Autonomous AI agents executed 138 million cryptocurrency payments on Coinbase's Base blockchain through the x402 protocol over 16 months ending September 2026, according to Chainalysis data published October 14. The headline figure represents a trajectory from roughly 1,500 transactions in mid-20...
"You have this perfect combination of speed plus low cost to settle, and that's the sort of thing that you would need if" doing AI-to-AI commerce. — Eric Jardine, Head of Research, Chainalysis
Autonomous AI agents executed 138 million cryptocurrency payments on Coinbase's Base blockchain through the x402 protocol over 16 months ending September 2026, according to Chainalysis data published October 14. The headline figure represents a trajectory from roughly 1,500 transactions in mid-2025 — a 92,000x increase. Over 87% of those payments fell between one-hundredth of a cent and ten cents, with the dominant category being AI model inference access. U.S.-linked wallets accounted for 88.1% of identifiable model-inference spend.
The numbers require context. TRM Labs, analyzing $52.7 million in x402 volume across Base, Solana, and Polygon, estimated that only 0.6% to 7.5% of screened commerce is plausibly genuine agentic AI spend — roughly $154,000 to $1.92 million total. The gap between the transaction count and the verified economic activity is the central question facing this market. Whether x402 represents the foundation of machine-native commerce or an infrastructure searching for real demand depends on which dataset proves more representative.
x402 is an open HTTP payment protocol released by Coinbase in May 2025. Its mechanism is straightforward: a server returns an HTTP 402 ("Payment Required") response with payment specifications; a client agent signs a USDC transfer on-chain; the server verifies settlement and fulfills the request. The cycle executes in milliseconds without bank accounts, credit cards, or checkout flows.
The protocol specification uses payment scheme fields including a network identifier in CAIP-2 format (eip155:8453 for Base), the USDC contract address, a recipient address, a maximum amount, an expiry timestamp, and an optional facilitator endpoint URL. By design, x402 is chain-agnostic and permissionless — any server can return a 402 response and any client can satisfy it.
Five named x402 deployments were running in production by April 2026: Coinbase Agent.market, Stripe Machine Payments, CoinGecko paid endpoints, Circle Wallets reference workflow, and the Cloudflare Agents SDK.
In July 2026, the Linux Foundation operationally launched the x402 Foundation as a standards body with 40 member organizations across three tiers. Premier members include Adyen, Amazon Web Services, American Express, Circle, Cloudflare, Coinbase, Fiserv, Google, Mastercard, MoonPay, Ripple, Shopify, Solana Foundation, Stellar Development Foundation, Stripe, and Visa. General members include Fireblocks, LayerZero Labs, NEAR Foundation, Polygon Labs, Quant Network, and World Liberty Financial, among others.
The membership roster signals something specific: traditional payment networks (Visa, Mastercard, American Express, Adyen, Fiserv) and technology platforms (Google, AWS, Cloudflare, Shopify) are participating alongside crypto-native infrastructure. The protocol is not solely a crypto project.
Chainalysis reviewed a 16-month window ending September 2026, extending its standard 12-month review period due to the protocol's novelty. Key metrics from the report:
| Metric | Value | |---|---| | Cumulative x402 transactions (Base) | 138 million | | July 2026 monthly volume | ~75 million payments ($24 million) | | Payment size (87% of transactions) | $0.001 – $0.10 | | Most common transaction range | $0.01 – $0.10 | | USDC share of transaction value | 99.6% |
Chainalysis categorized agent spending into several identifiable buckets:
Coinbase separately reported 69,000 active agents, 165 million transactions, and approximately $50 million in cumulative volume on x402 by late April 2026, when it launched the Agent.market platform directory of x402-paywalled services.
TRM Labs' analysis introduces substantial uncertainty into the headline figures. The firm analyzed 198.9 million individual settlements across Base, Solana, and Polygon from May 2025 onward, filtering down to $52.7 million in cleaned transaction data.
After removing self-payments, bulk flows, isolated sellers, sporadic meme-token activity, and irregular flow patterns, TRM estimated genuine agentic AI spend at 0.6% to 7.5% of screened volume — between $154,000 and $1.92 million total. The implied 2026 run-rate was $5,000 to $11,000 per month in actual agentic dollar volume.
The firm's methodology weighted facilitator-broadcast payments maintained consistently over several months more heavily than one-off patterns. TRM acknowledged a structural limitation: non-AI scripts can replicate identical transaction sequences, making definitive classification difficult. On-chain agent registries remain voluntary and sparsely populated, which constrains data reliability.
A separate 35-day study tracking 6.4 million transactions across agent payment protocols processed just $119,947 in total value. Some 90.8% of all recorded agent payments were microtransactions worth less than one cent.
The gap between Chainalysis' 138 million transaction count and TRM's economic value estimates is not necessarily contradictory — Chainalysis counts all x402 transactions while TRM applies filters for genuine economic activity. But the discrepancy is material. If TRM's upper-bound estimate of 7.5% holds, approximately 128 million of those 138 million transactions may not represent genuine autonomous commerce.
The wallet behavior data is less ambiguous than the transaction value data. Chainalysis tracked several behavioral indicators:
| Metric | Q3 2025 | Q2/Q3 2026 | |---|---|---| | Wallets receiving second funding within 30 days | 70.6% | 89.3% | | Median wallet age at first x402 payment | 73 days | 18 minutes | | Wallets spending within one hour of funding | 4% | 67% |
The compression from 73 days to 18 minutes in median time-to-first-spend suggests purpose-built wallets for automated commerce rather than human users experimenting. The 89.3% re-funding rate implies developers maintaining ongoing agent operations, not conducting one-off tests.
These patterns are consistent with programmatic deployment: create wallet, fund wallet, deploy agent, agent spends. The data does not, however, distinguish between a developer running a productive agent and a developer running a test script in a loop.
The agentic payments infrastructure has attracted parallel efforts from multiple companies:
Coinbase operates x402 and Base, the primary network. It launched Agent.market as a directory of x402-paywalled services and runs an accelerator for agent and payments startups.
Stripe launched its Machine Payments Protocol (MPP) in March 2026, supporting both stablecoins (via Stripe's Tempo chain) and fiat within a single protocol. Stripe reported that transactions grew 38-fold and payment volume grew 20-fold for AI agents using its Link payment service in September 2026, though it declined to disclose absolute figures. Approximately 200 agents were using the service at the time of reporting, including Meta's Muse and Grok Bot.
Google released the Agent Payments Protocol (AP2) in partnership with Coinbase and the Ethereum Foundation. AP2 sits atop Google's Agent-to-Agent (A2A) communication framework: A2A handles agent negotiation, AP2 handles settlement. The A2A x402 extension enables stablecoin payments within this framework.
Amazon launched AgentCore Payments with Coinbase and Stripe integration.
Visa, Mastercard, and American Express joined the x402 Foundation as premier members but have not publicly disclosed agent payment volume through their networks.
The IMF published a research note titled "How Agentic AI Will Reshape Payments" (IMF Notes Volume 2026, Issue 004), acknowledging the policy implications of autonomous machine commerce.
The geographic concentration is stark. According to Chainalysis:
| Category | U.S. Share | |---|---| | AI model inference payments | 88.1% | | Agent frameworks and tooling | 41.8% | | Agent-to-services (overall) | 30.0% | | Data and search APIs | 23.2% |
China-linked wallets represented 0.1% of model inference payments but approximately 90% of security and code-analysis services, predominantly through AgentLisa, an AI smart contract auditing platform.
Chainalysis head of research Eric Jardine attributed the Chinese pattern to open-weight models: "Open-weight models like DeepSeek and Qwen reduce Chinese developers' need for metered inference services." Chinese developers access model capabilities locally rather than paying per-query for inference on U.S.-hosted models, but purchase specialized security tooling.
The regulatory dimension is undefined. USDC comprises 99.6% of x402 transaction value. If agents are transacting autonomously in dollar-denominated stablecoins, questions about money transmission, sanctions screening, and tax reporting on automated micro-transactions remain unaddressed. The GENIUS Act (P.L. 119-27), enacted in July 2025, established a framework for payment stablecoin issuers but did not contemplate machine-initiated autonomous spending. The FDIC, OCC, and Federal Reserve have all issued proposed rules implementing the Act, none of which address agent commerce specifically.
The economic value distribution in agentic commerce follows a pattern recognizable from broader blockchain fee analysis. For each x402 transaction, value fragments across:
At $24 million in July 2026 monthly volume across 75 million transactions, the average transaction value was $0.00032. Base network gas fees at sub-cent levels mean network revenue from agent transactions is minimal in absolute dollar terms, even at 75 million monthly transactions.
The economic value proposition is not in per-transaction fees. It is in volume aggregation and ecosystem lock-in. Coinbase captures value through Base network activity, USDC velocity, and platform positioning. Stripe captures value through MPP settlement fees. API providers capture value through micropayment-enabled service monetization.
The Chainalysis data showed the broader North American crypto economy at $3.02 trillion in on-chain activity for the 12 months ending July 2026. Agent commerce, even at Chainalysis' headline figures, represents a fraction of one percent of that total. At TRM's filtered estimates, it is a rounding error.
The 138 million transaction figure and the TRM Labs 0.6-7.5% genuineness estimate are not competing narratives — they describe different aspects of the same phenomenon. The infrastructure is being built at scale. The institutional commitment is real: 40 organizations including the largest payment networks and technology platforms joined a standards body within three months of its formation. The protocol architecture works.
What remains unproven is whether autonomous AI agents will generate economically meaningful transaction volume — payments that represent genuine value exchange rather than developer experimentation, scripted activity, or self-dealing. At $24 million in July monthly volume across 75 million transactions, with an average transaction size of $0.00032, the current economic footprint is negligible relative to the $3.02 trillion North American crypto economy.
The infrastructure is ahead of the demand. That is not unusual in payments technology — card networks, ACH, and wire transfer systems all preceded the transaction volumes that eventually justified them. The question is whether AI agent commerce follows the same trajectory or whether the 138 million transaction count reflects a protocol stack generating its own artificial activity. The data, as of October 2026, supports both interpretations.