BlackRock just put out a paper saying AI agents need machine native money and tokenized compute. Stablecoins for payments. Claims on GPU time as a new asset class.
Raoul Pal’s point is they stop too early.
Agents don’t just need dollars and chips. They need energy to run the chips, data to know what’s true, contracts they can execute, and identity so they know who they’re dealing with.
His line is simple: EVERYTHING BECOMES A TOKEN.
The physical race right now is energy + compute. That’s the bottleneck.
Then the rails that make it usable:
ethereum:0x514910771af9ca656af840dff83e8264ecf986ca for verified data and contracts
sui:native for fast machine settlement
midnight-3:native for privacy so an agent can act without broadcasting its whole strategy.
Machines will buy kilowatts and GPU seconds the same way they buy dollars.
That’s the part most people are still underpricing.
Funnily enough I wrote about this yesterday...
BlackRock gets the rails right:
- AI is machine native intelligence, crypto is machine native money
- Agents cant use banks so they need blockchains
- Stablecoins are the money they pay with
- Compute becomes a tokenised asset class
But they stop at money and compute. The machine economy needs far more than that... identity, contracts, attention, energy, information itself. All of it gets tokenised, and most of those asset classes dont exist yet.
Everything will be a token. And I mean everything.