BlackRock’s 11-page Machine-Native Economy whitepaper presents a possible financial architecture in which autonomous software uses stablecoins to pay for data, computing capacity and application programming interfaces (APIs). The paper was written by leaders from BlackRock’s digital-assets, iShares and product-innovation teams.

The publication is a research thesis, not an announcement of a stablecoin, payment service, investment fund, machine-payment protocol or commercial partnership. It also does not establish that AI agents are already conducting payments at scale. Its significance lies instead in mapping, from an incumbent asset manager’s perspective, the infrastructure autonomous software would need in order to transact.

Software becomes the customer

Most digital payments assume that a person selects a product, enters payment details and approves a purchase. AI agents introduce a different type of customer: software that may need to buy small amounts of data, processing capacity or API access while completing a broader task.

For example, an agent conducting an extended financial analysis could estimate the computing power required, compare cloud providers by price and performance, rent capacity for one job and stop paying when the work is complete. A research agent could purchase a single database result, while a software agent could pay to use a specialized function.

Rather than maintaining subscriptions with every potential provider, an agent could purchase each resource when it is needed. That continuous consumption creates a payment challenge. A process involving hundreds of small purchases requires prices, spending permissions and settlement instructions that software can understand without opening a conventional checkout page for every transaction.

Stablecoins could handle small payments

BlackRock expects stablecoins to lead this transactional layer because their value usually tracks a currency such as the US dollar. An agent could compare a quoted price with its approved budget without accounting for the price volatility of Bitcoin or Ether.

Stablecoins can also settle throughout the day and support payments small enough to correspond to individual API calls or short periods of computing time. Protocols such as x402 can place a payment request inside the same online interaction used to access the underlying service. Coindoo’s guide to AI-native stablecoin payments explains how x402 and related protocols coordinate these transactions.

BlackRock raises a separate question: Could the computing capacity purchased through such systems eventually become a tradable financial asset?

Different purchases would continue to use different payment methods. BlackRock expects cards and modified bank rails to remain important when agents transact with consumer-facing businesses. Their existing acceptance, fraud controls and dispute procedures remain valuable for larger purchases, including travel and physical goods.

Permission remains outside the blockchain

Settlement explains how an agent transfers money. Authorization determines whether the agent had permission to spend it.

A valid blockchain transaction proves that the correct cryptographic key approved a payment. The transaction alone cannot establish who owns the agent, whether its instructions were legitimate or whether the purchase remained within an approved budget.

BlackRock uses the emerging term “know your agent,” or KYA, for controls that connect an AI agent with a verified person or company and a defined set of permissions. KYA is not yet a single, universally adopted compliance standard comparable with established know-your-customer requirements. The distinction matters because current compliance frameworks verify people and companies; they were not designed to vouch for software acting on a customer’s behalf.

The paper expects identity, anti-money-laundering and authorization checks to remain largely offchain. A verified result could then be passed to the blockchain when the system determines whether a transaction is eligible.

A functioning agent-payment system would need to establish:

• Who owns or controls the agent
• Which services it may purchase
• Its limit for each transaction
• Its total budget over a defined period
• Who is responsible when a service fails

Stablecoins can automate settlement after those decisions are made. They cannot determine whether the underlying purchase was appropriate.

Compute could become a tradable claim

Once agents can pay for computing power, the resource itself becomes the next question in BlackRock’s argument.

A cloud provider could sell the right to use a defined amount of capacity at a future date. A company expecting a large AI workload could buy that right in advance, securing access before demand or prices increase.

If those usage rights became sufficiently standardized, they could be represented digitally and transferred between holders. If realized, that would extend digital-asset markets beyond money-like instruments to standardized claims on physical infrastructure. BlackRock suggests that compute contracts might eventually support several financial functions:

• Locking in future processing capacity
• Hedging against increases in compute prices
• Trading rights linked to specific hardware or regions
• Pledging compute claims as collateral
• Settling contracts through programmable networks

Compute is harder to standardize than money. One hour on a newer GPU can produce more work than an hour on an older model. Electricity costs, latency, hardware availability and local regulation also vary between regions.

Any functioning compute market would need contracts that define the hardware, location, performance and delivery terms precisely. BlackRock views these differences as design problems that financial markets may eventually address through region-specific contracts, futures and other hedging instruments.

The figures show scale, not AI-agent adoption

BlackRock supports its argument with figures covering stablecoins and the broader computing market. These figures establish that large settlement and infrastructure markets already exist, but they do not show how much activity currently comes from autonomous agents.

The payment figures use different definitions. Stablecoin activity includes financial transfers that do not resemble ordinary consumer spending, while Visa, Mastercard and ACH publish figures based on their own network methodologies. The totals should therefore be used to understand scale rather than rank competing payment systems.

BlackRock also says adjusted stablecoin activity grew at an annualized rate of approximately 80% between 2020 and 2025, compared with roughly 8.5% for ACH. Stablecoins began from a much smaller base, so the faster percentage growth does not establish that they are overtaking the bank-transfer network.

Network use is not automatic token demand

Frequent stablecoin payments could increase demand for blockchain processing, blockspace and validator services. The effect on any native cryptocurrency would depend on the network’s fee model, staking design and use of sponsored transaction costs.

An application may pay gas on behalf of its AI agents, meaning the agents themselves would not need to hold the network token. Another participant would still have to pay the underlying fee.

BlackRock cites Circle’s Arc as an alternative design in which USDC is intended to serve as the network’s gas asset. Under that model, payment activity could increase the utility of the stablecoin without creating a separate token requirement.

The machine economy remains at an early stage

Agent payments remain small. The paper describes an emerging market rather than established mass adoption.

Stablecoin volume is not the same as agent spending. Existing activity includes trading, lending and other financial transfers. A wallet signature is not proof of intent, while identity, ownership and spending permission require additional controls.

Compute contracts also remain theoretical. Standards for hardware quality, delivery and regional pricing still need to develop. Blockchain activity may not reach every token equally, and value capture depends on each network’s economic design.

BlackRock’s publication announces no stablecoin, machine-payment protocol, commercial partnership or fund holding compute contracts. It sets out the firm’s view of where AI and digital assets may intersect as autonomous software assumes a larger role in economic activity.

Stablecoin payments are only the entry point

BlackRock’s paper is most significant for treating AI operating costs as a possible future digital-asset market. Stablecoins provide the proposed payment method, while standardized claims on computing capacity represent the more ambitious idea.

Cloud revenue projections show that compute is becoming a large economic resource. They cannot prove that AI agents will buy it autonomously or that those purchases will settle onchain. Actual agent spending, along with the controls governing it, will determine whether BlackRock’s machine-native economy develops beyond a research thesis. Concretely, that means watching whether KYA matures into a recognized compliance standard, whether compute contracts acquire shared specifications for hardware, delivery and pricing, and whether payment data begins isolating agent-driven transactions from trading and lending flows.

This article is provided for informational purposes only and does not constitute financial or investment advice. BlackRock’s paper discusses possible future developments and does not announce or recommend an investment product.