NVIDIA Guaranteed an OpenAI Data Center Lease Up to a $105 Billion Cap: What the 8-K Reveals About Triggers and Termination

NVIDIA entered into residual value guaranties with SB Energy on August 17, 2026 covering leases at the PORTS-Pike campus in Pike County, Ohio, and disclosed in an 8-K filed with the U.S. Securities and Exchange Commission that its aggregate payment obligation for the initial commitment is cumulatively capped at $105 billion. The guaranties cover leases representing approximately 4.25 gigawatts of IT load, with an affiliate of OpenAI Group PBC as tenant on a 20-year lease and SB Energy building, owning and operating the data center. NVIDIA also announced a $1.5 billion investment in SB Energy the same day, and CEO Jensen Huang used the company blog to pose the question of whether this is circular financing and answer no. ASAP reads the press release, the blog post and the legally binding 8-K side by side to separate what is now fixed from what remains open.

The legal form is a residual value guaranty, not a rent subsidy

The 8-K names the instrument a residual value guaranty and applies it to leases representing approximately 4.25 gigawatts of IT load in the aggregate at the Portsmouth Site. NVIDIA states it can also provide credit support to secure approximately an additional 3.8 gigawatts, exercisable in its sole discretion. The press release describes the same option as the remaining 3.75 IT-GW, leaving a roughly 0.05 gigawatt difference in how the two documents label it.

The ceiling on payment is explicit. NVIDIA's aggregate payment obligation is cumulatively capped at $105 billion for its initial commitment under the agreements, a figure that appears in the filing and in neither the blog post nor the press release. The blog describes the same arrangement in softer terms: support limited to defined portions of lease and power payments along with a specified residual-value commitment, and not the full cost of the site or all of the tenant's obligations.

Timing is also sharper in the filing. Each agreement generally becomes effective upon commencement of the applicable lease, and the lessor's ready-for-service conditions are expected to be satisfied beginning in 2028. NVIDIA filed the arrangement under Item 2.03 as well, the item covering creation of a direct financial obligation or an obligation under an off-balance sheet arrangement.

Two trigger events activate the guaranty

The 8-K defines a Trigger Event in two ways: OpenAI's insolvency resulting in a default under a lease, and OpenAI's failure to make payments under a lease. When either occurs, NVIDIA pays an amount generally equal to the shortfall between the guaranteed minimum value of a lease as set out in the agreements and amounts recovered through a replacement lease or sale.

Five remedies follow. NVIDIA may assume the applicable lease, require the lessor to seek to relet the premises, initiate a sale process, allow the lease to be terminated, or defer exercising these remedies for up to one year while paying specified project agreement costs. That last option is effectively a clock: it buys time rather than forcing a distressed disposal into a weak market.

One further sentence changes the economics. OpenAI has agreed to reimburse and indemnify NVIDIA for any and all amounts actually paid to the lessor under the agreements. The guaranty therefore does not extinguish OpenAI's obligation; it moves the timing of payment onto NVIDIA while leaving NVIDIA a claim against OpenAI.

Termination includes OpenAI attaining a credit rating

The third of four termination conditions states the purpose of this deal more precisely than any marketing sentence. NVIDIA's obligations terminate on the earliest of the 20th anniversary of the applicable lease commencement, termination of the lease by OpenAI in accordance with its terms, OpenAI achieving a satisfactory credit rating, or other customary termination events.

The credit rating clause bounds what is actually being lent. NVIDIA is extending creditworthiness rather than capital, and the bridge is removed the moment OpenAI can stand on its own rating. Huang's blog language about frontier labs that have strong demand and rapidly growing revenue yet lack the decades-long infrastructure contracts and investment-grade financing capacity to secure infrastructure independently is the same idea written as a contract term.

It also reframes the headline number. The $105 billion is a conditional ceiling that persists until the 20-year mark or an upgraded rating, and, as the blog notes, NVIDIA's remaining exposure declines as OpenAI makes lease payments and capacity comes online. Read the other way, the only path from ceiling to cash runs through OpenAI failing to pay.

Why the company wrote the circular-financing question itself

The blog contains a question and answer where NVIDIA asks whether this is circular financing and answers no. The stated grounds are that OpenAI pays the lease, and that NVIDIA is using its scale and long-term visibility to secure a site hosting NVIDIA compute, applying the same discipline it applies to supply-chain management.

Posing the question is itself a signal of where scrutiny sits. In this single transaction NVIDIA buys $1.5 billion of SB Energy equity, guarantees the residual value of the leases, and secures the site to exclusively host its own compute. Three relationships converging on one campus make revenue and support look like they flow in the same direction.

What separates the two readings is the direction and conditionality of cash. If NVIDIA funded OpenAI and that money returned as GPU purchases, the circularity charge would land; under the structure in the 8-K, NVIDIA's cash moves only when OpenAI fails to pay, and it moves to the lessor rather than to the customer. That said, a lease that likely would not have closed without the guaranty does close because of it, and NVIDIA revenue accrues on top, so the milder claim that the arrangement helps manufacture demand is not refuted by the filing. The company's answer and the skeptics' concern address different layers, and the 8-K supplies the material to tell them apart.

The rest of the campus numbers sit in power and in the county

PORTS-Pike reindustrializes the decommissioned Portsmouth Gaseous Diffusion Plant and surrounding land, spanning private and federal parcels and developed in collaboration with AEP Ohio, the U.S. Department of Energy and the U.S. Department of Commerce. SB Energy and SoftBank will build at least 10 GW of new generation, which yields 8 IT-GW of AI factory capacity, and at least $4.2 billion goes into new regional grid infrastructure through a partnership with AEP Ohio designed to protect ratepayers.

Community terms are itemized. OpenAI adds an incremental $40 million to SB Energy's originally announced $40 million community benefits fund, for $80 million total, directed at affordable energy, job creation and workforce development, and community and economic development. Goldman Sachs and JP Morgan advised SB Energy; Morgan Stanley advised NVIDIA.

The line from 10 GW of generation to 8 IT-GW of capacity is where this industry's numbers most often get misread. Gigawatts of generation and gigawatts of IT load share a unit but not a value, and the gap covers cooling, losses and design headroom. That is precisely why the 8-K specifies IT load.

The resale argument rests on CUDA fungibility

Asked in the blog what happens to PORTS-Pike if OpenAI does not use the site in the future, NVIDIA answers resale: its compute is versatile, fungible and broadly adopted, so capacity can be resold to another qualified tenant across an ecosystem of cloud providers, enterprises, AI labs and startups. The supporting chain is that CUDA gives a common platform, versatility makes compute fungible, and fungibility drives utilization and durability, making compute a productive asset that is rentable and financeable.

That chain is not written in a form anyone can test today. Resale requires that some tenant capable of absorbing multi-gigawatt capacity exists at whatever date it is needed after 2028, and that the GPU generation, power price and rent of that moment clear the guaranteed minimum value. Since the filing defines payment as the shortfall against that minimum, the resale market's price is what actually determines NVIDIA's exposure.

The generational upgrade claim shares the structure. Each generation deployed at PORTS-Pike could represent approximately 1.5 million NVIDIA GPUs, or roughly $150 billion to $200 billion in NVIDIA revenue, with the site supporting multiple upgrade cycles over 20 years. OpenAI's existing and planned commitments represent approximately 12 gigawatts, expandable to approximately 16 gigawatts if the PORTS-Pike option is exercised, which NVIDIA frames as roughly $600 billion of compute through 2030. These are opportunity figures the company put forward, not booked revenue, and they depend on OpenAI's ability to pay and on deployment pace.

Two angles for readers outside the United States

The first is where the bottleneck moved. The resource NVIDIA says it is securing is land, power and shell rather than chips or memory, and the company frames this as applying its semiconductor-sourcing discipline to a new input. For anyone planning AI data centers elsewhere, that means grid interconnection and site permitting timelines now govern schedules more tightly than equipment procurement. The separately budgeted $4.2 billion for grid infrastructure shows that who pays for transmission, and how ratepayers are insulated from it, is a live negotiating item rather than a footnote.

The second is credit. What actually changed hands here is not GPUs but the creditworthiness needed to make a 20-year lease signable, supplied conditionally by the equipment vendor. Any large AI infrastructure contract raises the same question: who guarantees that the counterparty can pay for the full term, on what conditions, and when does that guarantee end. PORTS-Pike answers with an observable metric, a satisfactory credit rating, which is what makes it worth studying as a template.

What has not been disclosed

The contracts themselves are not public. The 8-K states that its description does not purport to be complete and that the form of the agreements will be filed as an exhibit to NVIDIA's Form 10-Q for the fiscal quarter ended July 26, 2026. How the guaranteed minimum value is calculated, how it is allocated across leases, and how recoveries are computed are therefore unavailable today.

The accounting treatment is also unsettled in public. NVIDIA filed the arrangement under the direct financial obligation and off-balance sheet item, but how much is recognized as a liability versus disclosed as a contingency will only be visible in the next periodic report. Press reporting should be handled as reporting: CNBC previously reported discussions of a backstop of up to $250 billion for a 10-gigawatt buildout at the site, and The Wall Street Journal reported that the guarantee was set to be cut to less than $120 billion. The confirmed figure is the $105 billion cap in the 8-K.

Finally, time remains. Capacity phases in from 2028 and the guaranties take effect lease by lease from that point. If OpenAI's credit improves in the interim, the guaranty can end early; if power construction slips, effectiveness slips with it. What is fixed today is a ceiling and a set of conditions, and the actual amount will be decided by facts that do not exist yet.

Source: ASAP analysis based on NVIDIA's Form 8-K filed with the U.S. Securities and Exchange Commission on August 17, 2026, the NVIDIA newsroom press release and Jensen Huang's blog post of the same date, and reporting by CNBC and The Wall Street Journal