Nvidia’s $105B OpenAI Guarantee Turns AI Demand Into a Credit Bet
Nvidia has put up to $105 billion behind an OpenAI data-centre lease for hardware only Nvidia will supply. That is not chip sales. It is customer-credit risk.
Nvidia has put up to $105 billion behind an OpenAI data-centre lease for hardware only Nvidia will supply. That is not just selling chips. That is underwriting your own customer.
Anyone calling this a normal technology deal is either being polite or not paying attention.
Nvidia is no longer merely selling the picks and shovels
On August 17, Nvidia, OpenAI and SoftBank-backed SB Energy announced the PORTS-Pike Technology Campus in Pike County, Ohio. It is an enormous proposed AI campus on private land and remediated federal land once used by the Portsmouth Gaseous Diffusion Plant.
The headline numbers are ridiculous enough to deserve a slow read.
OpenAI has agreed to secure roughly 8 gigawatts of IT capacity under a 20-year lease. SB Energy will build, own and operate the site. Nvidia will be the exclusive AI-compute infrastructure provider. The project is designed around at least 10 gigawatts of new energy generation, with the first 800 megawatts expected to become available in 2028.
But the number that matters is $105 billion.
Nvidia’s SEC filing says it has entered residual-value guarantees tied to the initial 4.25 gigawatts of IT load. Its aggregate payment obligation is capped at $105 billion. Nvidia also has discretion to provide support for about another 3.8 gigawatts.
Let’s strip away the AI confetti. If OpenAI becomes insolvent or fails to make lease payments, Nvidia can be on the hook for the shortfall between the lease’s guaranteed minimum value and whatever SB Energy can recover by re-leasing or selling the asset.
Nvidia is not writing OpenAI a $105 billion cheque today. That distinction matters.
But it is putting its balance sheet behind the economic usefulness of an asset built to run Nvidia equipment, for a customer that will use Nvidia equipment. That matters even more.
This is a financing structure, not a press release
The old version of the chip business was beautifully simple. You designed the chip, manufactured it, shipped it, collected the money and let the buyer worry about whether they could turn the thing into profits.
The new version is messier.
The supplier wants the customer to build sooner. The customer needs a colossal amount of capital before revenue catches up. The developer needs confidence that the tenant will pay. The lender wants a creditworthy backstop. So the supplier with the best economics and the strongest balance sheet gets pulled into financing the whole machine.
That is what has happened in Ohio.
Nvidia is providing the technology stack: GPUs, CPUs and networking. It is investing $1.5 billion directly in SB Energy. It is providing credit support for the first 4.25 gigawatts. And OpenAI will use the campus under a 20-year lease.
Everybody has an incentive to say this is sensible. It may well be. Nvidia has enormous cash generation, and it is not irrational for a dominant supplier to secure long-lived capacity for its products.
But a good business can still make a dangerous financing decision. Those are not opposites. In fact, dangerous financing decisions are normally made by very good businesses at the top of a boom, when confidence is high and the downside looks theoretical.
Why Ohio is the real economy story, not just an AI story
The PORTS-Pike campus will not be built on vibes and chatbot subscriptions. It needs gas generation, transmission lines, skilled trades, construction crews, land remediation, grid connections, cooling systems and years of coordination.
OpenAI says the six-year buildout through 2032 is expected to create 35,000 construction jobs and 2,500 long-term operating jobs. SB Energy and its partners say they will invest at least $4.2 billion in regional grid infrastructure through AEP Ohio. OpenAI and SB Energy have put forward a combined $80 million community-benefits fund.
Those are material figures for Southern Ohio. They are also why data centres have become the defining industrial project of this era. The winners will not only be model developers and chip designers. They will include power producers, utilities, transmission contractors, turbine makers, electrical-equipment suppliers, engineers, builders, landowners and the towns smart enough to negotiate hard.
This is where too many investors get lazy. They buy the famous ticker because it is easy, then miss the plumbing because it is boring.
AI’s binding constraint is not imagination. It is power delivered to the right patch of land, on time, at a cost that does not destroy the economics of the workload.
Nvidia understands that. That is why it is securing what it calls land, power and shell capacity. The company is effectively saying that a chip is not much use if there is nowhere to plug it in.
Fair enough. But it also means the AI race is becoming less like software and more like railways, LNG terminals and industrial plants: capital-heavy, politically exposed and hostage to execution.
The overlooked risk: the product can age faster than the building
Here is the bit that should make operators sit up.
The campus may be designed for repeated upgrades as Nvidia releases new generations of infrastructure. That sounds sensible. Computing improves; you upgrade the kit.
The catch is that the expensive physical bits last decades while the valuable electronic bits can become old news very quickly.
A 20-year lease can be a terrific asset if demand compounds and the hardware keeps generating more economic value per watt. It can be a bloody awkward liability if AI demand slows, a cheaper computing architecture wins, regulation bites, power costs jump, or OpenAI’s economics disappoint.
Nvidia’s residual-value guarantee is meant to help bridge that risk. It tells SB Energy and its financiers that there is a serious party standing behind the project.
But that guarantee also reveals the problem: the market apparently values Nvidia’s credit support highly enough that it helps make the deal work.
That does not prove a bubble. It proves capital is starting to ask the correct question: who carries the risk if the demand forecast misses?
The answer, at least for the initial commitment in Ohio, is not simply OpenAI. Nvidia shares it.
Don’t call every circular deal a fraud. Do call it what it is.
There is a fashionable habit of yelling “circular financing” at anything involving the same group of AI companies. That is a bit lazy.
A supplier supporting a strategic customer is not automatically dodgy. Airlines have received manufacturer-backed financing. Car makers have supported dealers. Equipment vendors have helped customers buy equipment since before most of us were born.
The proper question is simpler: would the project still be financeable, on acceptable terms, if the equipment supplier was not standing behind it?
If the answer is yes, Nvidia has used its balance sheet to improve a decent deal.
If the answer is no, the supplier is doing more than selling. It is manufacturing demand by reducing the buyer’s cost of capital and assuming some of the downside itself.
That can still be a superb strategic move. Nvidia may believe the Ohio site will remain valuable because power, land and grid access are scarce. If OpenAI stumbles, Nvidia has options under the agreements, including assuming leases, seeking a replacement tenant or pursuing a sale.
But scarcity does not make an asset immune to a bad purchase price. I have learned that lesson the expensive way in business. A great asset bought on stupid terms can make you feel clever right up until it doesn’t.
What this means for you
If you are an investor, stop treating “AI capex” as one clean growth story. Split it into three buckets: the companies selling compute, the companies building physical infrastructure, and the companies carrying financing risk. They will not earn the same returns, and they will not break at the same time.
If you own a business, steal the useful lesson without copying the madness: secure your bottleneck before demand forces you to pay silly money for it. For Nvidia, that bottleneck is land and power. For you, it may be distribution, inventory, engineering talent, licenses or a single supplier. Know it before the market does.
If you are raising capital, remember this: a customer commitment is worth more when the customer can actually pay. Do not confuse a big contract with bankable revenue. Ask who funds it, what happens if they default, and whether the asset has another buyer.
And if you run infrastructure, property, power or industrial businesses, pay attention. The AI boom is not only creating software winners. It is dragging very old-fashioned assets back to centre stage.
Nvidia’s $105 billion commitment is the tell. The hottest company in technology is no longer satisfied selling the engine. It wants to help finance the road, the petrol station and the truck carrying its engine.
That is either brilliant strategic positioning or a sign the boom now needs increasingly creative financial engineering to keep moving.
Probably both.