"Nvidia's OpenAI backstop shrank by half — and that tells you exactly how markets price invisible promises"

Three weeks ago Nvidia was reportedly ready to stand behind $250 billion in financing for OpenAI's planned 10-gigawatt data center campus in southern Ohio. As of this week, that figure is down to less than $120 billion. The cut, reported by The Wall Street Journal and carried by Reuters, is being framed as a retreat. Read alongside the quarterly holdings filing Nvidia published the same week, it looks more like a clean lesson in how markets price what they can see — and punish what they cannot.

The first thing to get straight is what the backstop actually was. It was never an investment and never a purchase; it was a credit guarantee. OpenAI, for all its roughly $852 billion valuation, still has no investment-grade credit rating and is not profitable. To lease what would be the largest data center campus ever announced, it needed to borrow against somebody else's balance sheet — and Nvidia's was the one on offer. Lenders would price the debt against Nvidia's rating rather than OpenAI's. That distinction, guarantee versus stake, is the entire story.

Under the revised terms, Nvidia would initially backstop only half of the 10GW project and reserve the right to decide on the rest later. Its shares fell 5% when the $250 billion figure first surfaced in late July, and the guarantee has since been cut to under half its original size. Nothing about the Ohio campus itself changed in those three weeks. The physics of the project is identical; only the promise got smaller. Read those two facts in order and the sequence is plain: the market priced the promise, disliked the price, and the promise shrank.

Here is the first insight worth drawing out, and it is about transparency rather than technology. A 13F filing is a legal requirement, arrives quarterly, and is marked to market — anyone can read it. A financing guarantee is none of those things: it sits off the balance sheet until someone calls it. Investors could see Nvidia's equity book, so they priced it precisely. They could not see the guarantee, so they priced the fear instead, and the number came down. This is a specific instance of a general market rule: opaque commitments get discounted at the fear rate, not the fair rate. Companies that want credit for big, ambitious moves would do well to remember that invisibility is not neutral — it is expensive.

The filing Nvidia published on August 14 showed exactly what visibility looks like. Its second-largest position is SpaceX, at 122.8 million Class A shares worth roughly $21 billion at the end of June. Its largest is Intel — 214.8 million shares worth about $30 billion, built from an initial $5 billion investment. That Intel stake is a genuine triumph of timing: the stock has roughly quintupled over twelve months on the strength of its foundry turnaround, and Nvidia bought in less than a year ago. It bought into a recovery that then happened, which is a different thing from engineering one.

The SpaceX position has an unusual origin story. CNBC reports the shares trace back to Nvidia's $10 billion investment in xAI in January; Bloomberg puts a smaller, earlier sum into xAI during 2025 through a mixed equity-and-debt vehicle. SpaceX then acquired xAI in February at $1.25 trillion, which is how an xAI position quietly became a SpaceX position. The share count also shows how fast a 13F photograph ages: SpaceX closed at $140 on Friday, down from $170.86 at the end of June, so that stake is now worth closer to $17 billion. Intel has slipped too. Call it $12 billion of paper value gone in six weeks across two positions — and because it is all on the filing, the market can watch it evaporate in real time.

The second insight is about why Nvidia holds these particular names. Both of its two biggest holdings buy Nvidia silicon. Musk has said SpaceX will use Nvidia hardware exclusively and expects a significant allocation of next-generation Vera Rubin GPUs; Intel makes chips, and its own recovery runs partly through foundry work Nvidia sends its way. Nvidia spent more than $40 billion on equity in the first four months of 2026, and almost all of it went to companies that buy its hardware. This is not a diversified portfolio in any conventional sense — it is demand creation with a balance sheet. Every equity stake is a customer, and every customer is a way to move more GPUs. The flywheel is the strategy; the returns are almost a byproduct.

The same pattern shows up on the financing side. Six of the largest firms in finance signed up this week to build compute-financing platforms around Nvidia hardware, targeting more than $500 billion in outside capital. Those are memorandums rather than committed money — the caveat that keeps getting lost — but the direction is unmistakable. Nvidia is not just selling chips; it is building an entire financing layer whose whole job is to make it easier for customers to afford chips. The separate $350 billion agreement to finance OpenAI's chip purchases is the same idea at enterprise scale: money Nvidia would help arrange rather than guarantee outright.

Three things to watch from here. The first is the signature: either the first-phase deal is signed this weekend at under $120 billion, or the number moves again before anyone writes it down. The second is phase two — Nvidia has reserved the right to decide on the remaining 5GW later, and whoever ends up guaranteeing it will tell you who is actually carrying the risk. The third is the next 13F, which will show whether Nvidia held those SpaceX shares through the decline or quietly sold into it. All three are, in their own way, tests of the same question: is Nvidia a chipmaker with a venture arm, or a venture-style financier that happens to make chips?

None of this should be read as a knock on the deal. OpenAI still wants the full 10GW, the lease is still being negotiated, and a first phase of roughly 800MW is still on track for 2028. SoftBank's SB Energy subsidiary is developing the site, with Goldman Sachs and Morgan Stanley advising on either side. The power is the part nobody argues about — the U.S. government controls it, and Japan funds it separately under a recent trade deal. What changed is not the ambition; it is the market's insistence that ambition be priced transparently. In a capital-intensive industry moving this fast, the companies that can show their work will keep paying less for their risk than the ones asking investors to take the promise on faith.