Search for the size of the AI agent payments market and you hit a problem: there is still no single canonical number. What changed in 2026 is that triangulation no longer runs on forecasts alone. There is now measured transaction data, live network programs, and a widening spread of projections whose differences are themselves informative. Here is the honest read.
What is actually measured
Start with observed behavior rather than projections. Salesforce estimates AI and agents influenced roughly 20 percent of global online orders during the 2025 holiday season, about 262 billion dollars of commerce, with 67 billion dollars during Cyber Week alone. Adobe measured AI-referred traffic to US retail sites up 693 percent year over year in that window, still up 393 percent in the first quarter of 2026, and converting 42 percent better than other channels by March.
On the buyer side, roughly one in seven frequent US online shoppers has now let an agent complete a purchase, and Visa's research puts AI usage somewhere in the shopping journey at 47 percent of US shoppers. EY's comfort numbers explain the gap between those two figures: about half of consumers are comfortable with AI recommendations, while less than a third are comfortable with fully autonomous purchases. Influence is mainstream; autonomy is still earning trust.
What the forecasts say
The dedicated market-size estimates are much smaller than the commerce figures above, and that is not a contradiction; they measure different things. Grand View Research sizes the agentic commerce market at about 5.7 billion dollars in 2025, projected to reach 7.7 billion in 2026 and 65.5 billion by 2033. That is the infrastructure and platform layer, not the goods flowing through it.
The larger numbers are about the goods. McKinsey scenarios reach up to 1 trillion dollars of agent-orchestrated US retail revenue by 2030 and 3 to 5 trillion globally. Morgan Stanley is more measured, forecasting about 190 billion dollars of US agentic ecommerce by 2030 in its base case and about 385 billion in its bull case, roughly 10 to 20 percent of US ecommerce. Treat the largest figures as scenario ceilings rather than base cases; their value is showing how much is at stake for whoever holds the infrastructure when adoption compounds.
For context, all of this sits inside an AI spending boom: Gartner put global generative AI spending at 644 billion dollars in 2025. Budget is not the gating factor for agent payments. Trust infrastructure is.
Leading indicators from the rails
The strongest forward signal is that the payment networks are no longer piloting; they are processing. Visa's Intelligent Commerce counts more than 100 partners, completed live agent transactions, and an April 2026 network-agnostic integration layer. Mastercard ran its first live authenticated agentic transactions in Malaysia in March and launched machine-to-machine Agent Pay in June, and its commissioned research projects that one in four online shopping interactions could involve an AI agent by the end of 2026.
The cautionary indicator sits on the crypto side. x402 crossed 100 million cumulative transactions on Base, then saw settlement volume fall more than 90 percent this year as testing and incentive activity washed out. It is a useful reminder that transaction counts are not demand, and that this market's growth curves will be uneven even where the technology works.
Why the estimates diverge
Most of the spread between numbers comes from definitions, not disagreement. Influenced commerce (an agent recommended, a human bought) is hundreds of billions today. Completed agentic commerce (an agent executed the payment) is far smaller and only now becoming measurable through network programs. And infrastructure revenue (what the platforms and rails earn) is the single-digit-billions Grand View figure. Any market-size claim that does not say which of the three it means is not usable for planning.
The forecasts also underweight operations. Growth in completed agentic commerce is gated by three unglamorous constraints: verification quality (proving who acted and what was authorized), spend isolation (capping the blast radius per agent and workflow), and dispute readiness (explaining and challenging transactions quickly). Where those fail, merchants and finance teams withdraw permission, and the curve flattens regardless of model capability.
Influence is mainstream.
Autonomy is still earning trust.
What to track quarterly
If you operate in this market, the top-down numbers matter less than five operational series: AI-influenced order share, autonomous completion rate (purchases finished without a human handoff), approval-to-decline ratio by policy tier, dispute rate on agent-initiated transactions, and margin per completed workflow. Those predict real market capture better than any TAM narrative.
Signets sits on the completed-commerce side of the definitions above: the unified interface agents transact through, across every rail, with the verification, isolation, and records that decide whether the curve keeps compounding. If you are building for that market, tell us what you are building and we will get you sandbox access.
FAQ
How big is the AI agent payments market in 2026?
It depends on the definition. The agentic commerce infrastructure market is estimated around 7.7 billion dollars for 2026 (Grand View Research), while AI-influenced commerce is already hundreds of billions: Salesforce measured about 262 billion dollars of AI-influenced orders in the 2025 holiday season alone. Fully agent-completed payments sit between those figures and are only now becoming directly measurable.
How much online shopping do AI agents influence today?
Salesforce estimates AI influenced roughly 20 percent of global online orders in the 2025 holiday season, Adobe measured AI-referred retail traffic converting 42 percent better than other channels by March 2026, and about one in seven frequent US online shoppers has let an agent complete a purchase.
How large could agentic commerce become?
Scenario work from McKinsey reaches up to 1 trillion dollars of agent-orchestrated US retail revenue by 2030 and 3 to 5 trillion globally, while Morgan Stanley forecasts about 190 to 385 billion dollars of US agentic ecommerce by 2030. The largest figures are ceilings, not base cases, and all of them assume trust infrastructure keeps pace.
What could slow agent payments down?
Operational trust, not model capability: verification of who acted and what was authorized, spend isolation that caps the damage from agent mistakes, dispute readiness, merchant willingness to accept agent traffic, and unresolved liability rules. Weakness in any of these pulls adoption back.
