So Nvidia is putting up to $100 billion into OpenAI. That's the number that went around earlier this week, and it's still rattling around in my head five days later, which is longer than I usually let a tech headline live rent-free before I move on to something else.
The shape of the deal, as far as anyone's laid it out: Nvidia invests in chunks as OpenAI builds out data centers, tied to something like 10 gigawatts of Nvidia hardware getting deployed over the next several years. First tranche is supposedly $10 billion, connected to the first gigawatt going live sometime in the second half of next year, running on whatever Nvidia's calling its next chip platform by then. Jensen Huang and Sam Altman did the whole standing-next-to-each-other-smiling thing for the cameras. Nvidia's stock popped. Financial Twitter (I refuse to call it X in my own head, sorry) immediately split into two camps.
Camp one: this is just the AI buildout, get used to it, numbers this size are the new normal now.
Camp two, which is where I live: doesnt anyone else think its a little funny that the company selling the chips is also the company financing the purchase of the chips?
Because thats basically what this is. Nvidia gives OpenAI money. OpenAI spends the money on Nvidia GPUs. Nvidia books the revenue, and presumably at some point marks up the value of its OpenAI stake because OpenAI now has a pile of very expensive compute sitting in a data center somewhere. Everybody's balance sheet looks great and no actual outside dollar had to show up to make any of it happen.
Ive been half-jokingly telling anyone who'll listen that this is vendor financing with extra steps, and then getting mildly annoyed when I remember Ive been saying that about a different AI chip deal for like two weeks now, because Nvidia has done some version of this with more than one company this year. At some point "circular" stops being an insult and starts being the actual business model, which is its own kind of interesting, I guess.
I keep thinking about Lucent. Not because I lived through the dot-com crash as an adult (I was in elementary school, I have exactly one memory from that whole period and it's my dad muttering about his 401k at the dinner table), but because the Lucent story got repeated to me so many times later, as a cautionary tale, that it just stuck. Lucent lent money to telecom startups so they'd turn around and buy Lucent equipment with it. The equipment got bought, the revenue got booked, the stock went up, and then a bunch of those startups couldn't pay Lucent back because they never had real customers of their own to begin with. Nobody's saying OpenAI is one of those startups, to be clear. OpenAI has actual paying users, actual revenue, actual products people open every single day. Thats a real difference and I dont want to pretend its not.
But the financing structure rhymes enough that it bugs me, and I dont think "it's different this time because the underlying product is good" is actually a full answer to the circularity question. Those are two separate things. A product being good doesnt tell you whether the accounting wrapped around it is sturdy.
What actually gets me is how casually "up to $100 billion" gets reported as if it's a wire transfer that already landed. It's not $100 billion. It's a ceiling on a series of payments that depend on a bunch of future conditions being met, on a timeline that stretches out past 2030 in some of the read-throughs Ive seen. Somewhere between the press release and the headline, "up to, contingent on, spread over several years" turns into just a number, and the number is the only part anyone remembers by dinner.
My index fund has Nvidia in it, same as basically everyone's does at this point whether they picked it or not, so I'm not exactly rooting for any of this to go badly. I just want somebody, anybody, to publish the actual cash-flow schedule instead of the ceiling number and the smiling-founders photo. That's probably not going to happen. Doesnt mean I'll stop asking for it every time one of these gets announced, which given the pace this year, might be again before October even starts.