Nvidia's $105 Billion Guarantee Turns GPUs Into an Asset Class
An SEC filing reveals Nvidia providing up to $105 billion in residual-value guarantees for OpenAI's 8 GW Ohio campus — a financing structure that treats GPUs the way Ford Finance treats cars.
Nvidia has agreed to backstop up to $105 billion of the value of a data centre it does not own, for a tenant it does not control, on a campus it will not operate.
The disclosure came in an SEC filing dated 17 August. It covers OpenAI's planned campus at SB Energy's PORTS-Pike Technology Campus in Pike County, Ohio — and it quietly changes what kind of company Nvidia is in this buildout.
What the guarantee is
This is not a $105 billion investment. It's a residual-value guarantee, capped in aggregate at that number, covering an initial 4.25 GW of IT capacity. Nvidia holds an option to extend support to a further 3.75 GW, which would take the campus to roughly 8 GW.
SB Energy builds, owns and operates the site under a 20-year lease with OpenAI.
The mechanism only fires on specified default or insolvency by OpenAI. In that event, Nvidia covers the shortfall between the guaranteed minimum value under the leases and whatever SB Energy actually recovers by re-leasing or selling the facilities. OpenAI has agreed to reimburse and indemnify Nvidia for anything paid out.
Separately, Nvidia is putting $1.5 billion into SB Energy directly, has secured land, power and shell capacity at the site, and will be the exclusive AI compute infrastructure provider.
The idea underneath it
Jensen Huang's framing is about securing long-lived infrastructure that can be upgraded generation over generation. The more interesting reading came from Steven Dickens of HyperFrame Research, who compared it to Ford Finance: GPUs retain real residual value long after the first three years of use.
That comparison does a lot of work. Vehicle finance exists because a car has a predictable, liquid second-hand value — which is what makes leasing, guarantees and securitisation possible in the first place.
Applying that to GPUs is an argument that AI accelerators have crossed into being an investable asset class, with useful lives of six to seven years or more. If that holds, the financing pool for AI infrastructure widens well beyond the balance sheets of a handful of hyperscalers.
If it doesn't hold — if depreciation on AI silicon turns out to be steeper than the model assumes — the guarantee is where that error shows up.
The power question nobody has answered
8 GW of IT capacity is the part that should give anyone pause. SB Energy and SoftBank have separately announced plans to develop at least 10 GW of new generation to support the project.
Dickens also pushed back on the industry's habit of measuring AI infrastructure in gigawatts at all, arguing that cores, GPUs, racks or servers are more quantifiable units. He has a point: power is an input constraint, not an output measure, and using it as the headline number tells you what a campus consumes rather than what it produces.
What we still don't know
The actual guarantee agreements have not been filed. Nvidia has said detailed terms and guaranteed minimum values will appear in its quarterly report for the period ended 26 July.
Until then, the $105 billion is a ceiling rather than a description. The structure is public; the pricing is not.
For anyone building an AI infrastructure business case in India or anywhere else, forget the number for a moment. The chip vendor is now underwriting demand risk for its own customers. That is a very different market from one where compute is simply bought.
Source: ETDataCenters — Nvidia Backs OpenAI's Ohio Data Center With $105 Billion Guarantee