Compute and Collateral: Gresham’s Law in AI Finance

Fix the exchange rate between good money and bad, and the bad circulates while the good is hoarded. That is Gresham's law, and a version of it has been running through AI finance, with signatures in place of currency.

A lender to the AI build asks one question about the customer on the contract: is it investment grade? If yes, ninety cents on the dollar, at AAA and at BBB alike.

When a price is fixed by rule, the quantity is allocated by willingness, not by quality. So the market did not gravitate to the weakest name. The weakest name, Oracle, gravitated to the market, because it was the only one offering at that price.

For a while that cost nothing. A lease for a building not yet open was a footnote: off balance sheet, and off the agencies' ledgers. That changed last year. The agencies began treating obligations not yet performed as debt, in effect bringing them on balance sheet, much as the leverage ratio once did to repo at the banks.

The leverage ratio did not stop funding flows. It moved them, to the balance sheets the rule did not reach and into the forms it did not count. The same thing is happening here. The build has not slowed; it has changed address. To understand AI finance now is to understand this new address.

• 

  Overview · One customer, three names

• 

  1 · One customer, three names

• 

  2 · Microsoft issues its own promise

• 

  3 · CoreWeave borrows on Microsoft’s name

• 

  4 · Two routes to the same lenders

• 

  5 · Oracle’s name is on the lease

• 

  6 · NVIDIA adds its name in Ohio

• 

  7 · What lending its name costs each name

• 

  8 · What a rating on the customer would release

• 
• 
• 
• 
• 
• 
• 
• 
• 

Compute and Collateral - LinkedIn.pdf

Download PDF • 545KB