Nvidia CEO Jensen Huang would like everyone to know that his company’s extraordinary new arrangement supporting OpenAI’s Ohio data center is not circular financing.

Technically, he has a point.

Nvidia is not handing OpenAI $105 billion, so OpenAI can turn around and spend it on Nvidia GPUs. The money does not travel directly from Nvidia to its customer and then immediately back to Nvidia. It takes a longer and more complicated route through SoftBank’s SB Energy, banks, investors, power providers, construction companies and a 20-year lease.

It is less a circle than an oval. But Nvidia has positioned itself at both ends.

Under the agreement, SB Energy will build, own and operate the PORTS-Pike Technology Campus in southern Ohio. OpenAI has agreed to lease the capacity for 20 years. The completed campus could eventually support approximately 8 gigawatts of IT capacity, with the first 800 megawatts expected to become available in 2028.

Nvidia will invest $1.5 billion directly in SB Energy and provide credit support for the initial 4.25 gigawatts. According to Nvidia’s filing with the Securities and Exchange Commission, its cumulative payment obligation is capped at $105 billion. Nvidia can elect to support approximately another 3.8 gigawatts.

The resulting facility will exclusively host Nvidia AI infrastructure. OpenAI will deploy Nvidia’s full-stack DSX AI factory platform, including GPUs, CPUs, networking, systems and software.

Follow the road and see where it leads.

A Guarantee, Not a Loan

The details matter because calling this a $105 billion loan would be inaccurate. Nvidia is providing residual-value guarantees that make the project easier to finance.

OpenAI remains responsible for making the lease payments. But if OpenAI becomes insolvent or fails to pay, Nvidia could be responsible for the difference between the facility’s guaranteed minimum value and whatever SB Energy can recover by finding another tenant or selling the property.

Nvidia would have several choices. It could assume the lease, require SB Energy to find another tenant, initiate a sale or allow the lease to be terminated. OpenAI has also agreed to reimburse and indemnify Nvidia for any payments Nvidia must make.

That last protection looks comforting until you consider when Nvidia would most likely need it. An indemnification agreement from OpenAI becomes less valuable if OpenAI is insolvent or otherwise unable to meet its obligations. It is a little like having someone promise to repay you if they ever run out of money.

Still, Nvidia is not committing to write a $105 billion check today. Its actual cash investment in SB Energy is $1.5 billion. The much larger number represents contingent exposure that may never be triggered.

In exchange, Nvidia’s balance sheet helps make one of the world’s largest planned data centers financeable, and the facility is being designed around Nvidia technology.

That is a potentially remarkable use of financial leverage. Nvidia does not need to fund the entire mine. It needs to help persuade everyone else that financing the mine is safe. Nvidia then gets to sell the machinery that goes inside it.

The $200 Billion Refresh Cycle

The potential payoff explains why Nvidia is willing to accept the risk.

Nvidia estimates that each generation of AI factory systems deployed across the initial 4.25 gigawatts could include approximately 1.5 million GPUs and represent between $150 billion and $200 billion in Nvidia revenue. The company says the campus could support multiple hardware upgrade cycles during its 20-year life.

Huang has also suggested that Nvidia could generate as much as $600 billion in revenue from its broader OpenAI relationship by 2030.

Those are Nvidia’s projections, not completed sales or contractual guarantees. OpenAI has not irrevocably promised to hand Nvidia another $200 billion every time Huang introduces a new architecture. Construction must be completed, power must become available, financing must hold together, demand for OpenAI’s products must continue growing and the economics of ever-larger AI systems must remain attractive.

Nevertheless, Nvidia has done something strategically significant. It has secured the initial deployment and placed itself in the preferred position for every refresh that follows.

Replacing 1.5 million GPUs is not the same as replacing a few servers in a corporate data center. The surrounding networking, software, operating tools, developer expertise and workload architecture become part of the decision. Once a facility of this size is built around Nvidia’s full stack, moving to another platform becomes neither simple nor inexpensive.

The exclusivity agreement may govern the initial facility, but the architecture creates its own gravitational pull. Nvidia does not need an ironclad promise covering every future generation if the technical and economic costs of leaving become sufficiently high.

This is not just a hardware sale. It is an attempt to create a multigenerational Nvidia habitat.

The Indispensability Trap

Nvidia’s challenge is not simply to keep selling the world’s most valuable chips. It must remain indispensable even as customers search for ways to become less dependent on it.

That pressure is already visible. Hyperscalers are developing custom silicon. OpenAI has reached a major deployment agreement with AMD. Google continues advancing its tensor processing units. Amazon, Microsoft and others have their own accelerator ambitions. Meanwhile, model architectures, inference techniques, quantization and other efficiency improvements could change how much brute-force GPU capacity each unit of intelligence requires.

Nvidia’s answer has been to keep moving up the stack.

It moved from graphics processors to AI accelerators. Then from chips to complete systems, networking and CUDA software. Now it is moving beyond the technology stack into infrastructure design, equity investment, project finance and credit support.

Nvidia is no longer waiting for the AI market to arrive with a purchase order. It is helping create the financial and physical conditions that allow those orders to exist.

That is the Indispensability Trap: The more successful a technology company becomes, the more it must expand beyond its original product to preserve its central position. Each move makes the company more important, but it also increases the obligations and risks it assumes. This is one of the dynamics I examine in my forthcoming book, Indispensability Trap, due in September.

Nvidia is becoming indispensable not only to AI compute but also to the machinery financing the AI buildout. It is acting as chip supplier, systems company, software platform, infrastructure architect, investor, credit backstop and organizer of outside capital.

That is strategically brilliant—as long as the demand is real and durable.

When the Vendor Helps Create the Customer

The uncomfortable question is how much of Nvidia’s demand remains independent of Nvidia’s own financial support.

There is nothing inherently improper about vendor financing. Companies have used it for generations to help customers purchase everything from aircraft and heavy machinery to telecommunications equipment and enterprise software. Financing can remove a legitimate bottleneck and allow economically sound projects to proceed.

Nvidia also has a credible argument that demand is not the problem. The immediate constraints are available land, power, grid connections, buildings and capital. If Nvidia has better visibility into future compute requirements than conventional lenders, using its balance sheet to clear those obstacles may be entirely rational.

But scale matters. A residual-value guarantee capped at $105 billion is not a routine customer accommodation. It turns Nvidia into a material participant in the credit structure beneath one of its largest customers.

That changes the questions investors should be asking.

Would the Ohio project receive the same financing on the same terms without Nvidia’s guarantee? How much of OpenAI’s capacity is supported by projected operating cash flow, and how much depends on continuously raising outside capital? What happens if inference becomes dramatically more efficient, model scaling produces diminishing returns or OpenAI’s growth falls short of its lease obligations?

There is also the question of residual value. Nvidia’s guarantee assumes that the facility will retain a minimum value even if OpenAI defaults. But the value of a specialized multi-gigawatt AI campus depends heavily on the existence of another tenant capable of absorbing it. There are not many companies that need—and can afford—4.25 gigawatts of AI infrastructure.

If another qualified tenant does emerge, it will almost certainly need enormous amounts of compute. Nvidia would again be positioned to supply it. Even the recovery plan can lead back to Nvidia.

More Oval Than Circle

Huang is right about one thing: This is not the simplest version of circular financing.

Nvidia is not giving OpenAI money earmarked for Nvidia chips. Its guarantee supports the land, power and shell that must exist before those chips can be installed. Banks and investors will supply much of the project capital. SB Energy will construct and own the campus. OpenAI will pay the rent. Nvidia will provide the compute infrastructure.

But complexity does not necessarily eliminate circularity. Sometimes it merely stretches the circle until it looks like an oval.

Nvidia’s financial strength helps make the facility possible. The facility creates room for an enormous Nvidia deployment. That deployment generates revenue and reinforces Nvidia’s market position. Nvidia can then use the resulting profits and balance-sheet strength to support the next infrastructure project.

It may be a rational flywheel rather than a speculative shell game. The determining factor will be whether OpenAI and the broader AI economy generate enough sustained revenue to support the capital being committed on their behalf.

That answer will not be found in a press release or an SEC filing. It will emerge over years as these facilities come online, the systems are populated and customers decide how much they are willing to pay for the intelligence produced inside them.

For now, Nvidia has constructed a deal in which banks finance the project, SB Energy builds it, OpenAI leases it and power providers feed it. Nvidia supports the structure, supplies its most valuable technology and places itself first in line for the upgrades.

It may not be circular financing. But when Nvidia helps build the roads, guarantees the destination and sells the machinery waiting at the end, it should surprise no one that all those roads lead back to Nvidia.