Gotta Dance

The House Sets the Clock

Larry Fink stood on CNBC Monday, next to Jensen Huang and the bosses of Goldman, Blackstone, Apollo, KKR and Brookfield, and compared what they'd just built to the day he started the mortgage-backed securities market in the 1970s. He meant it as a boast. Read it as a warning. The six of them agreed to move over $500 billion into AI compute, and the pitch is that a GPU is now an "investable asset" — productive, long-lived, fungible, financeable like a toll road or an apartment building.

One problem. A toll road doesn't get obsoleted by a better toll road every three years. A GPU does. Bankers know it, which is why they've never really lent against chips before. So Jensen solved it the only way a chip vendor can: he offered to backstop up to 25% of the depreciation himself. He's insuring the collateral on his own product. And he can, because he's the one who decides when the successor chip ships. He sets the depreciation clock the whole $500 billion is priced against.

That's the tell, and it's the pattern of the week. Anthropic said it will watermark every word Claude writes — invisible ink only Anthropic can read, to satisfy an EU transparency rule. OpenAI shipped a cyber model that completes 95% of exploit requests, its own number, behind a list it controls. OpenAI also marked its own valuation flat at $852 billion by buying stock back with its own checkbook. In every one of those rooms, the referee who's supposed to price the risk turned out to be the party that profits from the answer.

Michael Burry says these chips really live two-to-three years, not the five-to-six the hyperscalers book, and if he's right the industry is overstating profits by something like $176 billion. He's probably right — in ten years. But Jensen controls the roadmap, nobody else can design, fab and mass-produce at his scale inside five, and the market can stay irrational a lot longer than Burry can stay solvent. So credit the man. It's a smart trade and it holds for a while.

This isn't 2008. Subprime worked because the risk was hidden and the correlation was a surprise. Here everybody can see it's one giant correlated AI bet, and they're piling in anyway, because Wall Street doesn't care what the asset is as long as it can slice it, rate it, and trade it. Prakash laid out the machine on X: reference-design data centers, fungible collateral, ABS, CLOs, CDOs. He didn't even get to the credit-default swaps that get written on top.

The kid in Singin' in the Rain sang "Gotta dance!" like it was the best feeling on earth. Chuck Prince, Citigroup, 2007, said the same two words as a confession: as long as the music is playing, you've got to get up and dance. One's joy, one's compulsion, and the AI trade is doing both. They're dancing. Just know who's holding the paper when it stops, because it isn't Jensen.

At COAI today: the full Signal/Noise — the securitization stack, the self-certification pattern across five rooms, and why the lender is the one who eats the loss — is live at getcoai.com.

Harry DeMott: [email protected]

Anthony Batt: [email protected]

$500 billion. That's what six of Wall Street's biggest firms just agreed to move into AI compute — and the whole structure is priced on a single number Nvidia itself gets to set: how long a GPU stays productive. Hyperscalers book five-to-six years. Michael Burry says two-to-three, a gap worth roughly $176 billion in overstated profit. Jensen writes the chip roadmap that decides who's right, and he's backstopping 25% of the bet. The house set the clock and dealt itself a hedge.

Re-run your biggest compute contract on a three-year clock. The entire $500 billion structure assumes these chips stay productive for five-plus years. Burry's bear case says two-to-three. Pull your largest AI-infra or GPU commitment today and redo the payback math on a 36-month life. If it only pencils at 60 months, you're the one quietly absorbing the depreciation risk Wall Street just tranched away.

If you're bullish on AI, own the equity — not the paper. We said it on August 5: credit is the wrong instrument for a fat-tailed asset. Nvidia backstopping a quarter of the loans tells you exactly where the residual risk sits, and it isn't with the lender clipping a coupon. Bullish means buying the upside. A substandard credit return on a chip that might halve in value is the sucker's side of the trade.

Ask what your pension or insurer now holds. Prakash's chain ends where these things always end — tranched, investment-grade-rated compute debt sold to pension funds and insurers. That's you. At the next statement or annual meeting, ask the plain question: how much of our "diversified private credit" is now AI-infrastructure paper? You're likely long this trade whether you chose it or not.

Wednesday, August 12

— Harry and Anthony

Sources: