Nvidia Cuts OpenAI Ohio Guarantee After Wall Street Platforms Launch

Nvidia is expected to guarantee less than $120 billion for the first phase of OpenAI’s planned Ohio data-center campus, down from the $250 billion figure that had been under discussion, according to a Wall Street Journal report citing people familiar with the talks. The chipmaker and OpenAI are nearing an agreement that could be signed as early as this weekend, with the backstop limited to the initial slice of a 10-gigawatt project.

The revision landed days after Nvidia announced partnerships designed to pull more than half a trillion dollars of outside money into AI infrastructure. That timing is not coincidental. It shows how the chipmaker wants headline deals to follow the new capital stack rather than precede it.

Nvidia Trims the Backstop to Under 120 Billion

The Journal reported that Nvidia will provide a financial backstop only for the first phase rather than the full multi-hundred-billion-dollar build-out. OpenAI is still negotiating a binding lease covering the entire 10-gigawatt campus in southern Ohio. The site is being developed by SB Energy, a SoftBank subsidiary, and would rank as the largest data-center project announced to date if finished.

Investors had raised concerns about Nvidia’s risk exposure from large financing commitments. Nvidia did not immediately respond to comment requests outside business hours. OpenAI declined to comment.

• Prior figure discussed: roughly $250 billion guarantee
• Revised initial backstop: less than $120 billion for phase one
• Project scale: 10 GW campus, total cost potentially exceeding $500 billion including chips
• OpenAI valuation context: $852 billion post-money after its latest round, still unprofitable

The change does not cancel the project. It changes who stands behind the early paper.

A smaller first-phase backstop still leaves Nvidia attached to the campus. It simply matches the guarantee to the slice that must close first, rather than to the full multi-year build.

Six Firms Now Underwrite the Asset Class

On August 10 Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent compute financing platforms. The goal is to mobilize over $500 billion of third-party capital for AI infrastructure over time.

Jensen Huang, Nvidia founder and CEO, framed the move as turning AI factories into an investable asset class. “In AI, compute is revenue,” he said. “NVIDIA compute is uniquely suited for this role. It is broadly adopted, flexible across models and workloads, fungible and transferable across customers and operators, and continuously improved through CUDA software.”

The partnerships create dedicated pools of capital at scale for Nvidia customers. Final agreements remain subject to execution. The structure lets long-duration capital underwrite the hardware and facilities while Nvidia continues to sell chips and software.

That architecture reduces the need for Nvidia itself to sit on the full guarantee stack for every large offtake. The Ohio revision is the first visible test of the new model in a headline deal.

Readers tracking the same shift can see how Nvidia’s new compute financing platforms recast GPUs and data halls as collateral rather than pure vendor risk.

In practice the six firms become the bridge between Nvidia’s product cycle and the multi-decade life of a power-and-land package. OpenAI still needs the chips. The platforms are meant to own more of the paper that pays for the halls around them.

What the Portsmouth Campus Includes

The campus sits on federal land at the former Portsmouth Gaseous Diffusion Plant in Pike County, Ohio, plus some private parcels. The Department of Energy is leasing the site to an SB Energy-affiliated entity. SoftBank Group and SB Energy plan to build 10 gigawatts of new power generation, including at least 9.2 GW of natural gas, specifically to serve 10 GW of data-center load.

SB Energy and AEP Ohio are investing $4.2 billion in new transmission infrastructure. SB Energy has committed to paying for those upgrades so the costs do not land on ordinary Ohio ratepayers. Excess generation and transmission capacity is supposed to flow back to the grid. Long-lead electrical equipment has already been secured.

Earlier reporting put first-phase capacity at roughly 800 megawatts targeted for 2028 under a 20-year lease structure in which OpenAI would control the equipment and payments would begin once operations start. OpenAI itself is an investor in SB Energy alongside SoftBank.

1. March 2026: DOE and Commerce announce the public-private partnership; SoftBank/SB Energy commit to the power and transmission package on the Portsmouth site.
2. June 2026: Reports surface that OpenAI is in advanced talks to lease the full 10 GW campus with potential Nvidia financial backing.
3. July 2026: WSJ details Nvidia talks around a roughly $250 billion backstop.
4. August 10, 2026: Nvidia launches the six-firm compute financing platforms.
5. August 14, 2026: WSJ reports the guarantee scaled to under $120 billion for phase one only; deal possibly this weekend.

The power side is deliberately designed so the data-center tenant and developer, not local families, carry the incremental plant and wire costs. That is the Ratepayer Protection structure the administration has highlighted.

Matching new gas generation to data-center load is the core engineering choice. The 9.2 GW natural-gas floor inside the 10 GW package is meant to keep the campus from leaning on the existing Ohio grid for its primary supply.

OpenAI’s Lease Still Targets the Full Ten Gigawatts

OpenAI closed a funding round on March 31 with $122 billion in committed capital at a post-money valuation of $852 billion. Amazon, Nvidia and SoftBank anchored the round; Microsoft continued as a long-term partner. The company said it is generating $2 billion in revenue per month and that enterprise now accounts for more than 40 percent of revenue.

It remains unprofitable. Cash burn on talent, chips and facilities continues even as usage and revenue climb. OpenAI lists Nvidia as the foundation of its training fleet and most of its inference stack, while also expanding cloud, silicon and data-center partners that include SB Energy and SoftBank.

Durable access to compute is the strategic advantage that compounds across the entire system: it advances research, improves products, expands access, and structurally lowers the cost of delivery at scale.

That sentence sits in OpenAI’s own funding announcement. The Ohio lease is one concrete attempt to lock that advantage under OpenAI’s control rather than through pure cloud contracts. The binding lease for the full 10 GW is still under discussion even as the Nvidia backstop shrinks to the first phase.

Revenue at $2 billion a month and an enterprise share above 40 percent show demand is broadening. They do not by themselves fund a campus whose total cost could exceed $500 billion once chips are included. The lease and the outside capital platforms are meant to close that gap.

The Downside Now Sits With Capital Markets

Crowd reaction on X treated the cut as a brake pedal rather than a stop sign. The sharpest takes asked who ultimately eats the downside once vendor guarantees shrink and institutional platforms expand. That question is now structural.

By packaging compute as an asset class with usage-linked economics and CUDA lock-in, Nvidia can sell more chips while the long-term capital providers underwrite the offtake and construction risk. The first large guarantee after the platform launch was trimmed by more than half for the initial phase. That is consistent with the design, not a retreat from the AI buildout itself.

The same pattern appears in the broader ledger of hidden AI debt reaching pension funds and other long-duration holders. Risk is migrating from corporate balance sheets into the portfolios that buy the new paper. Investors who pushed back on a $250 billion Nvidia exposure were effectively forcing that migration into the open.

OpenAI still gets a path to dedicated capacity. SoftBank still gets to develop the power and land package. Nvidia still supplies the GPUs. The difference is the size of the contingent liability that sits directly on Nvidia’s own books in the first years.

How Phase One Reshapes the Early Stack

The first phase is the narrow end of a much larger funnel. Roughly 800 megawatts targeted for 2028 is only a fraction of the planned 10 gigawatts, yet it is the slice that forces contracts, equipment orders and payment schedules into motion.

Under the 20-year lease structure described in earlier reporting, OpenAI would control the equipment and begin payments once operations start. That design ties cash outflow to working capacity rather than to a ground-breaking ceremony. It also explains why a backstop sized to phase one can still unlock the wider campus talks.

The table is a map of sequencing, not a retreat. Full-campus ambition stays on the OpenAI lease track. The guarantee is what got redrawn to fit the first operating block and the new financing platforms announced on August 10.

SoftBank and SB Energy still carry the power, land and transmission package. Nvidia still stands behind early paper, only at a lower ceiling. Long-duration capital is expected to take more of what used to land on a single vendor balance sheet.

Why the Weekend Timing Still Matters

A possible signing as early as this weekend would land four days after the six-firm platform launch and days after the Journal’s revised backstop figure. That is a tight public sequence for a project of this size.

• Platform memorandums first, so third-party capital is already the stated path
• Backstop cut next, so Nvidia’s own contingent liability matches phase one
• Lease talks continuing in parallel for the full 10 GW

None of those steps requires the others to finish on the same day. Together they tell investors how Nvidia wants large offtake deals to look after the August 10 announcements: chips and software on the vendor side, construction and offtake risk shared with institutions, and a tenant still chasing dedicated capacity at scale.

If the papers close near the reported terms, Portsmouth becomes less a one-company bet and more a template. The template can still fail in execution. Final platform agreements remain subject to closing, and the full lease is not done. The direction of travel is nonetheless clear from the figures already on the table.

Ten Thousand Construction Jobs and a Ratepayer Shield

For southern Ohio the numbers are concrete. More than 10,000 construction jobs over four years are projected across power plants, pipelines, transmission, substations and data halls. More than 2,000 operational roles are expected afterward. Tens of thousands of indirect jobs are forecast in manufacturing and services. SB Energy has also committed a $40 million community benefits agreement and to accelerated cleanup at the old uranium-enrichment site.

• New 765 kV transmission lines and four substations paid by the project
• Dedicated data-center rate structure so families and small businesses do not fund the incremental plants
• Commitment to reduce load or supply backup power to the local grid in emergencies
• Japanese capital supporting the bulk of the new gas generation

The campus is one of the clearest examples of data centers that still need the grid even while they claim private generation. The gas plants and wires are new, yet the interconnection and excess-capacity rules keep the facility tied to the regional system.

If the first-phase guarantee closes near the reported terms and the lease progresses, the Portsmouth site becomes an early case study of how AI factories get financed once the vendor steps back from full backstops. The chips will still come from Nvidia. The capital stack will look more like infrastructure finance and less like a single-company contingent liability.

That is the practical outcome of the week’s two announcements moving in sequence.