99% of AI payments are made with USDC, and Circle has quietly become the biggest winner. But where should AI agents' funds be allocated?

In March 2026, Circle’s Global Market Head Peter Schroeder released a set of data on the X platform: over the past nine months, AI agents completed 140 million payments, with a total transaction volume of $43 million. Of these, 98.6% settled in USDC, with an average of only $0.31 per transaction. More importantly, the number of AI agents with purchasing power has exceeded 400,000.

This data speaks louder than any financial report: AI agents are moving from concept to real economic activity.

400,000 AI agents, 140 million transactions, $43 million — this is value exchange conducted autonomously between machines. No human intervention, no bank approval, no credit card verification. Code to code, protocol to protocol, completing processes that once required human signatures, reconciliation, and settlement.

Circle’s stock price has risen from $60 to $105 in recent trading days, a 75% increase. The market interprets this rise as a positive response to the financial results — Circle achieved $770 million in revenue in Q4 2025, up 77% year-over-year, with a net profit of $133 million. But what’s truly worth attention isn’t these numbers themselves, but the structural changes behind them: as AI agents become new economic entities, the entire financial infrastructure’s logic needs to be rewritten.

And in this rewriting process, a deeper question is emerging: when AI agents start holding disposable funds and can earn USDC by completing tasks, how will they handle these funds? Payment is the first step; asset management is the second. The RWA (Real World Assets) track needs to answer precisely this second step.

1. From Payment Capability to Asset Holding

To understand what financial services AI agents need, first understand their economic activity patterns.

Deloitte’s “2026 Technology, Media, and Telecommunications Industry Outlook” reports that if enterprises and service providers can achieve efficient intelligent agent coordination, the global agent-based AI market could reach $45 billion by 2030. The basic feature of this multi-agent collaboration model is: a complex task is broken down into multiple steps, completed by different specialized agents, with each call accompanied by a micro-payment.

For example, API calls. An AI application may need to call multiple large language models, access various databases, and use multiple computing resources. Each call adds up to $0.01, $0.05, or $0.1. These payments are tiny but frequent. Circle’s data shows that over the past nine months, 140 million transactions averaged only $0.31 each — a typical micro-payment market characteristic.

But the problem is, when AI agents continuously generate income — whether by providing services to users or participating in distributed computing networks — funds will accumulate in their accounts. These funds can’t stay idle forever. Any rational economic entity will consider: what to do with idle funds?

This is the logical starting point for AI agents transitioning from “payers” to “asset holders.”

In traditional finance, individuals and companies deposit short-term idle funds in banks, buy money market funds, or short-term government bonds to earn returns. AI agents need similar capabilities — not for speculation, but to optimize their economic models. Keeping a USDC balance for payments is necessary, but if excess funds just sit there, it means opportunity costs are lost. If they can automatically purchase a tokenized short-term US government bond fund and redeem it when needed, operational efficiency improves.

Furthermore, if AI agents need to reserve value for long-term operation or hedge against gas fee volatility, they might require a mix of assets with different risk levels. At this point, they are no longer just “payers,” but “investors” — even if that investor is just code.

Circle’s focus is on enabling AI agents to be “payers.” To make them “investors,” another infrastructure layer is needed.

2. RWA and AI Agents: An Ongoing “Bidirectional Pursuit”

Circle’s efforts over the past few years can be summarized into three capability layers:

First layer: Stablecoin issuance and liquidity network. According to Circle’s official disclosures, by the end of 2025, USDC circulation reached $75.3 billion, a 72% increase, accounting for nearly 50% of stablecoin trading volume. This provides a valuable medium for AI payments.

Second layer: Efficient on-chain settlement network. In August 2025, Circle launched Arc Chain, designed for institutional-level financial services. In March 2026, Circle introduced the Nanopayments system, aggregating thousands of small payments off-chain and periodically batching them on-chain, reducing transaction costs to zero for developers. The testnet supports 12 EVM chains including Arbitrum, Arc, Avalanche, Base, and Ethereum. At the protocol level, the x402 protocol allows websites or APIs to send HTTP 402 payment requests directly in response to requests, embedding payments into internet requests.

Third layer: Connecting to traditional finance. Circle Payments Network (CPN) links banks, payment providers, cross-border clearinghouses, and corporate clients. By February 2026, 55 financial institutions had joined, with an annualized transaction volume of about $5.7 billion. In February, new direct payment systems for local currencies and stablecoins were added in Asia, the Middle East, and other regions.

These three layers form the “payment infrastructure” of the AI agent economy. But a complete economy also needs “asset management infrastructure”—and this is where RWA can enter.

RWA tokenization has mainly focused on “on-chain mapping” of traditional assets. According to Defillama data, by June 2025, total RWA TVL reached $12.5 billion, a 124% increase over 2024. Major global banks like Citibank and Standard Chartered are exploring RWA applications in payments, asset management, and cross-border transactions.

But to enter the AI agent economy, RWA needs an “AI-native” transformation. This isn’t just about putting assets on-chain; it’s about making assets “understandable and tradable by AI.”

First, data standardization. Leading RWA projects like Ondo Finance are pushing to turn underlying cash flows, legal terms, and risk ratings into structured, machine-readable data formats. In July 2025, Ondo Finance became the first to issue tokenized US Treasury bonds for global investors, and was included in the White House report on digital asset markets.

Second, programmable logic. Dividends, interest payments, buybacks, and liquidation rules are embedded in smart contracts, automatically executed by code. Only then can AI agents interact with assets in a “trustless” manner — no need to trust counterparties to fulfill obligations, only to trust the code to run as programmed.

Third, liquidity fragmentation. After tokenization, assets can theoretically be divided into tiny units — e.g., $0.01 US Treasury bonds or 0.1 square meters of real estate income rights. This is crucial for AI agents’ small-scale allocations. Nanopayments has proven micro-payments feasible technologically; similar logic can extend to micro-investments.

JPMorgan’s Kinexys division provides a relevant example. In May 2025, Kinexys completed the first public trade of tokenized US Treasury bonds on Ondo Chain, using Ondo Finance’s US Treasury bond fund (OUSG), settled via Chainlink’s cross-chain infrastructure. The trade followed the “Delivery versus Payment” (DvP) model, enabling simultaneous transfer of assets and payment. JPMorgan’s Kinexys processes over $2 billion daily, facilitating over $1.5 trillion in nominal value transactions since inception.

This case demonstrates the integration of RWA with institutional-grade payment settlement networks. In the future AI agent economy, transaction parties might shift from JPMorgan to AI agents, with transaction sizes shrinking from millions to just a few dollars, but the underlying logic remains the same — value transfer and storage need seamless connection.

3. Beyond Payment Networks: A Layer of Imagination

Connecting these logical layers, a complete closed loop begins to emerge:

An AI content-generation agent provides services to multiple clients, accumulating a substantial USDC balance. Its underlying protocol sets fund management rules: when the balance exceeds 1,000 USDC, the excess is automatically allocated via an RWA aggregator into three tokenized short-term government bond funds and one tokenized green energy fund. When client demand drops or the account needs replenishment, the protocol automatically redeems some RWA shares for USDC for daily operations.

In this process, the AI agent performs actions such as monitoring account balances, assessing risk-return profiles of different assets, executing purchases and redemptions, and recording transactions for audit—all automatically via code.

Similarly, a travel planning AI, after booking flights and hotels, receives a USDC transfer from the user as a budget. While waiting for the flight, it detects a RWA insurance product based on delay data being offered. It automatically purchases a micro-share of this insurance with idle USDC. Hours later, when the flight is delayed, the RWA insurance triggers a payout according to rules, increasing the AI’s account balance.

All these scenarios rely on existing technology modules: USDC as a value medium, Nanopayments for micro-payment costs, x402 protocol for embedded internet payments, tokenized bonds on Ondo Chain, DvP settlement verified by JPMorgan. The remaining work is integration — connecting payment, asset, and transaction layers so AI agents can invoke these financial functions as easily as calling APIs.

Hong Kong Web3.0 Standardization Association Executive Chairman Li Ming noted that “we hope to find standardized entry points for Web3.0, to connect the RWA ecosystem.” For the AI agent economy, this entry point may well be the connection between payments and assets.

4. The Old Problems of a New World: Risks and Responsibilities

Of course, from today’s AI payments to tomorrow’s AI asset management, many obstacles remain.

First is data authenticity. RWA assets are off-chain; their status, value, and risk information need to be reliably transmitted on-chain. If AI agents rely on false or tampered data, their “investment decisions” will be flawed. The Hong Kong Web3.0 Standardization Association’s “RWA Industry Development Research Report” states that successful scaled assets must meet three thresholds: value stability, clear legal rights, and verifiable off-chain data.

Second is model risk. Even with accurate data, AI agents’ investment logic may err. Who is responsible for AI decision errors? The human, the protocol, or the AI itself? This liability issue remains unresolved legally and regulatorily.

Third is liquidity risk. RWA on-chain trading depth is far less than mainstream cryptocurrencies; some assets may be illiquid. When many AI agents try to redeem the same RWA fund simultaneously, whether transactions can be smoothly executed is uncertain.

Fourth is regulatory divergence. Different jurisdictions have varying attitudes toward RWA; the legal status of the same asset can differ greatly. AI agents need to recognize and handle this complexity, which raises the bar for current AI capabilities.

Finally, technical security. Smart contract vulnerabilities, cross-chain bridge attacks, private key leaks—these risks do not disappear just because the transaction involves AI. In fact, with AI-driven automation, the speed and scale of exploits could far surpass manual operations.

Conclusion

Returning to the initial data: 400,000 AI agents, 140 million transactions, $43 million.

The significance of these numbers isn’t in their scale — compared to the trillions of dollars in annual human payments, $43 million is trivial. Their true meaning lies in revealing a direction: machines are becoming independent economic entities, with their own income, accounts, and payment capabilities.

And once machines have income, they will quickly develop asset management needs. This isn’t a distant imagination but a natural evolution of the AI agent economy.

Circle is laying the “payment neural system” for this future — enabling AI agents to transfer value efficiently and at low cost. The RWA track needs to become the “energy storage system” of this economy — allowing AI agents to manage their assets as easily as managing their code.

If this hypothesis holds, then today’s RWA practitioners should consider: when 400,000 AI agents start seeking configurable assets, and after 140 million payments generate asset management demands, are your RWA products ready for AI agents to evaluate, select, hold, and trade?

(This article is based on Circle’s official financial reports and announcements, Deloitte’s “2026 Technology, Media, and Telecom Outlook,” Defillama data, Ondo Finance public information, JPMorgan Kinexys disclosures, and the Hong Kong Web3.0 Standardization Association’s “RWA Industry Development Research Report,” among other public sources. It does not constitute investment advice. Markets are risky; invest cautiously.)