BlackRock's Machine-Native Economy paper argues AI agents will settle data, API, and compute purchases in stablecoins. The remaining usage gap, the unresolved liability question, and the identity of who owns the rails all point the same way.
AI Agents Will Use Stablecoins — and Crypto Still Loses
Key Takeaways
- BlackRock's Machine-Native Economy paper argues that AI agents will pay for data, API calls, and computing power using stablecoins.
- Stablecoins already settle more than 11 trillion dollars a year and grow around 80 percent annually, yet AI agents account for well under 1 percent of machine-payment volume today.
- The binding constraint is not payments. Identity, authorization, and legal liability sit outside the chain and are not solved by a token.
- The value accrues to the owners of the rails and the layers above them, the same pattern that enriched application companies on top of open internet protocols rather than the protocols themselves.
What is the machine-native economy?
BlackRock, the world's largest asset manager, published a research paper on September 22, 2026 arguing that AI agents will soon buy data, API calls, and computing power, and that they will settle those purchases in stablecoins. The paper, titled The Machine-Native Economy, was read widely across the crypto industry as a green light. The more careful reading is closer to a warning.
The framing is precise. AI is machine-native intelligence; digital assets are machine-native money. As agents move from answering questions to executing tasks, they need to pay for inputs, such as a data query, a burst of compute, or an API call, without a human approving every transaction. Stablecoins are digitally native, always on, and denominated in dollars, which makes them the obvious settlement candidate.
The paper's authors identify the firms already building this layer: Coinbase's x402 protocol for paying for API calls, Stripe's Machine Payments Protocol with Tempo, the Agentic Commerce Protocol built with OpenAI, Circle's agent wallets, and OKX's Agent Payments Protocol. The work is underway rather than theoretical.
The firms laying that plumbing are incumbents with compliance teams, legal resources, and balance sheets: Stripe, OpenAI, Coinbase, Circle, and BlackRock itself, which already operates the tokenized money-market fund BUIDL. They intend to own the rails they are building.
Who wins if AI agents start paying each other?
If the machine-native economy develops, the value accrues to four groups.
First, the hyperscalers. The paper cites analyst estimates that cloud revenue at Amazon, Microsoft, and Google reaches roughly 1.1 trillion dollars by 2030, growing about 29 percent a year. Compute is where the revenue sits, and crypto wraps the payment around it.
Second, the stablecoin issuers. Circle and Tether already earn on the float, the yield on the reserves backing their tokens. Every agent holding a stablecoin to pay for compute adds to that float. The yield accrues to the issuer, not to the agent.
Third, the payment firms that own the front end. When an agent pays for an API call, a processor takes a fee. The paper notes that existing payment networks are adapting to agentic commerce. Stripe is building its own agent payment rails and already holds the merchant relationships and the compliance stack.
Fourth, the asset managers. BlackRock's publication of this research is also a commercial position. The same firms distributing tokenized money-market funds and Bitcoin exchange-traded products are describing a future of machine payments and tokenized compute, and are positioned to underwrite and securitize a compute-contract market when it forms.
The losers are agent-branded tokens, AI-specific chains, and lending protocols pitching themselves as the network for machine commerce. Their thesis is that crypto captures the value of the AI economy. BlackRock's own paper treats crypto as settlement infrastructure, not as the asset worth owning.
Why the real bottleneck is not payments
An AI agent cannot open a bank account. It cannot pass a customer-verification check. It cannot be sued, and it cannot sign a credit agreement. When an autonomous agent overspends on compute or purchases a fraudulent data stream, it is unresolved who carries the legal liability: the model provider, the wallet custodian, or the operator who deployed the agent.
Stablecoins solve the easiest tenth of this problem, which is moving value programmatically. They do not address authorization, identity, or accountability. The paper states that customer-verification, anti-money-laundering, and emerging know-your-agent checks will still be required, and that most of that identity and compliance infrastructure stays off-chain. The defensible layer is identity and permission, and it is being built by regulated institutions and fintech companies rather than by crypto protocols.
The usage figures show the distance remaining. Blockchain analytics firm TRM Labs tracked about 52.7 million dollars of x402 settlement volume this year, and AI agents account for somewhere between 0.6 percent and 7.5 percent of it. The gap between agents that will transact and agents that are transacting is where most of the narrative currently sits.
Why this resembles the internet's protocol trap
The pattern has a precedent. The defining applications of the internet, including email, search, social networks, and commerce, ran on open protocols. Those protocols captured none of the value they enabled; the gains went to companies that built defensible products and data advantages on top of them.
Crypto's position has been that tokens let a protocol capture the value it creates. BlackRock's paper describes the opposite arrangement: crypto as a neutral settlement layer over which application providers, payment processors, and asset managers build the commercial businesses. Machine-native money is a utility, and utilities are regulated, commoditized, and low margin.
The stablecoin data already reflects this. Stablecoins recorded about 11 trillion dollars in adjusted transfer volume in 2025, growing roughly 80 percent a year against about 8.5 percent for automated clearing house payments, and their combined market value has passed 300 billion dollars. The economics have gone to issuers, payment processors, and the banks now receiving crypto trust charters. The rail expanded without producing corresponding returns for token holders.
Where this heads in three years
Three outcomes look most likely, in descending order of confidence. Agentic payments become a real but narrow category running mostly on stablecoin rails controlled by established payment firms, with the identity layer off-chain. Tokenized compute contracts emerge as a genuine new asset class, issued and securitized by cloud providers and asset managers rather than by decentralized networks. And the agent-and-crypto theme fades as a retail narrative as institutional capital concludes there is no distinct token to buy, because the value sits with cloud providers, payment processors, and money-market funds.
The contrarian position is that the largest beneficiary of a machine-native economy will not be a cryptocurrency. It will be the firms that already control the agents, the compute, and the compliance. Crypto's role becomes infrastructure: useful, necessary, and priced like a commodity.
What this means for market participants
Setting price speculation aside, the structural read is clear. Anyone accepting the machine-native thesis should look toward stablecoin float, tokenized money-market products, and the financing layer behind compute. These are cash-generating businesses, and they are increasingly available inside the regulated framework BlackRock describes.
The broader lesson concerns narrative risk. Each market cycle produces a thesis that sounds like obvious upside. Agents will need crypto is the current version, and it carries enough truth, that machines genuinely need programmable money, to be persuasive. The part that is real accrues to incumbents. The part marketed as a token purchase remains largely unproven. That is a reason to apply additional scrutiny to any asset whose primary justification is that AI requires it.
FAQ
Did BlackRock say AI agents will use stablecoins? Yes. Its September 2026 paper, The Machine-Native Economy, argues that AI agents will increasingly pay for data, API calls, and compute using stablecoins, which it frames as machine-native money.
What is the machine-native economy? A term BlackRock uses to describe a future in which AI agents rather than humans initiate and settle most micro-transactions for digital services, running on programmable, always-on payment rails.
Who benefits if AI agents transact on crypto rails? Primarily the stablecoin issuers that earn on float, the cloud providers that sell compute, the payment networks that process transactions, and the asset managers that structure the resulting products. Decentralized protocols and agent-branded tokens capture comparatively little.
What is the biggest obstacle to AI agents using crypto? Identity and liability. Agents cannot pass customer-verification checks, cannot be sued, and cannot sign credit agreements, so authorization and accountability are handled by regulated off-chain institutions, which is the layer crypto does not solve.
Is agent-and-crypto a viable trade? The more defensible exposure is stablecoin and tokenized money-market infrastructure rather than agent tokens, and the gap between thesis and usage remains wide: AI agents are still under 1 percent of machine-payment volume.