes online.
This is the open question worth tracking, and one no party to the deal can answer today: Nvidia's willingness to backstop a quarter-trillion dollars is either supreme confidence in AI demand, or a sign that even the world's most valuable chipmaker now has to underwrite its own customers to keep the growth story intact. The truth is probably both, and which one proves load-bearing depends entirely on whether OpenAI's revenue curves match its compute curves through 2028 and beyond.
What this means for you
- If you build on OpenAI APIs today: the rented-cloud capacity model still dominates near-term availability and pricing. Treat the Ohio 2028 first-phase milestone as a directional indicator, not a forecast you can plan contracts around.
- If you build on or compete with Nvidia: the chip-financing layer means Nvidia is increasingly a financial intermediary for its own demand, not just a hardware vendor. Watch its quarterly cash-flow warnings about data-center funding exposure (WSJ via Live Mint).
- If you run AI infrastructure or a small business evaluating AI tooling: the megaproject-tier headlines won't change your per-token costs in the next 12–18 months. What will change them is the broader supply-demand balance these projects are betting on. If the bets pay off, more capacity and price pressure downward; if they slip or stall, a tighter market and upward pressure. Either way, lock multi-year contracts where you can.
- If you're tracking AI investment risk: the circular-funding dynamic (Nvidia financing Nvidia buyers) is the structural risk to watch. It works on the way up and compounds on the way down.
For related analysis on the broader AI-investment and compute-spending picture, see our breakdown of Meta's free-cash-flow crash and what AI capex means for builders, our decision framework for open-weight vs. closed AI models after the Nvidia-Mistral shift, and our guide to why most enterprise AI projects never scale (and the 3-pillar fix).
FAQ
Q: How large is the Nvidia–OpenAI Ohio deal? A: Nvidia is in talks to guarantee roughly $250 billion in lease and debt financing for OpenAI to use a 10-gigawatt data center SB Energy (SoftBank) is developing on federal land in Piketon, Ohio. Separately, Nvidia is discussing financing up to $350 billion of OpenAI's chip purchases for the same site. Total project cost, including chips, could exceed $500 billion. The deal is not final and could fall apart.
Q: When will the Ohio data center open? A: The first phase — roughly 800 megawatts — is targeted for completion in 2028. Full build-out to 10 GW will take several years beyond that. Ground was broken in March 2026.
Q: Who controls the power for the project? A: The U.S. government controls power allocation. Japan pledged about $33 billion to build a 9.2 GW natural-gas plant on the federal land under a U.S.–Japan trade deal. Commerce Secretary Howard Lutnick is personally involved in deciding which companies — OpenAI, Anthropic, Microsoft, or Google — get access to the power.
Q: Why is Nvidia financing OpenAI rather than just selling chips to them? A: OpenAI is unprofitable and lacks an investment-grade credit rating, so it cannot raise tens of billions in debt on its own. By providing a "credit wrapper" guarantee, Nvidia lets SoftBank raise the construction financing at favorable terms — and in exchange Nvidia locks in multi-year demand for its chips without spending its own capital upfront. Nvidia previously invested $30 billion directly in OpenAI.
Q: Is the $250 billion a cash payment from Nvidia? A: No. It is a financing guarantee (backstop). Nvidia commits to step in if the debt/lease financing vehicles default, which is what lets lenders offer favorable terms. The chip-financing discussion (~$350 billion) is a separate, additional negotiation.
Q: What happens if OpenAI's revenue doesn't grow fast enough? A: This is the core risk: Nvidia would be exposed across at least three layers of the same bet (shareholder, financing guarantor, chip financier), and the underlying debt depends on OpenAI's revenue materializing. The WSJ explicitly flagged the "circular funding" pattern as a systemic risk if AI demand slows. The deal's feasibility hinges on whether OpenAI's compute-spend trajectory translates into matching revenue.
Discussion
0 comments