Blockchains to Become Settlement Rails for AI Agents: BlackRock

Key Insights

• BlackRock published a research paper arguing that AI agents will create structural demand for digital assets, framing AI as machine-native intelligence and digital assets as machine-native money.
• The paper sets out three convergence points: machine-to-machine payments, programmable tokenized assets, and computing power as a tradeable, collateralizable asset.
• It cites more than $11 trillion in adjusted stablecoin transaction volume during 2025, while cautioning that the figure is not directly comparable with traditional payment networks.
• Named payment protocols include x402, the Machine Payments Protocol from Stripe and Tempo, and the Agentic Commerce Protocol from Stripe and OpenAI.
• BlackRock does not put a timeline on any of it, and flags unresolved problems in hardware standards and agent identity.

The world's largest asset manager thinks blockchains are about to get a new customer base, and it is not human. BlackRock published a research paper on 22 September 2026 arguing that autonomous AI agents will need financial rails built for machines, and that digital assets are the natural fit.

The paper, titled "The Machine-Native Economy: How digital assets connect intelligence, commerce, and compute," makes a simple pairing: AI supplies machine-native intelligence, and digital assets supply machine-native money.

Its underlying argument is about shared architecture. Large language models turn human language into tokens a machine can process. Blockchains turn economic rights into tokens a machine can verify and settle. BlackRock's case is that the two systems already speak the same structural language.

Machine Payments Need Machine Money

The first convergence point is payments. As AI systems complete tasks on their own, they need to pay for data, software, application programming interfaces (APIs), and computing power, often in amounts too small and too frequent for a human to approve each one.

BlackRock points to emerging protocols built for exactly that: x402, the Machine Payments Protocol (MPP) developed by Stripe and Tempo, and the Agentic Commerce Protocol (ACP) from Stripe and OpenAI. Each is designed to let software agents initiate and settle payments without a person signing off on every transaction.

Stablecoins are the settlement instrument in that picture. They move on a blockchain while holding a value pegged to a currency such as the US dollar, which makes them usable by a machine that needs certainty about what it is paying.

The paper cites more than $11 trillion in adjusted stablecoin transaction volume during 2025, alongside regulatory frameworks now in place including the US GENIUS Act, the European Union's MiCA rules, and regimes in Hong Kong and Singapore.

Tokenized Assets Become Programmable

The second scenario moves from payments to portfolios. Tokenized real-world assets (RWAs) put financial instruments on a blockchain, where a smart contract can act on them directly.

That matters for AI because it removes the human step. An agent managing funds could execute trades or move collateral through programmable infrastructure rather than submitting instructions and waiting for a person or a settlement cycle to process them.

Computing Becomes Collateral

The third idea is the least developed and the most striking. BlackRock suggests rights to computing power, including graphics processing units (GPUs), could be represented as transferable digital contracts: traded in real time, or pledged as collateral.

The paper points to the Model Context Protocol (MCP) and Agent2Agent (A2A) as the kind of standards that could route those transactions.

BlackRock is direct about how far off this is. Standardized contracts and liquid markets for compute do not yet exist, and the firm says hardware metrics need standardizing and digital identity for autonomous agents needs resolving before any of it reaches commercial scale.

What the Paper Does Not Say

Three caveats deserve as much attention as the thesis. The $11 trillion stablecoin figure comes with BlackRock's own warning that the methodology is not directly comparable with traditional payment-network volumes.

Adjusted on-chain volume strips out some automated and internal transfers, but it is not the same measure as card-network throughput. There is no timeline. The paper describes what could develop, not when, and the compute section in particular reads as a sketch of a market rather than a description of one.

As such, it is important to consider that BlackRock is not a neutral observer. The firm runs some of the largest crypto exchange-traded products in the world and a tokenized money market fund, so a thesis in which machine demand expands the use of stablecoins and tokenized assets is one that suits its own product line.

None of that makes the argument wrong. What it does mean is that the paper is best read as the world’s largest asset manager explaining why it is building in this direction, rather than as a forecast.