Strategy

191 Million Settlements, 0.6–7.5% Agents: Reading x402's Adoption Numbers Honestly

By IdeaBosqueLast updated: November 9, 2026

Key takeaways

x402 has three independently counted settlement bases — 191.75M transactions as of October 11, 196M+ as of November 1, and 198.9M facilitator-mediated per TRM Labs — with $42–53M in settled value. The rail itself is production infrastructure; AWS Bedrock AgentCore Payments runs on it and Bedrock Bazaar carries 10,000+ x402 endpoints.
TRM Labs screens the $52.7M settled base down to $25.62M of plausible commerce and bounds the agentic share at 0.6–7.5%, with agents moving roughly $5,000–$11,000 a month and 99.6% of settled value in USDC.
Coinbase reported 97% of onchain agentic transactions in Q2 2026 ran on x402 — a share of a category, not a share of the rail's volume; the two readings must not be conflated.
x402 and MPP are diverging rails, not one standard — 191M+ x402 settlements against 46,198 MPP events, with Stripe settling both at minimums of $0.50 (card) and $0.01 (USDC), so any vendor claim should name its rail.
A buyer diligence kit drops out of the pair: which rail, settled-vs-agentic split, authorization provenance (ERC-8004 registry use is voluntary), controls below the value thresholds, and trend over a quarter — 25–30% of raw x402 activity is estimated to be test or leaderboard noise.

x402 has recorded 191,751,598 cumulative transactions across 12 chains as of October 11, 2026 — and Coinbase reported that over 97% of onchain agentic transactions in Q2 2026 ran on x402. But TRM Labs, tracing the same rail through its known facilitators, screens the volume down to $25.62 million of visible commerce and bounds the share that actually looks agentic at between 0.6% and 7.5%, with agents moving roughly $5,000–$11,000 a month. Both sets of numbers are true. This article stacks them against each other, because the difference between them is exactly the gap a buyer needs to probe when a vendor quote says "agentic payments are at production scale."

The pattern generalizes beyond x402. Agentic-payment infrastructure — x402, Stripe and Tempo's MPP, Mastercard Agent Pay, the MCP modules and A2A delegation that route orders into it — now has enough operating history that every vendor claim can be met with a count. The skill is knowing which count answers which question: a settled-transaction count answers "does the rail work," while a screened volume count answers "how much delegated purchasing is actually flowing." A procurement team that cites the first in a budget case and the second in a risk case is measuring honestly. A vendor deck that cites only the first is not necessarily lying — it is answering a different question than the one you asked.

The settlement counters: the rail works

Three independent counts now agree on the scale of the settled base:

Forkast, on Yahoo Finance: 191,751,598 cumulative x402 transactions across 12 chains as of October 11, 2026, per the agenteconomy.to tracker.
Fireblocks: more than 196 million transactions since May 2025, settling over $42 million total as of November 1, 2026, with about 45% on Base.
TRM Labs: roughly $52.7 million across 198.9 million settlement transactions mediated by known x402 facilitators since May 2025, across Base, Solana, and Polygon.

Three counters, one order of magnitude, no contradictions. The transaction counts (191.75M, 196M, 198.9M) climb in date order from October 11 to the facilitator-mediated total; the dollar figures ($42M via facilitators' public counters against $52.7M where TRM can trace both rails' flows) differ by the denominator each counter can see, not by error. The rail itself is settled infrastructure: AWS's own Bedrock AgentCore Payments runs on x402, the AWS Bedrock Bazaar carries over 10,000 x402 endpoints, and the x402 Foundation moved under the Linux Foundation in October 2026 with more than 50 members.

That is the case for the rail: production scale, hyperscaler operationalization, foundation governance. If a vendor says "x402 works," the counters have their back.

What the volume is not: the TRM screens

TRM's September 2026 measurement is the strongest honesty datapoint in the family, because it derives its bound from the onchain record instead of repeating vendor counts. The x402 protocol was built so an agent pays for a resource through an onchain transaction — but nothing in the protocol requires an agent. A scheduled script, a load test, a self-dealing loop, and an agent produce identical onchain records. So TRM screens the $52.7M settled base in three steps: remove addresses paying themselves, remove bulk flows concentrated in one or two payers, and remove sellers with fewer than ten distinct buyers. About half of the settled volume falls away.

What remains — $25.62 million of screened x402 commerce — then gets the harder test: how much of it is an agent? Agency cannot be read off a transaction, so TRM does not assert a single figure. A permissive test counts any payer that could plausibly be an agent (facilitator-broadcast, sub-dollar average, varying amounts); a strict test additionally requires the pattern to persist across months and either ERC-8004 registry registration or payments to more than one seller. The two tests bound the agentic share at 0.6% to 7.5% of screened commerce, with the volume agents actually move at roughly $5,000–$11,000 per month. TRM also flags its own modeling risk: the price-variation assumption may understate genuinely single-purpose agents that pay one service repeatedly.

Two more findings sharpen the picture. First, 99.6% of settled value ($52.47M of $52.68M) was in USDC — the agent-payment economy on these rails is today a stablecoin economy. Second, the number of agents transacting has been volatile through 2026, rising to a mid-year peak and falling back, while the volume they move stays small. Forkast adds a further discount with attribution: BigGo Finance estimates 25–30% of raw x402 activity is likely test or leaderboard noise — activity that pads the transaction counts without a human or an agent buying anything.

The rail split: x402 for scale, MPP for enterprise flows

The comparison piece reading the specs side by side is worth internalizing before reading any vendor claim, because the two stablecoin-native rails that replaced card checkout for agents (ACP retired its in-chat card flow in March 2026) are not converging — they optimize for different things:

	x402	MPP
Steward	x402 Foundation (Linux Foundation, 50+ members)	Stripe and Tempo
Optimization	Scale and open discovery	Session-based, high-trust enterprise flows
Cumulative count	191.75M+ transactions	46,198 events
Governance shape	Open protocol, open endpoints	IETF individual draft, launch partners incl. Anthropic, OpenAI, Shopify, Visa, Mastercard
Volume texture	Sub-dollar per-call, ~$5–11K/month from agents (TRM)	~115,000 transactions and ~$25,000 settled in its first weeks (Visa via Artemis)

The three-layer stack puts these rails in place: checkout (ACP/UCP), authorization (AP2 — now at v0.2 under the FIDO Alliance — plus Visa TAP and Mastercard Agent Pay), and per-request settlement (x402/MPP). Stripe settles both rails into a Stripe balance, with minimums of $0.50 for card tokens and $0.01 in USDC. x402's V2 (December 2025) added wallet-based identity, automatic service discovery, and reusable sessions — sessions are what lowered per-call friction and fueled the transaction counts. And the visibility gap is asymmetric: the string "x402" ran at roughly 667 original X posts per day in 2026 against 2.8 per day for the full "Machine Payments Protocol" name, which means mindshare is not a measure of either rail's economics.

For a buyer, the split matters practically: x402 is where per-call, sub-dollar settlement at volume lives today; MPP is where session-based flows with issuer-grade authorization are being built. A vendor claim about "agentic payments" should name the rail, because the two rails' numbers are not comparable.

How the two readings stack:

The diligence questions: what to ask a vendor that cites 191 million

The counters and the screens pair into a working diligence kit for any agentic-payment claim, whether it arrives in a procurement deck, a partner pitch, or an AI search summary:

Which rail? x402's 191M+ transactions do not transfer to MPP's 46,198 events. A claim that blends them (or omits the rail entirely) has already failed the first check. The rail determines the economics — x402 is per-call and sub-dollar, MPP is session-based — and the authorization layer (AP2 mandates, Agent Pay tokens) rides on top.
Settled or agentic? Ask the vendor to distinguish settlement counts from agent-driven volume. TRM's screen — removing self-payment, bulk flows, and thin-buyer sellers — removed about half of $52.7M before the agency test even began. "191 million transactions settled" is a claim about the rail, and the honest version of the agentic claim is a bound (0.6–7.5%), not a count.
Authorization provenance. Ownership claims onchain are voluntary — TRM found agent-registry declarations (ERC-8004) used by a minority of participants. Ask how a payment's agent is identified and who approved it: protocol-level identity (V2 wallet identity, AP2 v0.2 mandates, Agent Pay for Machines tokens) is verifiable infrastructure; an unattributed address is not.
Controls below the thresholds. TRM's structural finding is that x402 value and volume decouple: payments sit below nearly every value threshold and above nearly every count threshold at once, so controls calibrated for human-scale payments miss them from both directions. If your controller's approval workflow has a $10,000 review floor, an agent making 4,000 sub-dollar calls a week is invisible to it — and the settlement receipts prove payment, not that any screening, policy, or human intentionality ran before the money moved.
Stability over a quarter, not a launch week. Agent counts on x402 have been volatile (spring rise, mid-year peak, fall-back) while per-agent volume stayed small. Forkast's BigGo-sourced 25–30% raw-activity discount applies to any count presented without its noise floor. Ask for the trend, not the peak.

A vendor that answers all five concretely is selling measurement; a vendor that answers only with settlement counters is selling momentum. Both have a place in a market this early — but they belong on different slides.

What this does not settle

The measurement gap is not the same as the liability gap. Even a perfectly attributed agentic payment does not answer who was responsible for the agent's action — the four UK regulators' DRCF joint statement puts that burden squarely on deployers, with auditability as the compliance instrument. And settlement receipts — even x402's signed, portable ones — prove that a payment settled, not which screening version or policy the server applied before settling. That execution-proof gap is what the action_ref binding drafts close; our settlement-audit article maps the receipt composition in detail.