The Next Billion Customers Will Be AI: Stablecoins Disrupting the Market

[Mexico City = Shim Young-jae, Correspondent] There is a forecast that the 'next billion users' in the New York digital asset (virtual asset) market may not be humans but AI agents. As AI increasingly engages in transactions to purchase data, computing resources, and online tools, stablecoins are emerging as a means of micro-payment between machines. Coinbase is leading with its payment protocol 'x402' for AI agents, while Visa, Mastercard, Cloudflare, and MoonPay are also entering the infrastructure development for related payments.

AI Purchases Data via API... USDC Captures Early Market

According to a report by CoinDesk on the 22nd (local time), AI agents have begun to autonomously perform multiple steps based on user instructions, purchasing data, computing resources, and application programming interfaces (APIs). While traditional e-commerce has been designed around human card payments, a new market structure is emerging where software pays other software directly.

The x402 developed by Coinbase is a prime example. When an AI agent requests specific data or services, the seller presents pricing and payment information. The agent completes the payment and receives the output, eliminating the need for a human to create a separate account or enter card information.

According to Coinbase, x402 has processed over 165 million payments this year, with a transaction volume exceeding $50 million. Lincoln Mer, Coinbase's AI product lead, estimated that about 99% of x402 payments are made in the dollar-pegged stablecoin USDC.

The transactions primarily involve digital services like API calls, data, and computing resources rather than physical goods like shoes or travel products. Data released by Coinbase in April indicated that the x402 ecosystem includes over 480,000 agents. Based on the disclosed transaction counts and amounts, the average payment size is approximately 30 cents. Most transactions occurred on Coinbase's Layer 2 network, Base.

Mer stated, "The current key product-market fit is the machine-to-machine payments between agents and APIs," adding, "As AI agents move quickly across the internet, a standardized payment method is necessary."

However, the market size is still small. According to CoinDesk, x402 processed about $24 million over the past month, similar to the amount Visa processes in about a minute. Mer described the current state of agent payments as the "Napster and LimeWire era," indicating that while use cases are emerging rapidly like in the early internet, the final economic structure has yet to be established.

Stablecoins for Micro-Payments, Cards for Larger Transactions: The 'Multi-Rail' Competition

Stablecoins are gaining attention in AI agent payments because they are suitable for small, high-frequency transactions. According to Mer, the payment costs borne by card merchants are about 2-4%. As transaction amounts decrease, such as with a few-cent API calls, fixed fees further diminish the economic viability of card payments.

Stephanie Cohen, Chief Strategy Officer at Cloudflare, told CoinDesk, "Stablecoins are particularly well-suited for very small and frequent payments that AI agents are expected to perform initially, such as API calls, data, inference, and content."

Cloudflare recently unveiled 'Cloudflare Wallet' and 'cloudflare.pay'. Users deposit funds into their main wallet and set a separate limit for AI agents, determining the maximum payment amount with approved sellers. Within the established range, AI can make payments without needing human approval each time. The company plans to bundle multiple micro-payments and record them on the blockchain, utilizing escrow as well.

Circle is also testing a micro-payment product called 'NanoPayments' using USDC. MoonPay introduced 'PayBox' on the 29th of last month, connecting Claude and ChatGPT to card and digital asset wallets. Users can approve transactions with a passkey or grant autonomy to agents within a certain limit.

Card companies are not stepping back from the competition. Mastercard is developing 'Agent Pay for Machines' and 'Verifiable Vouchers'. Users set the purchasing scope and limits for agents, and sellers confirm this before providing services.

Sapan Matloi, Mastercard's Senior Vice President of Tokenization and Checkout Services, told CoinDesk, "Agent commerce will be an environment where multiple payment networks are used together," adding, "I believe stablecoins will complement rather than replace existing payment systems."

Cards already have the advantage of being usable at millions of merchants, with established credit, refund, and dispute resolution systems. In contrast, stablecoins have a competitive edge in transactions that are inefficient to process through existing card networks, such as API calls under a few cents.

Ultimately, the AI agent payment market is likely to see payment networks divided by use case rather than one dominating all markets. While stablecoins have secured an early advantage in micro machine-to-machine payments, cards still hold strength in high-value consumption like travel and shopping.

The remaining issue is control and accountability. This is because AI may misinterpret user instructions or follow malicious commands. Coinbase, Cloudflare, Mastercard, MoonPay, and others are designing common safety measures such as balance limits, approved sellers, transaction limits, and requiring human approval for significant transactions.

Bryce Ferguson, co-founder and CEO of Turnkey, told CoinDesk, "Wallet permissions must be restricted to prevent AI agents from causing harm." He likened the current market to the early stages of self-driving cars, where drivers kept their hands on the steering wheel.

For the economy where AI autonomously makes payments to take off, the inconveniences of wallet creation and fund loading must also be resolved. Mer explained that ultimately, he envisions a structure where users can instruct AI to "create a Coinbase agent wallet," and the AI would download the necessary programs from the internet and guide them through login and fund loading processes.