Payments 3.0: AI Agent Payment Marketplace

Author: Ekko, Ryan Yoon
Source: tiger-research
Translation: Shan Ouba, Jinse Caijing

1. The payment infrastructure of the agent era has begun to take shape

Over the past year, major global tech companies, card organizations, and exchanges have all released their respective agent payment standards (hereafter referred to as “agents”). In 2025 alone, eight standard protocols have been published, with more cooperation announcements to follow.

During this structural transformation period, controlling payment standards is crucial. In an economy primarily based on offline transactions, Visa and Mastercard established the card payment standards and dominated the market. All card transactions then ran through their networks.

As commerce shifted online, a new wave of participants emerged. PayPal built its online payment service based on email remittance, followed by Stripe. The next payment market will be dominated by agents. With the proliferation of AI, the arrival of the agent era is now widely acknowledged.

When people think of AI-powered agent payments, they usually envision two scenarios: one is agents acting on behalf of users to find and purchase products; the other is direct transactions between agents without human intervention.

These represent the near-term and longer-term futures, respectively. Orders executed by agents based on user instructions belong to the general agent business model, which has already begun to take shape. Direct payments between agents fall under a pay-per-use model, representing a more distant future.

These two types of participants may seem opposed, but in reality, they are solving different problems. This report will explore how these two categories are establishing standards in different fields within the agent payment industry.

2. General agent business

In the realm of general agent commerce, shopping behavior is delegated to agents. On specific platforms, users register a membership card and set delegation scopes; then, agents execute shopping operations within that platform.

For example, a user might tell an agent: “Please arrange my business trip to Tokyo next week, with a budget of 2 million Korean won (about $1,400).” This instruction grants the agent conditional payment authority. Within the budget, the agent uses the user’s bank card to sequentially select and pay for flights, hotels, airport transfers, currency exchange, and travel insurance.

To ensure this process runs smoothly, the agent must first interpret the user’s intent, find suitable products, and then securely complete payments. This structure involves two layers:

• Discovery Layer: The agent searches for products on behalf of the user within the platform.
• Payment Layer: The agent makes payments within the scope set by the user.

Some leading players focus on a single layer, while others aim to cover both simultaneously.

Alphabet Inc. (GOOG)

Core Technologies

Google is working to control both discovery and payment layers, based on two standards: UCP and AP2.

UCP (Universal Commercial Protocol) is a standard for communication between agents and merchants.

For an agent to handle shopping on behalf of a user, it must interact with various services. The problem is that each service’s structure differs. Every time an agent discovers a new service and initiates a transaction, it requires separate integration. Google aims to eliminate this inefficiency through a unified communication platform (UCP).

Once merchants are configured according to the standard, any agent can connect to them using the same method.

AP2 (Agent Payment Protocol) is the authoritative standard that guarantees clarity on “who authorized what and the maximum authorized amount” during the transition from discovery to payment.

When a user presses the payment button directly, the operator and responsible party are clear. But when an agent pays on behalf of a user, the scope of authorization and accountability become fuzzy. AP2 records the user’s initial instructions as an tamper-proof digital contract (authorization). The agent can only act according to the user’s instructions. After the transaction, the system retains a traceable record of who authorized what and when.

In short, if UCP is the discovery standard, then AP2 is the standard that enforces accountability in transactions.

Core Business

Google’s current revenue mainly comes from two pillars: advertising and cloud computing. By 2025, ad revenue will reach $262.7 billion, and cloud revenue will hit $58 billion, together accounting for most of the roughly $400 billion total revenue.

However, the market landscape is changing. As consumers begin to delegate purchasing to AI agents instead of typing keywords into search bars, the existing search advertising model faces threats. Google’s investments in UCP and AP2 are in preparation for this future transformation.

Google is developing AI into the next stage of search. Initially functioning as a Q&A layer, it will gradually evolve into agents that execute purchases on behalf of users. Once merchants join the unified consumer platform (UCP) and list their products, these products will fall within the scope of agent transactions.

• Advertising: The display mode shifts from discovery to recommendation. Agents compare and filter products based on user conditions; paid merchants appear at the top of search results. This looks like organic recommendation, but in fact, advertising is now in the recommendation phase. Advertisers reduce investment at stages with low conversion rates; Google earns higher ad revenue per transaction.
• Payment: The AP2-based Agentic Checkout allows users to confirm payment once. Google Pay becomes the payment channel, charging per transaction. Agent transactions are faster and more frequent than manual ones. Small fees accumulate into significant revenue.
• Cloud computing: A potential area, but not yet a revenue source. Merchants processing agent transactions need AI inference, data storage, and API integration. If these needs flow into Google Cloud, cloud revenue will grow with the agent ecosystem.

Future Outlook

Google’s advantage lies in its existing network.

Google pioneered the internet era and has built nearly complete payment infrastructure, including Google Pay and a large merchant base. In the AI era, Google also leads with its Gemini platform, demonstrating keen insight into technological shifts. Additionally, through Android and Chrome, it maintains extensive user engagement channels.

If UCP and AP2 are widely adopted, users will complete end-to-end purchasing within Google’s infrastructure. Merchant onboarding will follow naturally. The current systems are designed around humans, while UCP and AP2 are tailored for agents. Merchants that fail to integrate will be at a disadvantage compared to those that do.

For merchants, joining UCP and AP2 is the easiest way to reach buyers.

Google has done this before. In 2008, it open-sourced Android. Manufacturers joined, user numbers surged, and Google’s own infrastructure—like the Play Store and Google Pay—was built on top. The result: Google became the biggest beneficiary of the mobile market without manufacturing any phones.

Once agent transactions truly take off, Google is likely to follow suit. Buyers and merchants will transact on its infrastructure, and Google will profit at every stage of the process.

OpenAI Group PBC

Core Technologies

OpenAI co-developed the ACP (Agent Business Protocol) with Stripe, released on September 29, 2025.

ACP is an open protocol that allows agents to invoke merchants’ payment systems and purchase goods on behalf of users. It grants permissions through a four-party structure: 1) Buyer, 2) Seller, 3) Agent, 4) Payment Provider.

The core issue of ACP is: “How much payment authority should be granted to the agent?” In theory, giving the agent access to the user’s bank card info allows it to pay any merchant at any time and amount. But poorly trained agents might repeatedly buy unnecessary items, and hijacked sessions could be maliciously exploited.

ACP solves this with delegated payments. The user’s actual bank card info is never transmitted to the agent. Instead, the payment service provider (e.g., Stripe) receives the card info and issues a one-time token, which the agent only handles. This token has four constraints:

• Where can it be used?
• Maximum payment limit
• Expiry time
• Which checkout session does it apply to?

Thus, even if the agent malfunctions or is hijacked, damages are limited to that single shopping transaction.

Core Business

OpenAI’s current revenue mainly comes from three pillars. By 2025, annual revenue (recurring revenue) is expected to reach about $20 billion, with ChatGPT subscriptions accounting for roughly 85%. The rest comes from API usage fees and enterprise contracts. This is a subscription model that grows linearly with user numbers.

The problem is, this structure is nearing a bottleneck. As OpenAI competes for subscription users with Claude and Gemini, its growth depends on how many new users it can attract. ACP is an attempt to break through this bottleneck by adding transaction fees. Beyond subscriptions, it also charges per transaction, considering both user count and transaction volume—layered growth.

In September 2025, OpenAI launched Instant Checkout, allowing payments within ChatGPT. It charges Shopify merchants a 4% transaction fee. However, issues like real-time inventory sync, tax infrastructure, and low conversion rates hinder its development.

Merchants oppose this, citing difficulty in handling complex variables like stock status, taxes, and pricing directly within ChatGPT. Walmart, in particular, revealed that conversion rates via ChatGPT are only a third of those on its official website.

In March 2026, OpenAI discontinued Instant Checkout, returning payment functions to merchant apps and systems, limiting ChatGPT’s role to product discovery.

This is not a retreat but a recalibration. Acquiring personal finance app Hiro Finance is expected to play a key role in this adjustment. The plan is to upgrade the previous obstacles—consumer behavior analysis, financial data management, inventory, tax, and fraud detection infrastructure—and then re-enter the internal payments space.

Once this ecosystem is established, OpenAI will realize its original vision: making ChatGPT the starting point for every transaction, paving the way for intermediary fee models.

Future Outlook

Unlike Google, OpenAI must compete with a single platform—ChatGPT. Its strategy is to leave payment, order fulfillment, and customer relationship management to merchants, focusing solely on product discovery.

OpenAI’s success hinges on satisfying both merchants and consumers. For merchants, more cases like Walmart integrating discounts and payments into ChatGPT are needed. For consumers, ChatGPT’s recommendations must translate into actual purchases. If one side falters, the other stalls. Too few merchants, and product choices dwindle; low conversion rates cause merchants to withdraw investments.

OpenAI does not have the resources like Google to leverage other assets for extra time.

Ultimately, whether OpenAI can dominate shopping depends on whether ChatGPT can replace the shopping starting point as thoroughly as it replaced search. Since Google is also competing via the Gemini project, this will be the toughest challenge for OpenAI.

Visa (V)

Core Technologies

Even in the agent era, Visa is determined to maintain its position as the “most widely used payment method.” To adapt to agent payments, Visa has chosen to open its existing payment network to agents.

In April 2025, Visa launched the Visa Intelligent Business Solutions suite. It includes four components enabling agents to make payments as if they were real humans.

The commonality of these four components is that Visa does not directly participate in protocol competition.

Visa’s proprietary API for agents activates when using Visa cards. The Intelligent Commerce Connect (Intelligent Commerce Connect) is a strategy to accept other protocols simultaneously.

Core Business

Visa’s current revenue mainly comes from credit card transaction fees. By 2025, it’s expected to reach about $40.8 billion, with a total transaction volume of $14 trillion. Currently, Visa’s Intelligent Business does not generate independent revenue; it’s a strategic foundation to sustain the existing revenue structure as agent payments arrive.

Revenue sources remain unchanged:

• Payment fees: Whether the user presses pay directly or an agent does, as long as the payment runs through Visa, a fee is generated. This is why Visa designed the Intelligent Commerce Connect to be compatible with other protocols.
• Token infrastructure fees: Whenever a transaction uses a token issued via one of five major agent APIs, Visa handles credential conversion and authentication. When PSPs like Stripe (PSP) and AI platforms use this token service, Visa charges a network usage fee. The token layer built on top of cards becomes a new revenue source in the agent era.

Ultimately, Visa’s strategy is not to win protocol wars but to charge all participants—regardless of who wins. On the buyer side, Visa partners with AI platforms like OpenAI, Anthropic, and Perplexity. On the seller side, it collaborates with e-commerce platforms like Shopify and payment providers like Stripe (PSP). No matter which protocol develops, Visa will be present at both ends.

Future Outlook

Visa’s vision can be summarized as: rather than competing in protocol wars, it aims to become an inclusive payment infrastructure supporting all protocols.

This choice is critical because Visa’s stance differs sharply from others. Google, OpenAI, and Coinbase are all vying for protocol dominance. AP2, ACP, or x402 must become standards within their ecosystems to maximize profits. These protocol battles are almost zero-sum.

In contrast, Visa’s approach is that as long as payments run through its network, the final protocol doesn’t matter. The winner of the protocol war is irrelevant to Visa. It profits by collaborating with the winning side, even if the market shifts to other protocols.

This embracing strategy is not “compromise” but actually the most advantageous position for Visa. Its ability to do so stems from its vast existing assets: 4.8 billion cards and 150 million merchants. Only such strength allows this choice.

However, a variable exists: stablecoins. If agent payments bypass Visa’s network entirely—settling directly on-chain on blockchains—Visa’s core revenue stream from fees would be threatened. To counter this, Visa has acquired Bridge, launched stablecoin cards, and participates as a validator in the pure stablecoin chain Tempo.

For this “embrace” strategy to work, payments must flow through Visa’s infrastructure. Stablecoins are the only potential disruptor.

The reason Visa does not aim to win protocol wars is clear: maintaining its current card organization dominance is enough to make it the biggest beneficiary of the agent era. The only concern is how quickly stablecoins can bypass the card network.

Mastercard (MA)

Core Technologies

Mastercard adopts a similar strategy. Even in the agent era, it strives to maintain its position as a card organization. Mastercard chooses to open its existing global payment network covering over 210 countries and regions, designed to allow merchants easy access.

In April 2025, Mastercard launched the Mastercard Agent Payment, followed by developer tools in September, and the Agent Payment Acceptance Framework in October, steadily building its agent payment system.

The commonality of these components is that Mastercard does not seek to dominate the market with its own protocol. Its strategy is to ensure that regardless of market direction, Mastercard can intervene in payment and identity verification.

The real challenge for Mastercard is enabling agent payments to be automatically completed at merchants without any manual operation. Usually, merchants need to embed code on their websites when adopting new payment methods. Mastercard has eliminated this burden.

Through cooperation with Cloudflare, Mastercard has built a front-end architecture that can automatically distinguish “trustworthy agents from malicious bots” and only allow trusted agents to access merchant sites. Merchants can accept agent transactions without changing any code.

For merchants seeking deep integration, there is an alternative path. Those wanting to connect their systems directly with agents can do so via protocols like MCP, A2A, and ACP. Ultimately, merchants face two choices: do nothing and accept default traffic, or connect via protocols and build customized experiences—all through Mastercard.

Core Business

Mastercard’s revenue structure is straightforward. Every payment processed over Mastercard’s network incurs a fee. By fiscal year 2025, revenue is projected around $32.8 billion, with 175.5 billion transactions. Whether via agents or direct payments, as long as the transaction runs through Mastercard, the same fee applies.

The issue is that agent payments could bypass the card network entirely. If settlement occurs directly on-chain via stablecoins, Mastercard’s role diminishes. Its agent payment strategy aims to fill this gap.

Mastercard lowers merchant barriers. Normally, adopting new payment methods requires embedding new code. Mastercard has removed this requirement. Partnering with Cloudflare, it built a front-end layer that filters traffic from trusted agents.

Merchants can accept agent payments without extra steps. The more merchants connect, the more transactions flow through Mastercard.

The revenue model remains the same: each agent payment via Mastercard’s network generates a fee, similar to manual card swipes. The difference is that agent payments are faster and more frequent, accelerating fee accumulation as transaction volume grows.

For Mastercard, the growth in agent transaction volume is a natural extension of its fee income. It’s not a new business but an extension of the existing infrastructure into the agent era.

Future Outlook

Mastercard’s view can be summarized as: like Visa, it does not aim to win protocol wars but to dominate how merchants accept agent payments.

The core is enabling merchants to accept agent payments without writing any code. Its partnership with Cloudflare effectively reduces merchant onboarding barriers.

Visa’s approach is similar, but it is also dealing with stablecoins. With Bridge acquisitions and Tempo validator participation, its defenses are expanding. Mastercard remains focused on merchant acceptance. Currently, defending one side is an advantage.

The question is how long this centralized strategy can last. If agent payments start settling directly on-chain via stablecoins, existing payment networks’ transaction volumes could decline. Visa has begun addressing this risk; Mastercard has not publicly responded.

Stripe Inc.

Core Technologies

Over the past 15 years, Stripe has become the standard for internet commerce payments thanks to its developer-friendly APIs. To maintain this position in the agent era, Stripe has chosen to build a new, directly usable payment channel on top of its existing infrastructure.

In September 2025, Stripe jointly released ACP as an open standard with OpenAI, and announced its own payment primitive SPT (Shared Payment Token).

Stripe’s agent payment stack consists of four parts.

The commonality is that only ACP is an open standard; the others are proprietary to Stripe. If ACP is the “universal language” between agents and merchants, then SPT is Stripe’s proprietary payment token responsible for actual fund flow.

The SPT process:

1. User says “Buy this” in ChatGPT.
2. Stripe issues a one-time payment token (SPT).
3. The agent passes this token to the merchant.
4. The merchant verifies the token and completes the payment.

The card number is never directly handed to the agent. Stripe generates a single-use token (SPT) for this transaction, which the agent only passes along. This reduces the risk of card data theft. Merchants already using Stripe can enable SPT payments with a single line of code.

In other words, for merchants already integrated with Stripe, this is the fastest route. For those on other PSPs (PSP), separate integration is required.

Core Business

Stripe’s uniqueness lies in designing SPT as a hub, not just a simple payment token, thus integrating agent payments into the Stripe ecosystem. Once a merchant adopts SPT for agent payments, the entire suite of Stripe products becomes bundled.

The operation:

• All agent transactions via SPT follow the same fee structure as before.
• Agent payment speed and frequency are much higher than manual payments.
• As transaction volume increases, the speed of fee accumulation accelerates.
• Payment methods change, but fee structure remains.

Visa and Mastercard only charge once per payment over their networks; subsequent transactions are free. Stripe, however, uses the payment as an entry point, connecting merchants to its full financial services ecosystem.

From the merchant’s perspective, adopting a single agent payment method ultimately ties them into the entire Stripe ecosystem.

Future Outlook

Stripe’s vision can be summarized as: a comprehensive coverage strategy, occupying all layers of agent payments.

Stripe has established a solid position across all layers: protocol (ACP), payment token (SPT), and stablecoins (Tempo). Visa and Mastercard see stablecoins as a defensive challenge, while Stripe views bypassing card networks as an offensive asset.

But the core question remains: can Stripe surpass the card network?

Currently, Stripe’s profit model still relies on Visa and Mastercard networks. Payment service fees (SPT), transaction fees, and merchant bundling all depend on the card infrastructure. Only when stablecoins truly replace card networks will this model change, but no one knows when that will happen.

Stripe is investing in both directions. If Visa and Mastercard control the highway, Stripe is the most successful logistics company on that highway. Becoming the top logistics provider is different from controlling the highway itself.

First, consider the relationship with OpenAI. ACP was co-developed, but real-time payments for ChatGPT are effectively monopolized by Stripe. If OpenAI’s traffic grows large enough to establish its own payment channels or partner with other PSPs, Stripe could lose its biggest AI channel. The current partnership is necessary but not forever.

Second, the openness of ACP is crucial. As an open standard, competitors like Adyen and Worldpay (PSP) can issue compatible tokens. Currently, Stripe’s market share makes merchants naturally choose SPT; once competitors issue similar tokens, the protocol’s premium will decline rapidly.

3. Pay-per-call

As mentioned, agentic commerce is a market where humans decide the payment method, while pay-per-call is a market where the agent itself becomes the payer. As long as the account has sufficient funds, the agent autonomously settles. When an agent calls another agent’s API, data, or performs calculations, this mechanism activates.

For example, a broker is asked to produce a market analysis report:

1. User requests the agent to write a market analysis.
2. The agent queries OpenAI API and receives a response.
3. To verify data, it requests info from Dune and Nansen APIs.
4. The completed report is sent to another agent for review.

A report might involve dozens of payments. Unlike general agent commerce, these payments are precisely tracked to fractions of a cent. In traditional payment systems, just the fee alone could cause losses per transaction. That’s why pay-per-call models rely on ultra-low-fee stablecoins.

Currently, three foundational standards exist for pay-per-call infrastructure:

• Protocol Layer: How agents express payment intent via messages.
• Settlement Layer: Which chain handles settlement, and in what assets.
• Trust Layer: Who verifies who paid what, and what permissions they have.

Coinbase Global, Inc. (COIN)

Core Technologies

Coinbase is the creator of x402, a standard allowing agents to pay without using card networks.

In May 2025, Coinbase released the x402 protocol, officially activating the previously dormant HTTP 402 (Payment Required) status code. Later that year, Coinbase and Cloudflare established the x402 Foundation, expanding x402 into a general payment layer.

x402 is an open protocol composed of five elements.

Its operation is simple: when an agent calls an API, the server responds with an HTTP 402 containing payment terms: amount, recipient address, asset, and network info. The agent signs the payment authorization with its wallet, adds the signature to the response header, and resubmits the request. The server’s coordinator verifies and settles the payment, then returns the resource.

Process

1. Agent requests resource from server.
2. Server responds with HTTP 402 and payment terms.
3. Agent signs payment authorization with its wallet, resends request with X-PAYMENT header.
4. The payload is passed to the facilitator (verification and settlement service), which broadcasts it on-chain.
5. After settlement confirmation, server returns a 200 response with the resource.

The key difference from existing payment systems is that payment itself includes identity verification. Agents can autonomously discover, pay, and use new APIs without human intervention.

Version 2 (V2) extends x402 from a single-call payment protocol to a general layer supporting multiple authentication, session, and chain types.

Most importantly, it introduces wallet-based session mechanisms. V1 required payment for each call; V2 introduces session-based payments (using X login). After a single payment, the session can be reused within a set period. This change enables workloads like LLM calls and multi-call agents, previously too slow or costly to pay per call, to operate on x402 for the first time.

Core Business

Coinbase’s current revenue mainly comes from transaction fees. By 2025, annual revenue is projected around $7.2 billion, mostly from crypto trading fees. x402’s revenue structure is not directly tied to Coinbase’s current income. That’s why Coinbase open-sourced x402: to set a free standard and profit from the underlying infrastructure.

This strategy is similar to Google’s: open-source Android, then earn from Play Store and Google Pay.

x402’s target market differs fundamentally from Visa, Mastercard, and Stripe. Credit card payments incur a fee of $0.30 plus 2.9%. Micro-payments like $0.01 API calls, $0.005 image classifications, or $0.50 per minute GPU time cannot exist on credit card networks.

The x402 market involves agent-to-agent or API provider payments that humans cannot initiate with a button press.

When this distant future market opens, Coinbase will gain two revenue streams:

• Basic transaction fees: Each x402 payment settles on Coinbase’s own Layer 2 chain, Base. Coinbase earns sequencing fees from each transaction. As x402 volume grows, so do Base’s transaction fees and locked assets. The standard is free to set, but the chain is operated by Coinbase.
• CDP infrastructure fees: Each x402 payment requires an intermediary (payment service provider) to verify “the payment is real” and transfer funds. The default PSP is operated by Coinbase, part of Coinbase’s developer platform (CDP). As developers adopt x402, related tools—wallet creation, gas sponsorship, data analytics—are naturally introduced.

From version 2 onward, third-party service providers also participate, weakening Coinbase’s lock-in ability. Coinbase’s real strategy is elsewhere: its Base blockchain, optimized for x402, offers minimal costs, high speed, and is explicitly recommended in its white paper.

In short: Stripe binds payments to its system; Coinbase builds payment channels allowing payers to naturally dock at Coinbase’s “rest area” (CDP tools).

Future Outlook

Coinbase’s vision: it’s not about beating competitors but about rapidly opening a market that doesn’t yet exist.

This is crucial because x402 does not compete with credit card networks. Micro-payments like API calls, image classifications, or GPU minutes are impossible on credit card networks.

Even if Visa and Mastercard accept agent payments, that only covers human-like shopping. The x402 market opens a lower layer: agents paying other agents or service providers in ways humans cannot.

The challenge is that the market for inter-agent transactions remains small. For agents to autonomously buy data, compute resources, and work products from other agents as part of daily workflows, several conditions must be met: low-cost LLM inference, widespread production use of agent frameworks, and support for x402 by data and API providers.

When will this market explode—within six months, two years, or five? No one knows. Coinbase’s success depends on market dependence it cannot control. The key question: can Coinbase maintain its position as the first to build agent payment infrastructure, and will it persist until the market opens?

Stripe (MPP)

Core Technologies

Stripe is the only provider embedding its own standards into both agent commerce and pay-per-use models. Previously discussed protocols—ACP and SPT—are card-based, delegating shopping to agents. MPP (Machine Payment Protocol) is a separate protocol designed specifically for pay-per-use, enabling agents to autonomously pay other agents’ APIs, data, and compute resources.

Released on March 18, 2026, as an open standard, MPP was developed jointly with Stripe and Paradigm’s mainnet, Tempo.

Like x402, MPP is an open payment protocol built on HTTP 402. The main differences are:

First, protocol neutrality. While x402 is built around stablecoins, MPP can handle stablecoins, bank cards, or other fiat payments within the same protocol. Visa extended MPP to bank card payments; Stripe supports bank cards and e-wallets via its platform. Different payment methods act as plugins on the same protocol.

Second, session support. When an agent initiates a session, on-chain funds are provisionally deposited. During the session, credentials are exchanged off-chain; final settlement occurs at session end.

This contrast is visually clear: one-time payments (Charge) resemble x402, with a single on-chain transaction; continuous payments (Session) involve only two on-chain records—initial deposit and final settlement—while intermediate calls are off-chain with credentials.

This difference enables MPP to achieve a throughput target of over 1 million transactions per second.

It’s akin to fueling at a gas station: one authorization at start, one final deduction at end, regardless of volume. Based on this, MPP aims to process over 1 million transactions per second.

Core Business

Stripe divides the agent payment market into two layers. The first is human recommendation of products—covered by ACP and SPT. The second, which MPP targets, is agents autonomously paying for other agents’ APIs, data, and compute resources. Like x402, it’s designed for pay-per-use, but with different architecture.

x402 supports only stablecoins; MPP supports bank cards, stablecoins, and Bitcoin Lightning Network, creating diverse payment options. The initial partner list for Tempo reveals market features: Anthropic, OpenAI, Deutsche Bank, and Visa are onboard. Protocol extensions for MPP are implemented by Visa (bank cards), Stripe (bank cards and wallets), and Lightspark (Lightning Network).

The reason: AI model calls are the most frequent payments for agents. Whether conducting research or coding, agents constantly call OpenAI or Anthropic services. These companies are involved from the design stage, so transaction volume paths are well established.

In this structure, Stripe has two revenue sources:

• Payment processing fees: When using MPP for bank card payments, Stripe handles actual processing. The protocol remains open for all; fund transfer is integrated into Stripe’s infrastructure. As agent payments grow, so does transaction volume through Stripe.
• Tempo ecosystem revenue: The default settlement layer for MPP is Tempo, a blockchain built by Stripe and Paradigm, with Stripe as a validator. As MPP volume increases, transaction fees are distributed to validators, with Stripe holding one seat. The open protocol allows participation in the protocol’s settlement chain, earning a share of fees.

Ultimately, Stripe replicates the agentic commerce design in the pay-per-use domain. ACP and SPT are based on card rails; MPP and Tempo on stablecoin rails. Protocols are open; infrastructure is closed. Regardless of the payment rail, all transactions go through Stripe.

Future Outlook

Stripe’s outlook: this protocol causes a shift in Visa’s strategy.

Visa once aimed to build an independent ecosystem with its own trusted agent protocols and smart commerce platform. Its strategy was “use our standards.” But with MPP’s emergence, Visa adopted MPP, becoming a partner in its card rail expansion. Visa’s crypto head Cuy Sheffield commented that Visa “sees MPP as another way to define how agents and merchants communicate.”

This indicates Visa’s shift from fighting native crypto standards to embedding card payment systems within them.

The agent payment market is fragmenting. Regulated human transactions happen via bank cards; inter-agent transactions (API calls, compute purchases, micro-payments) happen via stablecoins. While x402 is the crypto-native standard for the latter, MPP acts as a bridge, integrating both channels into a single protocol. Stripe has built the architecture to operate this bridge.

But there are two variables:

First, the standard competition between Coinbase’s x402 and the x402 protocol. Coinbase’s x402 has accumulated over 100 million payments and ensures neutral governance via transfer to the Linux Foundation. Stripe has joined the x402 Foundation as a partner. The question is whether these two protocols can coexist long-term or eventually merge.

Second, how fast will Tempo gain adoption? MPP’s settlement layer is Tempo, but how much agent payment traffic will it attract? Trust capital from Visa and Scotiabank is significant, but whether developers and service providers will choose Tempo remains uncertain.

Circle Internet Group, Inc. (CRCL)

Core Technologies

Circle started as the issuer of USDC and has evolved into a full-stack provider of stablecoin infrastructure. In the agent payment era, USDC is no longer just a payment method. Circle has adopted a vertical integration strategy, combining payment primitives, wallets, and settlement chains into its tech stack.

In September 2025, Circle released developer-controlled wallets and x402 integration examples. In October, the Arc public testnet launched, followed by the Nanopayments testnet in March 2026. Circle’s call-based payment tech stack includes three components.

The core of Nanopayments is off-chain aggregation and batch settlement. Many micro-payments are aggregated off-chain and settled on-chain as a single batch, drastically reducing gas costs per transaction. Circle bears the gas fees during batch settlement.

Arc is a Layer 1 payment service designed for stablecoin finance. Its key feature is using USDC as native gas. It offers sub-second finality based on the Malachite consensus engine, an integrated FX engine, optional privacy, and EVM compatibility.

Developer-controlled wallets allow developers to create and manage proxy wallets via a single API. Built on multi-party computation (MPC), it does not expose private keys and can operate across multiple chains (including Base, Ethereum, and Arc), viewing USDC as a unified balance.

When these three technologies are combined, agents can enjoy a complete payment experience within Circle’s stack: USDC issuance (USDC) → payment layer (Nanopayments) → settlement chain (Arc) → wallet management (Wallets).

OpenMind’s autonomous robot dog pays itself USDC via Nanopayments to charge—an early proof of this tech stack’s effectiveness.

Circle’s revenue structure is simple. In 2025, total revenue and reserve income will reach $2.7 billion, with over 95% from interest on USDC reserves. As USDC is used, reserve interest accumulates in Circle’s accounts. Notably, Circle has agreements with major distribution partners like Coinbase, sharing part of the interest income. Overall, as long as USDC is used for payments, Circle’s revenue pool will keep growing.

While Stripe built the Tempo chain, it does not issue stablecoins on that chain. Coinbase’s situation is similar. Regardless of who wins the on-chain race, growth in stablecoin payments will boost Circle’s reserves.

As agent payments increase, this structure will keep accumulating. Each API call, data purchase, or computation increases USDC circulation. As USDC circulation grows, reserves and interest income will also increase.

Additionally, Circle has a second revenue stream via Arc. After transitioning to mainnet, each on-chain transaction generates gas fees paid in USDC. As USDC demand rises, new USDC is issued proportionally, increasing reserves and interest income. Tempo’s architecture treats USDC as an external asset, while Arc’s design uses its own asset as gas, directly translating on-chain activity into USDC demand.

In essence, Circle’s game differs from competitors. Stripe and Coinbase aim to capture agent payment flows via blockchain and protocols; Circle holds the assets used within those flows.

Future Outlook

Circle’s vision: the only full-stack player with asset issuance rights.

Its key advantage is its monopoly on USDC issuance. While Stripe built the Tempo chain, it does not issue stablecoins on it; USDC on Tempo is issued by Circle. Coinbase faces a similar situation with Base.

Whichever chain ultimately prevails, stablecoin payment growth will increase Circle’s reserves. This grants Circle a structural advantage at the asset level.

This advantage involves two factors:

First, USDC’s market share. The stablecoin market is currently dominated by USDT and USDC. Tether, PayPal, and bank-issued stablecoins are entering the scene. Whether USDC will become the default in agent payments or other stablecoins will take market share remains uncertain.

Second, the adoption speed of Arc. Circle’s move to embed Arc into the chain layer is significant. But the pure stablecoin chain market is crowded: Tempo has trust from Visa and Scotiabank; Base leads in transaction volume. The key question: how much USDC (Circle’s unique asset) can be issued into the Arc ecosystem? Only when USDC usage on Arc increases can this vertical integration truly work.

If stablecoin payments become the default in the agent era, Circle will be the biggest beneficiary. But it must also build this premise itself.

Ethereum Foundation (ETH)

Core Technologies

Ethereum chooses to provide a trust layer for the agent economy via open standards. Other participants handle agent payments within their fee structures, while Ethereum builds a core protocol to record agent identities, reputation, and verification of work results on-chain in a standardized way.

On August 13, 2025, the Ethereum Foundation officially submitted ERC-8004 (Trustless Agent), deployed on mainnet on January 29, 2026.

ERC-8004 aims to extend the protocol’s trust layer. The list of co-authors reflects its nature: Marco De Rossi (MetaMask), Davide Crapis (Ethereum Foundation), Jordan Ellis (Google), and Erik Reppel (Coinbase). This lineup brings together AI infrastructure, crypto exchanges, wallets, and the Ethereum Foundation.

ERC-8004 consists of three on-chain registries.

When these registries are combined, agent identities, transaction histories, and verification results are accumulated on-chain. Previously, reputation earned on platform A could not transfer to platform B. Under ERC-8004, reputation becomes a transferable asset for the agent itself.

The trust model choice correlates with task risk. For low-risk tasks (e.g., ordering pizza), only reputation queries suffice. For high-risk tasks (e.g., medical diagnosis), the model requires re-execution with collateral, or even TEE attestation. The standard is flexible: verification rigor varies with task risk.

Crucially, payment is not part of ERC-8004