Nvidia Reportedly Scales Back Ohio OpenAI Data Center Guarantee

Nvidia has reportedly cut its proposed guarantee for OpenAI’s Ohio data center from $250 billion to less than $120 billion. For anyone following the story through Google News, the reduction is more important than another enormous infrastructure headline. It shows Nvidia placing a limit on how much financial risk it will accept to support future chip demand.

The revised plan would initially cover about five gigawatts of the proposed 10-gigawatt campus, according to reporting based on people familiar with the negotiations. Nvidia would decide later whether to support the remaining capacity. OpenAI was still discussing a binding lease for the full project when the change was reported.

That structure creates the central conflict. OpenAI wants infrastructure on a scale that requires unusually large and durable financial commitments. Nvidia benefits when that infrastructure buys its systems, but guaranteeing the project also exposes the chipmaker to its customer’s long-term ability to pay.

The project has not been canceled, and the reported reduction is not the same as Nvidia withdrawing. However, the shift turns a sweeping commitment into a phased decision. It also gives investors, lenders, and enterprise AI buyers a clearer test for separating announced capacity from financed capacity.

Nvidia’s Reported Guarantee Fell Below $120 Billion

The reported change reduces Nvidia’s immediate exposure while leaving the wider Ohio project alive.

The original proposal emerged in late July. Nvidia was reportedly considering a guarantee of roughly $250 billion to help OpenAI lease a data center campus developed by SoftBank subsidiary SB Energy.

A financial guarantee is not an immediate cash payment. It is a promise that the guarantor will cover specified obligations if the primary borrower or tenant cannot meet them. That promise can help a developer obtain financing on better terms.

The proposed guarantee reportedly covered lease obligations and debt used to build the campus. It did not include the Nvidia hardware planned for installation inside the facility. Separate discussions reportedly considered financing as much as $350 billion in chip purchases.

Those figures made the original proposal exceptional. The campus could require more than $500 billion when construction and computing systems are considered together, according to the initial reporting summarized by project coverage.

The revised outline is narrower. Nvidia would reportedly guarantee less than $120 billion and focus first on approximately five gigawatts. The company would preserve the option to evaluate the rest later.

That distinction matters because a 10-gigawatt lease and a five-gigawatt guarantee are different commitments. OpenAI can continue negotiating for the entire campus, while Nvidia limits the amount it initially places behind the financing.

The change also introduces a real checkpoint. Before expanding its support, Nvidia can assess construction progress, power availability, OpenAI’s finances, lender participation, and demand for the resulting computing capacity.

Earlier reporting said the first phase was expected to deliver about 800 megawatts in 2028. That initial phase is only a fraction of the planned campus, so the full build would still require years of construction and additional financing decisions.

The latest report therefore describes a sequencing change, not a completed financing package. No public announcement from Nvidia, OpenAI, or SB Energy has established final terms for the guarantee.

Reuters attributed the reduction to a Wall Street Journal report citing people familiar with the matter. That sourcing requires careful language. The proposal remains reportedly under discussion, and its structure can still change before contracts are signed.

This uncertainty is particularly important for Google News readers encountering several versions of the story. A proposed guarantee, a signed guarantee, a funded construction phase, and an operating data center represent four different milestones.

The current evidence supports only the first stage: companies are negotiating a smaller initial guarantee tied to a much larger proposed project.

Why Nvidia Reduced Its Risk

Nvidia is trying to protect its balance sheet without surrendering a major source of future GPU demand.

According to the Journal’s reported account, the reduction addressed investor concern about Nvidia’s financial exposure. Those concerns intensified after the earlier $250 billion figure became public.

The original proposal asked investors to look beyond ordinary semiconductor sales. Nvidia would not simply sell hardware to a customer. It would also place its credit behind infrastructure that could generate demand for that hardware.

That arrangement can be commercially rational. Large AI campuses require financing before they produce revenue, and lenders prefer creditworthy parties behind long-term obligations. Nvidia’s participation could help transform OpenAI’s demand forecast into a financeable construction plan.

Yet the same arrangement produces a conflict. Nvidia benefits when the campus orders Nvidia systems. If Nvidia also supports the financing, reported demand becomes harder to separate from demand enabled by the supplier.

Axios described this concern as circular financing, where capital moves among suppliers, customers, and infrastructure partners that depend on one another. Circular does not automatically mean illegitimate. It does mean investors must examine who ultimately supplies independent cash flow.

OpenAI is privately held and does not have an investment-grade credit rating. A developer and its lenders must therefore evaluate a long lease differently from one backed by a highly rated, profitable public company.

Nvidia can improve that credit profile by accepting contingent liability. However, doing so transfers part of OpenAI’s execution risk to Nvidia’s shareholders. That exposure becomes more consequential as guarantees approach the scale of Nvidia’s annual business.

The smaller initial guarantee limits this transfer. It also lets Nvidia support construction without making one decision about the entire 10-gigawatt campus.

Phasing aligns financial commitments with physical delivery. Data centers need land, permits, transmission equipment, generation, cooling systems, networking, and servers. Delays in any component can prevent completed buildings from producing usable computing capacity.

A five-gigawatt starting framework is still immense. The reduction should not be interpreted as Nvidia losing confidence in AI demand altogether. The chipmaker remains deeply connected to OpenAI and the broader expansion of accelerated computing.

In September 2025, the companies announced a systems partnership covering at least 10 gigawatts of Nvidia infrastructure. That letter of intent said the first gigawatt would use Nvidia’s Vera Rubin platform and begin deployment during the second half of 2026.

The Ohio negotiations are related to the same broader demand cycle, but they are not interchangeable with that announcement. A strategic target does not automatically identify the location, financing vehicle, lease terms, or construction schedule for every gigawatt.

The reduction therefore signals financial discipline more clearly than a strategic retreat. Nvidia still wants OpenAI to expand. It simply appears less willing to guarantee the entire Ohio plan at once.

That choice places more responsibility on OpenAI, SoftBank, outside lenders, and other capital providers. If the project has attractive economics, the financing group should be able to distribute the risk instead of concentrating it at Nvidia.

The Promise Versus Reality Behind the Ohio Campus

The main reversal is the gap between a 10-gigawatt infrastructure promise and the smaller commitment Nvidia will reportedly guarantee today.

OpenAI’s demand for computing capacity is real, but proposed scale is not the same as delivered capacity. The Ohio project demonstrates how many financial and physical dependencies sit between the two.

SB Energy is developing the campus in southern Ohio, including land associated with the former Portsmouth Gaseous Diffusion Plant near Piketon. The federal site has a long industrial history and continues to undergo environmental cleanup.

The campus is expected to combine federal and private land. Earlier reporting described plans for extensive new generation and transmission infrastructure to support the proposed load.

Ten gigawatts represents the electrical capacity required by a very large regional system, not an ordinary corporate data center. Reaching that level requires far more than ordering GPUs.

The developer must secure dependable power generation, grid connections, substations, cooling water or alternative cooling designs, construction labor, and long-term equipment supply. Each dependency has its own regulatory process and delivery schedule.

Ohio has also become a focal point in the political debate over data center costs. In May, the state suspended a tax incentive used by qualifying projects while officials examined the expanding cost of the program.

The Ohio policy debate includes electricity demand, utility investment, tax treatment, and community opposition. Those questions can affect project economics even when developers secure private financing.

OpenAI’s position adds another layer. The company has traditionally obtained substantial capacity through cloud providers such as Microsoft, Oracle, and Amazon. A direct lease would give it more control over infrastructure planning, but it would also create a larger fixed commitment.

Cloud contracts transfer part of the construction and utilization risk to infrastructure providers. A direct campus lease places more of that risk with the tenant and whichever parties guarantee its obligations.

This is why Nvidia’s role matters. Its guarantee could allow SB Energy to borrow against a stronger credit profile than OpenAI can provide alone. Reducing the guarantee makes the other participants more visible.

The project now has to answer several practical questions. Who guarantees the remaining lease exposure? Which lenders fund later construction phases? How will equipment purchases be financed? What happens if power arrives later than the buildings?

The reported plan has not publicly answered those questions. It instead gives Nvidia a way to proceed with the first portion while retaining discretion over the second.

That reversal also affects how readers should interpret headline numbers. An announced 10-gigawatt lease can describe an end-state ambition. It does not prove that all 10 gigawatts have financing, power, equipment, or a binding delivery schedule.

Google News aggregation can compress these distinctions because headlines must summarize a complex deal in a few words. Readers should look for verbs such as “discussing,” “guaranteeing,” “leasing,” “financing,” and “building.” They describe different stages of commitment.

In this case, “scales back” refers to Nvidia’s reported guarantee. It does not establish that OpenAI reduced its requested lease, that SB Energy cut the campus design, or that construction stopped.

The project can still reach its original scale. It must now do so through phased guarantees, additional capital, or both.

OpenAI and SoftBank Face the Harder Financing Test

Nvidia’s smaller backstop shifts pressure toward OpenAI’s creditworthiness and SoftBank’s ability to organize a broader capital group.

OpenAI remains the prospective tenant whose demand supports the project. A lender must therefore assess whether future OpenAI revenue can sustain years of large lease and computing payments.

Private-market valuations do not settle that question. A valuation estimates what investors will pay for ownership. It does not provide the recurring cash needed to service lease obligations or construction debt.

OpenAI can raise more equity, attract strategic partners, issue debt through project entities, or secure guarantees from additional counterparties. Each approach changes who absorbs the risk if demand grows slower than expected.

The company can also negotiate phased delivery. That approach would align its obligations with product adoption and revenue growth, but it may reduce the certainty developers and lenders want.

SoftBank and SB Energy face a related challenge. They must convert an ambitious campus concept into financeable phases with identifiable sources of repayment. Nvidia’s guarantee can support that process, but it no longer appears to cover the entire planned site initially.

One reported option involves Nvidia making a separate investment in SB Energy. That would give the chipmaker another financial connection to the developer, although such an investment would not replace the need for project debt or lease support.

The financing group will need to distinguish equity, guarantees, loans, equipment credit, and long-term purchase commitments. Combining them into one total creates an impressive headline but hides their different risk profiles.

Equity absorbs losses first and generally has no required repayment schedule. Debt requires repayment. A guarantee becomes relevant after another party fails to meet specified obligations. Equipment financing is tied to hardware, while a lease supports access to buildings and power.

Those distinctions explain why reducing a guarantee matters even if the project’s total projected cost remains unchanged. Nvidia is reportedly limiting the portion for which its credit provides direct protection.

The move can pressure OpenAI to demonstrate stronger revenue visibility. Enterprise contracts, paid consumer usage, API demand, and long-term capacity agreements would all help lenders evaluate whether the campus has support beyond optimistic forecasts.

It can also pressure SoftBank to recruit more institutional capital. Infrastructure investors often prefer projects with phased construction, contracted demand, and measurable delivery milestones. The revised plan moves closer to that structure.

Microsoft, Amazon, Google, and Oracle provide another reference point. These companies finance data centers using established cloud businesses, diverse customers, and substantial operating cash flow. OpenAI does not yet have the same financial profile.

That does not make its plan impossible. It does make direct comparison with hyperscaler spending incomplete. A private AI company dependent on external capital faces different constraints than a diversified public cloud operator.

The competitive question extends beyond OpenAI. Anthropic and other model providers also need large amounts of computing capacity, but many continue to rely heavily on cloud partners. That arrangement can restrict control while reducing direct infrastructure exposure.

OpenAI is attempting to gain more control without assuming every risk alone. Nvidia’s scaled-back guarantee shows the difficulty of that balance.

For developers and enterprise buyers, the financing outcome affects more than corporate balance sheets. Delayed capacity can influence model availability, API quotas, reliability, and the rate at which inference costs fall.

Organizations planning AI products should therefore track actual capacity delivery, not only model announcements. Teams can preserve project decisions, vendor commitments, and infrastructure assumptions inside a searchable technical knowledge base as these plans change.

The Ohio negotiations are a reminder that AI service roadmaps depend on physical and financial infrastructure. A model provider cannot serve unlimited demand simply because its software improves.

What the Reduced Guarantee Does Not Prove

The revision exposes uncertainty, but it does not prove that the AI investment cycle is collapsing or that the Ohio project will fail.

A cut from $250 billion to less than $120 billion is significant. It reduces the size of Nvidia’s reported initial commitment by more than half. It also answers investor concerns with a tangible change in structure.

However, the original figure was itself under negotiation. Comparing two proposed amounts does not have the same meaning as canceling a signed obligation.

The public record still lacks final contracts. It also lacks complete details about duration, triggers, collateral, project entities, lender protections, and Nvidia’s rights if the guarantee is called.

Without those terms, observers cannot calculate Nvidia’s expected loss. The headline amount represents maximum or potential exposure under a proposed structure, not an estimate of the cash Nvidia expects to pay.

The likelihood of loss depends on OpenAI’s ability to meet its obligations, the value of the underlying infrastructure, additional guarantees, and Nvidia’s ability to reuse or sell the capacity.

A completed data center has economic value, but specialized AI infrastructure can be difficult to repurpose. Hardware generations change quickly, and a campus designed around one tenant’s needs may require modifications for another.

Power access could make the site attractive even if the original tenant withdraws. Yet the value of that option depends on when power becomes available and whether another customer wants capacity at the same scale.

The revised plan might therefore be prudent risk management. Nvidia can support a customer and preserve future sales while avoiding an immediate commitment to the entire campus.

A more skeptical reading is also reasonable. The change may indicate that investor resistance, financing complexity, or uncertainty about OpenAI’s future cash generation made the earlier proposal difficult to defend.

Neither interpretation has been independently established. The reported motivation includes investor concern, but the companies have not publicly released a complete explanation.

The stock market’s response to the earlier guarantee discussions showed that investors were paying attention to contingent liabilities. Nvidia shares reportedly fell after the larger proposal became public. A share-price move, however, does not reveal whether the project will earn an adequate return.

The deeper issue is demand quality. Nvidia’s revenue is strongest when independent customers compete for its systems using cash generated from their own businesses. Supplier-supported demand can still be valid, but it deserves additional scrutiny.

Investors should ask how much Nvidia-supported capacity ultimately serves paying end users. They should also ask whether guarantees are temporary bridges or recurring requirements for each new construction phase.

The same discipline applies to OpenAI. User growth, product adoption, and headline revenue matter, but large fixed infrastructure commitments require durable cash generation.

It would be equally misleading to treat the project’s enormous figures as proof of inevitable demand. Infrastructure can be announced years before utilization becomes visible.

This is the verification gap at the center of the story. Nvidia, OpenAI, and SoftBank appear committed to pursuing the campus, but the final allocation of risk remains unresolved.

Readers should resist two extremes. The reduction is not meaningless financial engineering, and it is not conclusive evidence that the AI market has reached a breaking point.

It is a measurable sign that even the sector’s strongest supplier has limits. Nvidia appears willing to enable demand, but only through a structure that gives it another decision before supporting the full campus.

Three Signals That Will Decide What Happens Next

A signed lease, disclosed financing terms, and physical power milestones will show whether the Ohio plan is advancing beyond negotiation.

The first signal is a binding OpenAI lease. Reports said an agreement for the proposed 10 gigawatts could arrive soon, but no public contract was available when the guarantee reduction emerged.

A signed lease would strengthen the case that OpenAI remains committed to the full campus. Its value would depend on duration, delivery conditions, termination rights, and which obligations receive Nvidia’s guarantee.

If OpenAI signs only for an initial phase, the revised plan will look like a broader reduction in near-term ambition. If it signs for all 10 gigawatts, the financing burden for the unguaranteed portion will become the next central question.

The second signal is Nvidia’s financial disclosure. The company’s filings and earnings commentary can show how it accounts for guarantees, investments, and other commitments connected to partner facilities.

Nvidia’s next earnings discussion should draw questions about maximum exposure, risk controls, and the relationship between financing support and recognized revenue. Specific disclosure would strengthen the view that the company has bounded its exposure.

Limited disclosure would leave investors dependent on anonymous-source reporting. It would also make comparisons between the Ohio proposal and Nvidia’s existing partner guarantees more difficult.

The third signal is physical project execution. Watch for completed financing on the first phase, utility agreements, generation approvals, transmission construction, and a credible path toward the expected 2028 delivery.

These milestones matter because capital cannot substitute for unavailable electricity. A guaranteed lease supports financing, but the campus cannot operate until power and grid infrastructure arrive.

Progress on an approximately 800-megawatt first phase would support the argument that the 10-gigawatt plan has a realistic starting point. Delays would weaken the project even if the parties announce additional financial commitments.

Google News will continue surfacing large totals because they capture attention. Readers should instead track the sequence: signed lease, financed phase, available power, installed systems, and paying utilization.

For enterprise buyers, the practical action is straightforward. Record which AI roadmaps depend on unbuilt capacity, identify alternative providers, and revisit assumptions after each financing or construction milestone.

The Ohio project remains one of the clearest tests of the AI infrastructure boom. Nvidia has reportedly reduced its immediate guarantee, but OpenAI has not abandoned its scale target. The next contracts will show whether other investors accept the risk Nvidia chose not to carry at once.