BlackRock Wants AI Agents to Pay in Crypto
The world's largest asset manager argues autonomous agents need stablecoins, blockchains, and eventually a futures market for compute.
By The crypto-money Desk
BlackRock has published a paper arguing that autonomous AI agents will need payment infrastructure built for 24/7, high-frequency, often tiny transactions — and that blockchains and stablecoins are the logical answer. [1]
The firm frames the thesis around what it calls the "Machine-Native Economy." In that model, AI agents constantly purchase APIs, data, and compute — payments too fast, too frequent, and too small for conventional banking rails to handle cleanly. [1]
Compute is the central resource. As it becomes a larger economic input, BlackRock expects markets to emerge for pricing, financing, hedging, and trading access to it. [1] That could eventually mean standardized claims on future compute capacity that can be tokenized, transferred, used as collateral, and settled programmatically, alongside markets like compute futures. [1]
The argument is less about crypto ideology than about plumbing. When the transaction volume is high enough and the unit cost small enough, the infrastructure has to match — and that infrastructure, in this vision, looks less like a bank and more like a protocol.
The urgency sharpens as AI inference grows relative to training. BlackRock cites projections showing inference rising from 25% of data-center power demand in 2025 to 43% by 2030, creating a much more fragmented, continuous market for compute. [1]
In that environment, agents could shop for compute in real time and pay for it per use, per job, or even per model token through protocols like x402. [1] The compute market begins to resemble a commodity exchange, with settlement happening at machine speed.
There is something clarifying about seeing the argument come from an asset manager rather than a crypto-native firm. The case is not aesthetic — it is structural: money that moves at human pace will struggle to serve agents that think at machine pace.
BlackRock's paper does not name a timeline for when compute futures or per-token payments become standard. But by anchoring the thesis to inference growth — a metric already in motion — the firm grounds a speculative architecture in a trend that data centers are already tracking. [1]
The question is not whether AI agents will need to pay for things. They already do. The question is whether the financial system gets rebuilt around that fact before it becomes a bottleneck.
agents could shop for compute in real time and pay for it per use, per job, or even per model token through protocols like x402BlackRock