OpenAI's Ohio Lease Pays It $688M a Gigawatt • The Weighted Average

SB Energy issued OpenAI warrants valued at $3.6 billion in January that were worth $5.5 billion by the end of June, granted to secure OpenAI as the anchor tenant on a 20-year lease at its Ohio campus, according to draft IPO documents reviewed by The Wall Street Journal and summarized in TNW’s account of the warrant grant. The direction is the story: the landlord paid the tenant, not the other way around.

Set that against the capacity involved. Nvidia’s August 17 announcement, filed as an exhibit to its 8-K, states that SB Energy and SoftBank will build at least 10 GW of new generation yielding 8 IT-GW of AI factory capacity at the PORTS-Pike Technology Campus, that OpenAI is the customer for all 8, and that SB Energy builds, owns, and operates the site under a 20-year lease. Divide the warrant value by the committed capacity and OpenAI’s signature is worth roughly $688 million per IT-gigawatt — an inducement nobody published, and the cleanest available price for what an anchor tenant’s credit is worth in this market.

The tenant is becoming the shareholder

The instruments cost SB Energy nothing today, which is the point of a warrant. They dilute whoever buys the IPO rather than the balance sheet now, and SB Energy is targeting $5 billion to $7 billion in proceeds with OpenAI expected to hold a single-digit percentage stake afterward. Public investors will therefore be asked to value a company whose largest customer is also a shareholder whose stake appreciates when the lease is deemed credible.

Nvidia sits in the same structure from a third angle. Its release confirms a $1.5 billion direct investment in SB Energy plus credit support on land, power, and shell for the initial 4.25 IT-GW, with an option on the remaining 3.75. Separately, CNBC reported Nvidia weighing as much as $3 billion in SB Energy — half at signing, half at the IPO — as part of roughly $100 billion in credit support for the campus. Chip supplier, power developer, model company, and the parent that owns two of the four all hold positions in the same buildout. This paper’s arithmetic that Nvidia now captures 47 cents of every hyperscaler capex dollar measured the concentration; the Ohio structure shows the financing mechanism that produces it.

SB Energy’s own framing is power-first rather than compute-first, and its company newsroom has been running the Ohio project as a reindustrialization story since the March announcement of a campus on Department of Energy land. That framing matters commercially: the scarce asset in this deal is not silicon but interconnection, permits, and generation, which is why a developer can extract a 20-year commitment from the most capital-rich buyer in the industry and still need to sweeten it with equity.

The community terms are real money and worth reading as risk pricing. The filing commits at least $4.2 billion in new regional grid infrastructure through a partnership with AEP Ohio designed to protect ratepayers, and an $80 million community benefits fund — SB Energy’s original $40 million plus a matching $40 million from OpenAI. Developers do not spend that on goodwill; they spend it because local opposition is now the binding constraint on gigawatt-scale siting.

What breaks the $688M-per-gigawatt read

Three things. The capacity denominator is a plan, not a delivery: the campus phases in beginning in 2028, and 8 IT-GW is what SB Energy intends to build, so the per-gigawatt figure will move as phases slip or expand. The numerator is a mark, not cash — the $1.9 billion of appreciation between January and June reflects a rising private valuation, and OpenAI cannot sell a warrant in a company that has not listed. And the reporting rests on draft documents and unnamed sources; neither party has commented on the terms.

The structural risk is more interesting than any of those. Warrants tied to a 20-year lease convert a liability into a partial hedge for the tenant: if the campus proves valuable, OpenAI captures some of the value its own commitment created; if demand disappoints, it still owes twenty years of rent, but its position in the landlord’s equity is worthless at exactly the moment it can least afford it. That is correlated risk dressed as diversification.

There is also a timing asymmetry worth naming. The warrants were granted in January, appreciated $1.9 billion by June on a private mark, and are being disclosed now, in draft documents, weeks before a listing. Every one of those steps increases the value of OpenAI’s position without OpenAI doing anything beyond signing, and each one lands in a window where the public price does not yet exist to contradict it. Private marks are comfortable precisely until a public price arrives.

For operators, the useful transfer is the benchmark. Anyone negotiating capacity at scale now has a published number for what an anchor commitment is worth to a developer — $688 million per gigawatt in equity value, on top of the lease — and should ask why a smaller tenant’s signature is treated as free. The paper’s earlier work on what a Rubin-class deployment costs per megawatt priced the hardware side of that negotiation; this prices the credit side. Watch the IPO prospectus: it will be the first document to state the warrant terms, the lease escalators, and the customer concentration in the same place, and it is the evidence that would confirm or kill this estimate. Today’s lead on vendors repricing risk around outcomes describes the same migration one layer up the stack — whoever absorbs the failure case sets the price.