When the Vendor Co-Signs the Lease: Nvidia's $250 Billion Guarantee and the Circularity Problem

The Wall Street Journal reported on July 26 that Nvidia is negotiating a guarantee of roughly $250 billion to help OpenAI lease a planned 10-gigawatt data-center campus in Piketon, Ohio — a project being developed by SoftBank's energy subsidiary on the site of a former uranium enrichment plant, with a total cost that could exceed $500 billion. A separate $350 billion financing structure for the Nvidia chips that would fill the site is reportedly also on the table.

Strip away the zeros and the mechanism is simple: if OpenAI cannot pay its rent, Nvidia pays it. The world's most valuable chipmaker would be co-signing the lease of its own largest customer. That is not how healthy demand gets financed. It is how demand gets manufactured when nobody else will underwrite it.

What the guarantee actually tells us

The stated rationale is technical. OpenAI carries a non-investment-grade credit profile, and lenders financing the Piketon construction — first phase around 800 megawatts, targeted for 2028 — want a stronger counterparty behind the lease payments. Nvidia's backing lets the developer borrow on better terms.

But read that backwards and it becomes an admission: debt markets, presented with OpenAI's roughly $1.4 trillion in accumulated compute commitments, declined to price the risk on OpenAI's name alone. The AI buildout has reached a scale where the only balance sheet big enough to absorb its credit risk belongs to the company selling the shovels.

There is precedent for skepticism about how firm these arrangements are. In September 2025, Nvidia and OpenAI announced a letter of intent for up to $100 billion of investment tied to 10 gigawatts of Nvidia systems. Nvidia's own CFO conceded in December that it remained non-binding, and by February 2026 the structure had collapsed into something smaller — a $30 billion equity stake in OpenAI's $110 billion round. The $250 billion guarantee is likewise unfinalized, and the WSJ was explicit that talks could still fall apart. The pattern is headline first, structure later, and the structure is usually less than the headline.

The loop, and who sits inside it

Circular financing has a precise shape here: Nvidia backs OpenAI's obligations, OpenAI uses that backing to lease capacity, the capacity is filled with Nvidia chips, and the chip orders are booked as demand that justifies Nvidia's valuation — which is what makes the guarantee credible in the first place. Investor Michael Burry and critic Ed Zitron have both flagged the Ohio arrangement as the clearest instance yet of the loop underwriting itself.

Institutional voices are now saying the same thing more politely. Gary Tan of Allspring Global Investments notes that capital is "increasingly being used to fund future AI customers and infrastructure deployments." Billy Leung of Global X calls it a deepening of "vendor financing that's already under scrutiny." Bloomberg tallies Nvidia's announced financing-linked deals this year at over $540 billion before the Ohio talks — with the new arrangements pushing the running total toward $750 billion.

The uncomfortable comparison is the telecom bust of 2000-01, when equipment makers like Lucent and Nortel financed their own customers' network buildouts and discovered, when traffic growth undershot projections, that they had been booking their own money as revenue. Nvidia's deals are larger by an order of magnitude, though its margins and cash generation are also incomparably stronger.

The Asian balance sheets holding the other end

What makes this an Asian story is who carries the residual risk. SoftBank is not just Piketon's developer: it has committed nearly $65 billion to OpenAI, financed partly through a $40 billion bridge loan, and its earnings now swing on OpenAI's paper valuation and eventual IPO. Masayoshi Son has effectively bet the firm a second time — and the Ohio campus concentrates the wager, since SoftBank's development returns and its equity stake both depend on the same tenant staying solvent.

Korea sits one layer down. Nvidia's separate arrangement with SK Group — a partnership worth more than $500 billion covering two-plus gigawatts of Korean data centers, SK's chip purchases and Nvidia's HBM memory buying, with the first SK Telecom facility due next year — plus a $1 billion stake in Naver's data-center expansion, means Korea's memory supercycle is increasingly a derivative of the same financing loop. SK Hynix reports earnings on July 29; its record backlog is real, but a growing share of the end demand behind it is being financed, directly or indirectly, by Nvidia itself.

What breaks, and what doesn't

The bull case is that none of this matters if the tokens keep selling. Guarantees only crystallize into losses if OpenAI actually misses lease payments, and OpenAI's revenue, whatever its losses, is still growing at a pace no software company has matched. On that view, Nvidia is doing rationally what Detroit's financing arms did for a century: using a fortress balance sheet to smooth its customers' capital access.

The bear case is about correlation. Nvidia's equity value, its chip demand, its guarantee exposure and its investees' solvency are all long the same single variable — AI revenue arriving on schedule. Vendor financing does not create risk, but it relocates it to the one place a shock would be most systemic. If inference prices keep falling faster than volumes rise, the entity that absorbs the miss is no longer a diversified bond market. It is the $4-plus trillion company at the center of the index, with SoftBank and Korea's chipmakers chained to the same anchor.

The guarantee may never be signed; the last megadeal wasn't. But the fact that it is being discussed at all is the tell. The AI buildout has outgrown the credit markets' willingness to fund it at arm's length. From here, the boom is being underwritten by its own beneficiaries — and the distance between demand and belief in demand has never been thinner.