A guarantee costs nothing until it does. Nvidia books revenue on chips shipped into Pike County while the promise to cover a portion of the lease and power payments, and to hold the site to a minimum value, sits outside the income statement until the tenant stops paying or the asset drops through its floor [3]. Spread the ceiling across the term and the shape gets clearer. Up to $105 billion [1] against a 20-year lease [5] averages about $5.25 billion a year [1] of rent and power carrying Nvidia's name.
Set that against what Nvidia is actually spending. The $1.5 billion of equity going into SB Energy [6], added to the $1 billion OpenAI and SoftBank put in months earlier [7], is $2.5 billion of disclosed equity [2] sitting under a contingent commitment 42 times its size [3]. Per unit of power that is roughly $13.1 billion for each of the eight gigawatts [4], of which 800 megawatts, a tenth of the eventual site [5], is due in 2028 [8]. And measured against the more than $500 billion of third-party AI infrastructure funding that Nvidia and six financial institutions including BlackRock were targeting a week earlier [10], this single tenant's Ohio lease is 21 percent of the programme [6].
Equity gets defined first, possibly including a SB Energy IPO and direct SoftBank money [11], and only afterwards comes a debt tranche of project finance loans and perhaps bonds [12]. That sequence matters more than the ceiling itself: the guarantee is an input to the debt pricing rather than an output of it, and whoever sizes that tranche is underwriting Nvidia with OpenAI's usage as the upside case. Huang says this is not circular financing, and that Nvidia is using its scale and long-term visibility to help [9]. On the narrow point he is right. But the sharper version of the objection is not about circularity at all, it is about seniority: the circular-funding argument was always about revenue quality [2], while a guarantee is about seniority, and Nvidia has just moved itself behind its own customer in the payment waterfall of a site where it is also the exclusive chip supplier [4].
The counter-thesis rests on land and power being the binding constraint on new data centres, given a strained grid and local opposition over electricity prices and water use [13], so a floor under eight gigawatts of interconnected Ohio capacity may be the cheapest guarantee Nvidia ever writes, and if interconnect is scarcer in 2035 than in 2028 the minimum-value clause never triggers and the whole thing cost $1.5 billion and a signature [6]. This is probably wrong, but the case I would price is the middle one, where OpenAI survives and renegotiates, because a portion of lease and power still leaves shareholders standing behind a tenant they do not control. What would break the thesis is size: if the equity layer lands large enough that the guaranteed slice is a small fraction of the $105 billion ceiling [1], this is ordinary credit enhancement and the arithmetic above is theatre. Either way, the scarce resource Nvidia is allocating in Ohio is its own credit.
