Nvidia Scales Back Its Proposed OpenAI Data Center Guarantee

Nvidia has reportedly cut its proposed OpenAI data center guarantee from $250 billion to less than $120 billion. The change, now prominent across Google News, is more than a smaller financing commitment. It reveals how quickly ambitious AI infrastructure plans collide with balance-sheet risk.

According to the revised financing report, Nvidia would initially back only the first half of a planned 10-gigawatt campus in Ohio. The chipmaker would then decide whether to support later phases. The companies have not publicly disclosed final terms, and the negotiations could still change.

That distinction matters because a guarantee is not the same as an immediate cash investment. It is a promise to cover defined obligations if another party cannot pay. Nvidia would therefore support OpenAI’s access to infrastructure while accepting a potentially significant contingent liability.

The original proposal would have tied Nvidia’s balance sheet more closely to one of its largest customers. The revised structure preserves the partnership while limiting Nvidia’s initial exposure. It also shifts more of the project’s financing burden toward OpenAI, SoftBank, outside lenders, and future investors.

What Nvidia Actually Changed

Nvidia appears willing to support the Ohio campus, but no longer wants to guarantee the entire project at the outset.

The reported negotiations concern a large data center campus being developed by SB Energy, a SoftBank subsidiary, in southern Ohio. OpenAI is discussing a lease covering up to 10 gigawatts of capacity. Nvidia would provide financial support that helps the developer raise money on more favorable terms.

The earlier proposal reportedly contemplated a guarantee of roughly $250 billion. That backstop would have supported the lease and the debt used to construct the campus. It would not necessarily have paid for the Nvidia processors installed inside the buildings.

The new proposal reportedly limits Nvidia’s initial guarantee to less than $120 billion. That support would cover a first phase representing approximately five gigawatts. Nvidia could later decide whether to participate in financing the remaining capacity.

This is a phased commitment, not a complete withdrawal. Nvidia still has a commercial incentive to help the campus proceed. A functioning OpenAI facility of this scale would require vast numbers of accelerators, networking products, and supporting systems.

However, the smaller guarantee changes who carries the early risk. Under the original concept, creditors could rely more heavily on Nvidia’s financial strength. Under the revised structure, they must examine the creditworthiness and commitments of every project participant.

OpenAI is a private company without an investment-grade public credit rating. Its infrastructure demand may be substantial, but lenders still need confidence that lease payments will continue for many years. A guarantee from Nvidia can reduce that uncertainty.

The guarantee would become relevant if the covered tenant or project entity failed to meet defined obligations. The precise triggers remain undisclosed. Investors therefore do not know whether Nvidia would assume lease payments, acquire capacity, support refinancing, or accept another remedy.

Those details matter more than the headline amount. A narrowly written guarantee with collateral and termination rights presents a different risk from an unconditional promise covering decades of payments.

Nvidia has experience with smaller arrangements. Its latest annual filing says it guaranteed certain partners’ facility lease obligations in exchange for warrants. The maximum gross exposure across those disclosed agreements was $3.5 billion, with $712 million held in escrow.

A guarantee below $120 billion would still sit far outside that disclosed historical scale. It would not mean Nvidia immediately owes the entire amount. It would mean a single relationship could create a contingent exposure many times larger than its existing guarantee book.

That gap helps explain the revision. Nvidia can support the first phase, observe construction progress, and evaluate OpenAI’s capacity needs before backing the rest. Phasing also gives lenders time to test whether the project’s economics match its enormous physical footprint.

The change does not reduce OpenAI’s reported ambition. OpenAI is still discussing a binding lease for the full 10 gigawatts, according to the report. What changed is the amount of that ambition Nvidia would support from day one.

Why the Google News Headline Matters

The Google News headline captures a financing reversal that challenges the assumption that AI demand automatically justifies unlimited infrastructure commitments.

Nvidia sells the processors that make large AI systems possible. OpenAI needs those systems to train models and serve users. That relationship gives both companies a reason to accelerate construction, even when conventional lenders hesitate.

The conflict begins when Nvidia helps finance the demand for its own products. Such support can unlock genuine customer demand that capital markets cannot efficiently fund. It can also blur the line between independent sales and supplier-supported expansion.

This concern is often described as circular financing. The term refers to arrangements where money or credit flows from a supplier toward customers that use the support to purchase the supplier’s products.

Circularity does not prove that a transaction lacks economic value. Equipment makers have long offered financing to qualified buyers. Energy developers also use long-term purchase commitments to secure project debt.

The risk grows when the same participants depend on one another at an exceptional scale. OpenAI needs chips and facilities. Data center developers need a credible tenant. Lenders want stronger credit support. Nvidia benefits when the completed facility purchases Nvidia systems.

If AI revenue eventually supports those obligations, the structure can work. If usage, pricing, or model efficiency changes faster than expected, several parties may discover that their protections depend on the same underlying customer.

The initial market response reflected that concern. Nvidia shares fell 5% after the earlier guarantee talks became public, according to the Journal. A one-day move cannot establish investors’ reasoning, but the subsequent restructuring reportedly addressed concerns about Nvidia’s exposure.

Nvidia’s own language recognizes that its OpenAI plans remain unfinished. Its annual filing says the company was finalizing an investment and partnership agreement with OpenAI. It also warns that no agreement or completed transaction was assured.

That disclosure is important because the companies previously announced a much broader strategic vision. In September 2025, they described a partnership involving at least 10 gigawatts of Nvidia systems and an investment of up to $100 billion.

The systems partnership said Nvidia’s investment would occur progressively as each gigawatt was deployed. It also targeted the first gigawatt for the second half of 2026 using the Vera Rubin platform.

Progressive funding already formed part of the public framework. The Ohio proposal now appears to apply similar staging to financial guarantees. Nvidia can support deployed or near-term capacity without assuming every later phase immediately.

This approach also reflects the project’s timing. A 10-gigawatt campus cannot become fully operational overnight. Construction, power generation, transmission equipment, cooling systems, networking hardware, and permitting all develop on separate schedules.

A gigawatt measures electrical power, not computing performance. Five gigawatts still represents a huge energy requirement. Actual model capacity will depend on hardware generations, utilization rates, cooling efficiency, and the amount of power available for computing.

Those variables can change during construction. New processors may deliver more computation per unit of energy. Models may become more efficient. Customer demand may grow faster than expected, or inference prices may decline.

A phased guarantee lets Nvidia reassess those conditions. It also gives OpenAI a chance to demonstrate that expanding usage can support the resulting infrastructure obligations.

That is why the story deserves more than a passing Google News summary. The smaller number changes the project’s risk allocation, even if the campus and partnership continue as planned.

The Real Contest Is Ambition Versus Bankable Demand

The central contest is not Nvidia against OpenAI. It is the AI sector’s infrastructure ambition against the demand that lenders can confidently finance.

OpenAI has repeatedly argued that more computing capacity will support better models and wider adoption. Nvidia benefits from that thesis because each new facility creates demand for accelerators, networking equipment, software, and complete server systems.

The companies’ interests remain closely aligned. OpenAI wants access to hardware before rivals secure it. Nvidia wants major AI developers to standardize their systems around Nvidia’s platform rather than competing accelerators.

Yet aligned commercial interests do not eliminate credit risk. Lenders need evidence that project cash flows can cover debt and operating costs across changing technology cycles. A long-term lease must remain valuable even after today’s chips become obsolete.

This creates a mismatch between infrastructure and hardware. Buildings, electrical connections, and generation assets can operate for decades. AI accelerators have much shorter replacement cycles because newer generations improve performance and efficiency.

The facility may retain value after its original processors age, but upgrades require more capital. Each replacement cycle can change power density, cooling requirements, and networking architecture. A campus designed today must accommodate systems that do not yet exist.

Nvidia can reduce this risk because it understands its hardware roadmap. It can coordinate power, cooling, networking, and deployment plans with OpenAI. However, its technical knowledge does not guarantee that end-user revenue will cover the resulting obligations.

OpenAI’s usage provides one encouraging signal. Its 2025 partnership announcement said the company had more than 700 million weekly active users. That measure includes people who may contribute very different levels of revenue and computing demand.

User growth alone does not answer the financing question. Lenders must consider enterprise contracts, consumer subscriptions, advertising, application programming interface usage, and the cost of serving each workload.

Inference, which is the computation used when a deployed model answers requests, can generate recurring demand. Training runs are more episodic. The economic mix between those workloads affects how consistently a large campus earns revenue.

Efficiency improvements complicate the picture. Cheaper computation can reduce the resources needed for each response. It can also encourage customers to use AI more frequently, potentially increasing total demand.

This is a version of the rebound effect. When a resource becomes less expensive, usage can grow enough to offset efficiency gains. Nvidia and OpenAI are effectively betting that lower costs will expand applications faster than they reduce hardware requirements.

The revised guarantee does not reject that bet. It sets a checkpoint. Nvidia can observe whether real workloads fill the first phase before supporting the second half on similar terms.

That checkpoint protects Nvidia from committing too early to assumptions about demand several years ahead. It also pressures OpenAI to convert product adoption into financial commitments that outside capital providers accept.

SoftBank faces its own pressure. Its subsidiary is developing the campus, but project financing depends on credible tenants, available power, and contractual protections. A smaller Nvidia backstop can require SoftBank to contribute more capital or secure additional guarantees.

Banks and private-credit investors also gain influence. They can demand collateral, completion guarantees, minimum lease payments, or restrictions on future borrowing. Their terms will indicate how conventional capital markets evaluate AI infrastructure risk.

The broader competitive setting makes those decisions urgent. Microsoft, Amazon, Alphabet, Meta, and Oracle can fund substantial infrastructure through established businesses and public debt markets. OpenAI lacks the same operating history and financing structure.

Those companies are not simply OpenAI’s infrastructure providers. Several also develop models, cloud services, or custom processors that compete with parts of Nvidia’s and OpenAI’s businesses.

Alphabet has tensor processing units. Amazon has Trainium accelerators. Microsoft is developing its own AI hardware, while Meta has pursued internal chips for selected workloads. Wider adoption of those systems would reduce dependence on Nvidia.

Nvidia therefore has a strategic reason to help independent AI laboratories and infrastructure operators compete with cash-rich cloud companies. Financing can become part of its platform strategy, alongside hardware, networking, and software.

The company reinforced that approach in August by announcing financing relationships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The plan seeks to mobilize more than $500 billion for AI infrastructure serving Nvidia customers.

Under the capital pool plan, outside institutions would supply much of the funding. Nvidia said it might provide residual-value support covering part of selected opportunities.

Residual-value support protects against the possibility that financed assets are worth less than expected later. It is narrower than guaranteeing every payment, although final risk depends on each contract.

Using outside capital could reduce Nvidia’s direct exposure while keeping customers on its systems. It also distributes AI infrastructure risk across insurers, asset managers, private-equity firms, banks, and other institutional investors.

That distribution does not make the risk disappear. It changes who absorbs losses if utilization or asset values disappoint. It may also reduce financing costs if investors believe Nvidia’s involvement improves project quality.

For developers and enterprise buyers, this financing debate can seem distant. It still affects product availability and market structure. Large infrastructure commitments influence which models receive compute, which services can lower prices, and which platforms can promise capacity.

Developers should not assume that announced gigawatts equal usable near-term computing resources. Construction schedules, grid connections, chip deliveries, and financing conditions can delay capacity long after a headline appears.

Enterprise buyers should also distinguish technical capability from supply assurance. A provider can offer an attractive model while remaining dependent on unfinished facilities and complex partner agreements.

Anyone tracking these commitments can use an AI knowledge base to connect announcements, filings, deployment milestones, and later revisions. The Ohio guarantee shows why the first headline rarely tells the full story.

What the Smaller Guarantee Does Not Resolve

Reducing Nvidia’s initial exposure improves the structure, but it does not prove that the first phase is financially sound.

The central report relies on unnamed people familiar with negotiations. Nvidia, OpenAI, SoftBank, and SB Energy have not released a final contract describing the guarantee’s triggers, duration, collateral, or termination rights.

Readers should therefore treat less than $120 billion as a reported proposal, not a completed obligation. The amount can change before signing. The parties could also separate the project into multiple agreements with different risk profiles.

A final guarantee might cover debt service, lease payments, construction costs, or only specified shortfalls. It might give Nvidia rights to take over capacity. It could include escrow funds, warrants, or claims against project assets.

Without those terms, comparisons based only on headline values remain incomplete. A large maximum exposure can carry limited expected losses if triggers are narrow and collateral is strong. A smaller guarantee can still create material risk if its conditions are broad.

The reported project size adds another uncertainty. Ten gigawatts describes an eventual target, not necessarily funded or operational capacity. The first phase reportedly covers about five gigawatts, but its internal deployment schedule remains unclear.

Earlier reporting said an initial portion of the campus could enter service in 2028. Power generation and transmission construction must keep pace with the data halls. Delays in either system can leave expensive equipment underused.

Local infrastructure also matters. Large campuses need water or alternative cooling systems, natural gas or other generation sources, substations, transmission connections, and skilled construction workers. Each component carries permitting and execution risks.

Technology risk is equally important. The proposed facility would likely host several hardware generations during its life. Changes in rack density can force redesigns of electrical and cooling systems before construction finishes.

Nvidia benefits if the campus remains optimized for its roadmap. OpenAI benefits if it can obtain newer systems without being trapped by outdated facility designs. Lenders need agreements that preserve asset value through those upgrades.

Competition introduces another variable. OpenAI could diversify among Nvidia, AMD, cloud-provider chips, or future accelerators. Diversification might reduce supply risk, but it can weaken Nvidia’s incentive to guarantee infrastructure that supports competing equipment.

Conversely, a financing agreement could encourage long-term Nvidia purchases. That outcome would strengthen Nvidia’s platform position, yet it could limit OpenAI’s flexibility when alternative chips become attractive.

Regulators might also examine how financing affects competition. A dominant supplier supporting customer infrastructure is not inherently unlawful. Questions arise if contractual terms restrict rival hardware or make market access harder for other chip vendors.

No public evidence currently establishes such restrictions in the reported Ohio proposal. Any competition analysis must wait for actual contract terms or regulatory disclosures.

The project also faces concentration risk. A giant campus tied closely to one AI developer depends on that developer’s continued growth, credit access, and product relevance. Diversifying tenants can reduce that exposure, but OpenAI reportedly seeks a substantial dedicated lease.

The proposed guarantee partly addresses concentration by substituting Nvidia’s support for OpenAI’s standalone credit. That protection works only while Nvidia remains willing and able to perform under the agreement.

Nvidia’s financial position is much stronger than that of many infrastructure customers. Still, contingent obligations can accumulate. The company’s annual filing already warns that investments and partnerships can expose it to losses and changing market conditions.

The new Wall Street financing initiative expands that question. Nvidia may support many projects through limited residual-value commitments rather than one enormous guarantee. Investors will need to evaluate those commitments collectively.

Accounting disclosures will become essential. Maximum exposure, fair value, collateral, payment triggers, and concentration by counterparty can reveal more than promotional announcements.

Nvidia previously reported $3.5 billion in maximum gross exposure for partner lease guarantees. If the Ohio agreement closes, future filings should explain whether it enters the same category or receives different treatment.

The absence of immediate cash spending can make guarantees look harmless. That interpretation is too simple. A guarantee transfers tail risk, meaning losses associated with less likely but severe outcomes.

At the same time, treating the full maximum value as an expected loss would also be misleading. Nvidia would owe money only under specified conditions, and it could receive assets or other protections.

The prudent conclusion sits between those extremes. Nvidia is not writing OpenAI a check equal to the guarantee. It is considering using its credit strength to make a customer’s infrastructure easier to finance.

The smaller commitment indicates that this credit support has limits. It does not establish where those limits ultimately sit.

Three Signals to Watch After the Google News Cycle

The next evidence must come from contracts, construction, and financial disclosures, not another round of ambitious announcements.

The first signal is a binding lease and a signed guarantee. OpenAI’s reported discussions with SB Energy must become enforceable agreements before the project gains a stable financial foundation.

A signed deal would strengthen the view that the reduced guarantee is a deliberate risk structure rather than a retreat. It should identify the covered phase, the responsible entities, and the conditions that activate Nvidia’s support.

If the parties delay signing or repeatedly revise the scope, confidence in the project’s timeline would weaken. A delay would not prove that OpenAI lacks demand, but it would show that allocating the financial risk remains difficult.

The second signal is measurable progress on the first phase. Readers should watch for financing closes, construction milestones, grid agreements, equipment orders, and a credible schedule for operational capacity.

Groundbreaking ceremonies offer limited evidence by themselves. The stronger proof is committed capital paired with power delivery dates and contracted hardware. Those milestones show that the project has moved beyond planning.

Deployment should also be measured in usable capacity, not only announced gigawatts. A campus can have planned power without completed data halls, installed systems, or production workloads.

If the first phase advances on schedule and reaches high utilization, Nvidia’s phased approach will look disciplined. The company could support later capacity using evidence rather than forecasts.

If construction or power connections slip, the revision will look prescient. Nvidia would have avoided guaranteeing the full build-out before major execution risks became visible.

The third signal is Nvidia’s disclosure of its total financing exposure. Its coming filings and earnings calls should show whether large guarantees remain exceptional or become a recurring sales tool.

Investors need maximum exposure, collateral, counterparty concentration, and the fair value assigned to each commitment. They also need to know whether Nvidia can assume or repurpose facilities after a customer default.

The $500 billion institutional financing initiative adds urgency to that disclosure. Nvidia can mobilize outside capital while retaining limited support obligations. However, dozens of limited commitments can still create a meaningful aggregate exposure.

If Nvidia provides detailed disclosures and keeps its support bounded, the model can appear more sustainable. Outside investors would fund infrastructure while Nvidia concentrates on hardware and platform coordination.

If guarantees expand faster than independent customer revenue, concerns about circular financing will intensify. The chipmaker would increasingly depend on financing mechanisms to preserve the demand reflected in its own sales pipeline.

Google News will likely continue surfacing enormous numbers around OpenAI, Nvidia, SoftBank, and AI infrastructure. Readers should resist treating each number as equivalent.

An investment, lease commitment, debt facility, guarantee, and projected construction cost represent different obligations. Combining them without examining their terms can exaggerate both spending and risk.

The Ohio proposal is important precisely because the terms changed. Nvidia reportedly moved from guaranteeing the whole planned campus to backing an initial half. That is a material revision, even though the companies still appear committed to the broader project.

The sharper question is now whether staged financing produces staged evidence. Can OpenAI translate usage into durable revenue? Can SB Energy deliver power and facilities on schedule? Can Nvidia support customers without making its balance sheet the foundation of their demand?

Track the signed agreements, operational megawatts, and disclosed guarantee exposure after the Google News attention fades. Those three signals will show whether this is disciplined infrastructure financing or a warning that AI ambition has outrun bankable demand.