# According to Visa, AI now buys on its own: more than 109 million machine-to-machine transactions

> Visa and Artemis reveal that AI agents now buy on their own: more than 109 million machine-to-machine transactions
> By Covadonga Fernandez · AI Agents · 2026-07-16

Commerce between artificial intelligence agents has begun to become a reality. That is the conclusion of the joint report Agentic Payments from the Ground Up, produced by Visa and blockchain analytics firm Artemis, which argues that the x402 and Machine Payments Protocol (MPP) protocols are already processing real transactions and that a new infrastructure for automated machine-to-machine payments is beginning to consolidate.

## AI now buys on its own

The study analyzes the evolution of so-called agentic commerce, a model in which artificial intelligence agents can discover services, compare prices and make payments autonomously to access data, computing capacity, software tools or other digital resources.

The report's main conclusion is that commerce between agents is beginning to take shape. x402 and Machine Payments Protocol are already processing real transactions; merchants and agent platforms are beginning to develop services on these protocols, and the standards ecosystem around them is starting to converge toward a common set of needs: trust, identity, authorization and settlement.

**[The x402 Foundation launches under the Linux Foundation to integrate payments into the internet](https://observatorioblockchain.com/en/technology/x402-se-lanza-en-linux-foundation-para-integrar-pagos-en-internet/)**

For now, the data collected by Artemis shows that x402 is the protocol with the most activity within this new ecosystem. Initially created by Coinbase and Cloudflare and currently under the governance of the Linux Foundation, the protocol has recorded since its launch in May 2025 approximately: 109.6 million transactions, 15 million dollars in adjusted volume, more than 422,000 unique buyers and about 5,300 active sellers

The [report](https://www.visa.com/en-us/thought-leadership/innovation/agentic-payments-from-the-ground-up) also shows that most of this activity is concentrated on Base, followed at a considerable distance by Solana and Polygon. For Visa, these data indicate that payments between agents are no longer merely a technological test, but an activity that is beginning to generate a significant volume of real operations.

## MPP: connecting blockchain and traditional payment systems

The second protocol analyzed is Machine Payments Protocol (MPP), developed by Stripe and Tempo with contributions from Visa and presented in March 2026. According to the report, during its first weeks of operation it had settled approximately: 115,000 transactions, 25,000 dollars in volume, about 2,800 buyers and around 90 sellers

Although its activity is still very small compared with x402, Visa highlights that MPP is born with a different approach. While x402 is oriented toward native machine-to-machine payments on blockchain, MPP has been designed to support different settlement mechanisms, including both payments via stablecoins and payments using tokenized payment credentials, which represent traditional payment methods without directly exposing the user's original data.

The report does not state that MPP will use tokenized bank deposits or that it will replace stablecoins. What it points out is that the protocol is designed to work with different settlement infrastructures, both inside and outside blockchain.

## Two types of agent-driven commerce

Visa distinguishes two clearly different categories within commerce between agents. The first corresponds to macrocommerce, where an agent acts on behalf of a person to book a trip, manage a subscription or make purchases similar to traditional e-commerce.

The second is microcommerce, made up of thousands of small payments that one piece of software makes directly to another to access digital resources. This second group includes operations such as API queries, access to databases, use of artificial intelligence models, image generation, purchase of computing capacity or use of specialized tools.

Many of these operations cost just a few cents or even fractions of a cent, but a single agent can execute hundreds or thousands of them during a single task.

## AI creates new demand for payments

The report argues that the real change comes not only from blockchain, but from the advance of artificial intelligence. Until recently, models could recommend actions but not complete a purchase. The new agents are capable of locating a service, evaluating its price, deciding whether it is worth using and making the payment automatically to continue with their work. According to Visa, this capability generates for the first time real demand for programmatic machine-to-machine payments.

The study recalls that in 1997 the designers of the HTTP protocol reserved code 402 Payment Required with the idea that one day paying for content and services on the internet would be as simple as loading a web page. That scenario never arrived because the fees of traditional systems made micropayments unviable. Now, the combination of AI agents and blockchains with very low settlement costs makes this type of operation economically possible.

Trust remains the main challenge

Although the technology is advancing rapidly, the report identifies several still-open problems. Such as who is responsible if an agent makes a wrong purchase. How to resolve disputes when thousands of automatic micropayments are involved, who bears responsibility if an agent is manipulated through an attack, or how to manage chains of agents that pay other agents to complete the same task.

According to Visa, current claims and refund systems were designed for purchases made by people and not for software networks capable of executing thousands of operations per hour.

## Standards begin to converge

In this regard, the report notes that different standards aimed at providing identity, authorization, trust and payment mechanisms for AI agents are beginning to appear. Among them it cites x402, Machine Payments Protocol (MPP), Trusted Agent Protocol (TAP) and Visa Intelligent Commerce.

The payments multinational believes that these developments will not necessarily compete with each other, but could end up complementing each other to build the infrastructure of future automated commerce.

Another of the report's conclusions is that the evolution of payments between agents does not involve choosing between cards or stablecoins. Visa argues that both models can play different roles. According to the report, agents will mainly use cards to make larger purchases on behalf of consumers or companies, while stablecoins offer advantages for automated micropayments thanks to their low settlement costs.

The study also notes that protocols capable of working with different payment mechanisms are beginning to emerge, pointing toward a future where blockchain infrastructures and traditional systems can coexist.

## A new infrastructure for the AI economy

For Visa, commerce between agents has ceased to be a hypothesis, since the data analyzed by Artemis shows that operational protocols, real transactions and companies building services on this new infrastructure already exist. Although the economic volume is still small compared with the global payments market, the report concludes that the ecosystem is beginning to organize itself around four fundamental elements: identity, trust, authorization and settlement.
