The Battle for the AI Payment Race: Traditional Card Networks Face Off Against Coinbase

marsbitPublished on 2026-06-08Last updated on 2026-06-08

Abstract

With the rise of AI agents conducting transactions, a battle for the underlying payment infrastructure is underway. Two distinct and incompatible approaches have emerged for enabling autonomous AI payments. The first approach is championed by traditional card networks Visa and Mastercard. They leverage their existing tokenized card credential systems, extending them to allow verified AI agents to make purchases within user-defined limits. Services like Mastercard's Agent Pay and Visa's Intelligent Commerce integrate with major AI platforms (e.g., OpenAI, Anthropic) and keep transactions within the established, decades-old card payment model. This system offers advantages for consumer retail, including robust fraud protection, chargeback mechanisms, and extensive merchant networks. The second approach, led by Coinbase, utilizes stablecoins on open internet protocols. Its x402 protocol reactivates the HTTP 402 status code for machine-to-machine micropayments, using USDC for settlement directly on-chain. This method eliminates the need for accounts or card fees, making it highly efficient for high-frequency, low-value, cross-border transactions between AI agents—such as paying for API calls, data streams, or computational resources—where traditional card fees and settlement times are impractical. While card networks excel in consumer-facing scenarios requiring dispute resolution, stablecoin protocols are tailored for machine economies. A key challenge for both is agent identi...

Written by: Zennon Kapron

Compiled by: Chopper, Foresight News

As AI agents increasingly undertake various commercial transactions, a battle for the underlying payment infrastructure has already begun.

Currently, the technical path to enabling AI agents to make autonomous purchases has diverged into two incompatible solutions: through which channel transactions are ultimately cleared and settled when software programs make payments on behalf of users. One camp builds payment rails based on tokenized bank card credentials controlled by Visa and Mastercard; the other, led by Coinbase, adopts stablecoins for settlement based on open internet protocols. While the surface-level focus of AI agent commerce is on shopping assistant applications, the core game behind the scenes is about who will dominate the next-generation payment system.

Two Payment Channels, Adapted to Different Application Scenarios

Traditional card networks have taken the initiative and moved swiftly. Mastercard launched its Agent Pay service in April 2025, built on its proprietary agent tokenization system. This tokenization technology, originally designed for contactless payments and card-on-file payment scenarios, has been extended to allow verified AI agents to make transactions on behalf of users within authorized limits.

From its launch, the service assembled a host of industry partners, signaling a clear strategic intent: collaborators include Microsoft, IBM's watsonx intelligent orchestration platform, and payment service providers Braintree and Checkout.com. A day later, Visa introduced its Visa Intelligent Commerce service, opening its payment network to AI developers, with AI-adapted bank cards as the core vehicle. This solution replaces raw card numbers with tokenized credentials, proving user authorization for a specific AI agent and defining transaction boundaries. Visa also enlisted several leading AI companies, including Anthropic, OpenAI, Perplexity, Mistral, and Samsung.

The solutions from both major card networks keep transactions within the decades-old bank card payment model. While AI agents are a new entity, the payments flow through traditional channels that have served global commerce for half a century.

The stablecoin camp adopts an architecture fundamentally different from the card networks. Coinbase launched the x402 protocol in May 2025, reviving the long-dormant HTTP 402 "Payment Required" status code. The protocol leverages this code to settle transactions using the USDC stablecoin directly on the internet. The specific process involves: a client requesting access to a resource, the server returning a payment instruction; the client attaching signed stablecoin payment information in the request header; once the on-chain transaction is confirmed, the corresponding resource becomes accessible. The entire process requires no account registration, card binding, and incurs no bank card transaction fees.

This solution is designed specifically for machine-to-machine (M2M) transactions. An AI agent might need to make thousands of micro-payments for API calls, data streams, or interactions with other agents. Processing such transactions through traditional card channels is completely unfeasible from a cost perspective.

Both technical routes have their strengths. The bank card channel excels in personal retail consumption scenarios, which place high demands on chargeback handling, fraud prevention, and dispute arbitration mechanisms. The stablecoin channel holds significant advantages for high-frequency, small-value, cross-border machine transactions, where traditional card fee structures and settlement times are ill-suited. The core of the contest lies in which type of scenario will become the mainstream for AI agent commercial transactions.

A major challenge common to both routes is identity verification. When a software program initiates a payment, merchants need to confirm that the operator is a legitimate agent authorized by a real user, not a malicious bot using stolen credentials. Simultaneously, users need a mechanism to dispute transactions erroneously initiated by an AI agent.

Visa, noting a 47-fold surge in AI traffic to US retail websites, collaborated with cloud service provider Cloudflare to launch a Trusted Agent Protocol for distinguishing legitimate AI programs from malicious crawlers. This also highlights the structural advantage of traditional card networks: five decades of accumulated risk scoring systems, chargeback rules, and dispute resolution mechanisms are well-suited to handle issues like AI agents purchasing the wrong item. In contrast, stablecoin transactions are immutable once on-chain, with no native reversal solution within the system.

Looking ahead, the key factor determining the direction of the consumer market may not be which payment channel has lower fees, but who can solve the challenges of agent identity verification and transaction dispute resolution.

Dual-Track Strategy: Card Networks Betting on Both Arenas

A telling signal is that Visa and Mastercard are not betting everything on their own rails but are also simultaneously investing in the stablecoin arena.

By April 2026, Visa's stablecoin settlement business had reached an annualized transaction volume of $7 billion, a 50% increase quarter-over-quarter. The company added support for 5 new public blockchains, bringing its total number of partner chains to 9, while launching over 130 "stablecoin + bank card" linkage projects in more than 50 countries globally. In October 2025, Visa doubled down, jointly launching the Trusted Agent Protocol with Cloudflare to help merchants identify legitimate agents versus malicious programs. It also publicly announced a collaboration with Coinbase to promote interoperability between its network and the x402 protocol. Seemingly competing bank card systems and stablecoin protocols are now building bridges.

Mastercard has adopted a similar dual-track strategy. In March 2026, Mastercard announced its intention to acquire stablecoin platform BVNK for up to $1.8 billion. Prior to this, its Agent Pay service had expanded to Latin America and the Caribbean, with local issuer adaptation completed in early 2026.

It's clear that the core strategy of the two major traditional card networks is: no longer merely defending the bank card channel, but striving to become the fee-collecting gateway for all payment rails, whether proprietary or stablecoin-based. This layout strongly indicates their assessment: if the industry ultimately settles on bank cards as the mainstream for AI payments, they would not need to invest heavily in acquiring stablecoin infrastructure.

Diverging Application Scenarios

Based on currently launched products, the application boundaries of the two technical routes are quite clear.

Mainstream products targeting ordinary consumers mostly choose the bank card channel. The "Checkout with ChatGPT" feature launched in September 2025, co-developed by OpenAI and payment service provider Stripe, relies on a shared payment token for bank card settlement. This token is limited to specified merchants and shopping orders, initially connecting with Etsy sellers and later covering over a million Shopify stores. Amazon's "Buy for Me" feature, which allows AI agents to make purchases on third-party websites on a user's behalf, automatically populates the user's stored bank card for checkout.

AI shopping services for personal consumption commonly choose bank cards due to the system's mature anti-fraud tools, extensive merchant network, and long-established user trust.

Meanwhile, the stablecoin channel firmly occupies the machine transaction market. Amazon integrated the x402 protocol into its Bedrock agent core payment service, using Coinbase's Base blockchain for settlement, with transactions taking about 200 milliseconds and fees under one cent. Stripe also joined the service as a payment gateway. According to Coinbase data, in its first year, the x402 protocol processed over 169 million payment orders, covering 590,000 buyers and 100,000 sellers.

These transactions are not about users shopping for clothes online, but AI agents paying for services like computing power, data, and API calls. Their frequency and individual amounts are incompatible with the design logic of bank cards. In September 2025, Coinbase, jointly with Cloudflare, spearheaded the establishment of the x402 Foundation, aiming to promote the development of universal industry standards rather than building a closed proprietary product.

Summarizing five flagship AI commercial payment projects launched by early 2026: 3 used bank card settlement, 2 used stablecoin settlement, with application scenarios largely divided along the lines of personal consumption and machine transactions.

Future Industry Direction

In the short term, the industry landscape in 2026 is likely to maintain the status quo: bank cards dominating personal retail payments, stablecoins specializing in inter-machine transactions, with both coexisting and developing. However, by 2030, this situation may be disrupted, as both camps are vying for the convergence zone of the two scenarios.

The ultimate decider will depend on whether AI-driven commercial transactions ultimately lean more towards traditional retail forms or evolve into a massive network of micro machine transactions. If it's the former, traditional card networks will retain dominance. If it's the latter, stablecoin channels will capture a significant volume of new transaction flows.

Visa and Mastercard have made the safest choice: a dual-track investment in both arenas, ensuring they can collect fees regardless of where future flows go. The entities that truly need to be cautious are those betting on a single payment channel. The two major card networks have already hedged this risk, which intuitively reflects their judgment about the industry's future.

Related Questions

QWhat are the two main technical approaches for enabling AI agents to conduct autonomous payments, as discussed in the article?

AThe two main approaches are: 1) The traditional card network model led by Visa and Mastercard, which uses tokenized card credentials for settlement. 2) The stablecoin-based model led by Coinbase, which uses the x402 protocol and USDC stablecoin for settlement directly over the internet.

QAccording to the article, what is a key structural advantage that traditional card networks like Visa possess in the AI payment space?

AA key structural advantage is their 50-year history of developing risk scoring systems, chargeback rules, and dispute resolution mechanisms. This existing infrastructure is well-suited for handling issues like incorrect purchases made by AI agents, which are difficult to address in native, irreversible stablecoin transactions.

QWhy do personal consumer AI shopping services typically prefer the card network payment channel?

APersonal consumer AI shopping services prefer the card network channel because it offers mature anti-fraud tools, an extensive merchant network, and long-established user trust. This makes it more suitable for retail purchases where issues like refunds and disputes are common.

QWhat is the strategic rationale behind Visa and Mastercard's 'dual-track' approach of investing in both their traditional networks and the stablecoin sector?

AThe strategic rationale is to ensure they remain the fee-collecting gateway for all payment flows, regardless of which channel (traditional card or stablecoin) ultimately dominates AI agent commerce. This approach hedges their bets and shows they are not relying on a single payment future.

QWhat is the primary use case for the stablecoin-based x402 protocol, and what specific transaction characteristics make it suitable?

AThe primary use case for the x402 protocol is machine-to-machine (M2M) transactions. It is particularly suited for high-frequency, low-value, cross-border micropayments, such as payments for API calls, data streams, or interactions between AI agents, where the traditional card network's fee structure and settlement speed are impractical.

Related Reads

Jensen Huang 'Saves' South Korean Stock Market: Locks In SK Hynix Memory, Chip Shortage to Continue

On June 5th, South Korea's stock market experienced a sharp decline, with major chipmakers like Samsung and SK Hynix dropping nearly 10%. Amidst the turmoil, NVIDIA CEO Jensen Huang's visit to Seoul played a dramatic role in boosting market sentiment. Following a dinner meeting with SK Group Chairman Chey Tae-won and SK Hynix CEO Kwak Noh-Jung, Huang confirmed that NVIDIA's new Vera CPU will utilize SK Hynix DRAM. The companies announced a multi-year technical partnership to co-develop next-generation memory for NVIDIA's AI infrastructure, covering products from data centers to personal AI and robotics. This collaboration extends beyond memory supply. SK Hynix is integrating NVIDIA's AI and Omniverse platform into its own semiconductor design and manufacturing processes, including computational lithography and creating digital twins of its fabrication plants for autonomous operation. While strengthening ties with SK Hynix, NVIDIA is diversifying its supply chain for the upcoming HBM4 memory, with Samsung, SK Hynix, and Micron all certified as suppliers for its Vera Rubin platform. Despite this, Huang warned that the global chip shortage, driven by relentless demand from AI factory construction, is expected to persist for several years across the entire supply chain. His visit underscores NVIDIA's systematic effort to deepen integration with South Korea's broader tech industry.

marsbit24m ago

Jensen Huang 'Saves' South Korean Stock Market: Locks In SK Hynix Memory, Chip Shortage to Continue

marsbit24m ago

Nasdaq Plunges 4.2% in a Single Day: Does "Black Friday" Burst the U.S. Stock Market Bubble?

The Nasdaq plunged 4.18% on June 5, 2026, its worst single-day drop in over a year, as a much stronger-than-expected US jobs report triggered fears of economic overheating and delayed Federal Reserve interest rate cuts. The selloff, centered on high-valuation tech and AI stocks like Nvidia and Broadcom, spread across major indices. The article examines whether this signals a market top. The strong May non-farm payrolls data, nearly double expectations, pushed bond yields higher, directly hurting rate-sensitive tech stocks. This exposed vulnerabilities in the crowded AI trade, where valuations had soared on narratives of infinite growth, despite emerging signs of slowing order momentum and corporate AI monetization challenges. Prior to the drop, market indicators flashed warning signs: historically high valuations (e.g., Shiller CAPE ratio near 39.5), extreme bullish sentiment, and high levels of leverage. Technical charts showed key support levels being breached. Wall Street is divided on the outlook. Bears, citing risks of "stagflation" and AI bubble comparisons to the dot-com era, warn of a potential significant correction. Bulls view the drop as a healthy correction within a bull market, underpinned by a strong economy and expected corporate earnings growth of around 7% in 2026. The immediate future hinges on upcoming key events: the May CPI inflation data and the mid-June FOMC meeting. Their outcomes will critically shape market expectations for the Fed's rate path. The article concludes that conditions for a major market top are aligning, marking a fragile transition from narrative-driven gains to a phase demanding validation from macroeconomic data and corporate fundamentals. Caution is advised.

marsbit28m ago

Nasdaq Plunges 4.2% in a Single Day: Does "Black Friday" Burst the U.S. Stock Market Bubble?

marsbit28m ago

Nasdaq Plunges 4.2% in a Single Day, Did 'Black Friday' Pop the U.S. Stock Bubble?

The Nasdaq Composite plummeted 4.18% on June 5, its biggest single-day drop since April 2025, triggering widespread debate over whether the U.S. stock market has peaked. The sell-off was sparked by a stronger-than-expected U.S. non-farm payrolls report, which fueled fears of economic overheating and pushed back market expectations for Federal Reserve rate cuts, leading to a sharp rise in Treasury yields. The AI sector, the primary driver of the recent bull market, suffered severe losses, with the Philadelphia Semiconductor Index crashing over 10%. Stocks like Nvidia, Broadcom, and Micron led the decline. Concerns are mounting about the sustainability of AI capital expenditures and high valuations, with signs of order cuts for next-generation chips emerging. Analyses point to several warning signs: historically high market valuations (e.g., elevated Shiller CAPE ratio, Buffett Indicator), extreme bullish sentiment indicators, and significant insider selling. The sell-off also caused a key technical breakdown, with the S&P 500 breaking below its short-term moving average and testing its 200-day moving average. Wall Street is divided on the outlook. Bears warn this could be the start of a bubble deflation or a "stagflation" scenario, while bulls view it as a healthy, overdue correction within a bull market driven by solid corporate earnings growth. A more moderate view suggests the easy liquidity-driven rally is over, and markets are entering a phase of fundamental stock-picking with potential for consolidation. The immediate future hinges on key upcoming events: the May CPI report and the mid-June FOMC meeting. Their outcomes will be critical in determining whether this is a temporary pullback or the beginning of a more significant trend reversal. The consensus is that the era of one-directional market gains may be ending, requiring increased investor caution.

Odaily星球日报33m ago

Nasdaq Plunges 4.2% in a Single Day, Did 'Black Friday' Pop the U.S. Stock Bubble?

Odaily星球日报33m ago

The First Case on AI Agents: What Was Adjudicated?

"The First 'Agent' Ruling: What Was Decided?" On April 30, the Guangzhou Internet Court issued a ruling—China's first behavior preservation order in the intelligent agent (AI agent) field. The defendant, an open-source AI agent software, was ordered to stop downloads, cease actions that bypassed a platform's technical protection measures, and delete related tutorials and data. The core issue: the software used the operating system's "accessibility service" permissions to automate user interactions within other apps without those platforms' authorization. This mirrors a recent US case where Amazon sued Perplexity for similar reasons—bypassing Amazon's API to directly scrape and interact with its pages—and won a preliminary injunction. Both rulings establish a crucial legal boundary for the AI agent era: agents cannot operate unchecked. The article argues the fundamental legal principle emerging is one of **dual authorization**. An AI agent requires both **user consent** AND **platform consent** to operate legitimately within that platform's ecosystem. Bypassing platform rules through system-level permissions, even with user permission, undermines platform responsibilities for content moderation, data security, and user privacy, creating liability issues. The piece uses the evolution of "Doubao Phone" (an AI-integrated smartphone) as a case study. Its initial, aggressive version that bypassed platform controls faced roadblocks. Its upcoming 2.0 version is reportedly pivoting to negotiate API access and authorization deals with major platforms (like Alibaba's ecosystem), seen as a strategic adaptation to the new regulatory reality. A global trend is identified: the era of unregulated, "wild west" growth for AI agents is ending, replaced by a **compliance race**. This raises barriers to entry, as securing platform authorizations becomes a new cost. Open-source status is also not a legal shield if the code facilitates bypassing technical protections. In conclusion, these first rulings target not the largest, but the most **aggressive and representative** cases. By setting precedent with them, regulators are efficiently steering the entire industry towards a new, more regulated operating paradigm defined by dual authorization and platform cooperation.

marsbit38m ago

The First Case on AI Agents: What Was Adjudicated?

marsbit38m ago

Fired by Google Over a 14-Page Paper, Over 4,000 Rallied for Her. 6 Years Later: She Almost Predicted the Entire AI Era Back Then.

In late 2020, Google AI researcher Timnit Gebru was effectively dismissed following a conflict over a 14-page, unpublished research paper she co-authored titled "On the Dangers of Stochastic Parrots." The paper, which has since been cited over 14,000 times, raised critical early warnings about the risks of large language models (LLMs). It argued that these models, trained on vast, biased internet data, are essentially "stochastic parrots" that mimic language without true understanding, potentially amplifying societal biases, generating plausible but false information (later termed "AI hallucination"), consuming massive energy, and obscuring their training data contents. Gebru's stance led to a clash with Google management, who requested the paper's withdrawal. Her subsequent internal criticism of the company's diversity efforts and handling of the matter culminated in her termination, which sparked protests from over 4,000 Google employees and researchers. Six years later, the paper's predictions have proven remarkably prescient. Issues like AI hallucination, embedded bias (evident in resume screening and healthcare algorithms), soaring energy consumption from AI data centers, unvetted training data containing harmful content, and the risk of "model collapse" from AI-generated internet content have become central industry challenges. The incident also highlighted concerns about AI development being driven primarily by commercial competition within a handful of powerful tech companies, often at the expense of ethical considerations. After leaving Google, Gebru founded the Distributed AI Research Institute (DAIR) to explore these issues independently. The controversy underscores how her early, critical insights into the fundamental limitations and societal impacts of LLMs anticipated many of the most pressing dilemmas in today's AI era.

marsbit39m ago

Fired by Google Over a 14-Page Paper, Over 4,000 Rallied for Her. 6 Years Later: She Almost Predicted the Entire AI Era Back Then.

marsbit39m ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片