Why Stablecoins Have Become the Default Asset for Machine Payments
A counterintuitive fact: AI agents choose stablecoins not because "crypto people like using cryptocurrency," but because traditional card payment systems simply cannot handle AI's bills — transactions below $0.30 are unprofitable for card networks, and over 90% of AI agent transactions fall within this range.
The Math Doesn't Add Up for Bank Cards
Payments from AI agents are high-frequency, low-value, and continuous. An agent may make thousands of API calls per hour, each costing a few cents.
Card networks have a fixed cost per transaction — Visa and Mastercard's fixed fees start at about $0.30 per transaction, plus a 2–3% processing fee. If your agent initiates one $0.05 API call per second, the card network swallows $0.35 in cost, you actually pay $0.40, and 87% of that goes into the payment channel's pocket. This business model simply does not work in the machine economy.
The alternative is cross-border wire transfers (SWIFT), where fees can reach up to 6.5%, settlement takes days, and the process relies on manual authorization and KYC verification. AI agents cannot "swipe a card" like humans, nor can they tolerate such waiting times.
A report by Keyrock (crypto market maker) shows that over the past year, AI agents completed approximately 176 million on-chain transactions with a total settlement amount exceeding $73 million, and an average per transaction of only $0.31 to $0.48. About 76% of these transactions were below Visa's fixed-fee threshold of $0.30.
Traditional payment systems were never designed for the scenario of "machines paying machines a few cents." The problem isn't that they are "not good enough" — they are fundamentally unfit for purpose.
Why Stablecoins Work for the Machine Economy
Stablecoins resolve four structural conflicts.
Settlement costs approaching zero
On the Base network, a USDC transfer costs about $0.0001, accounting for only about 0.03% of the average transaction amount. On some Layer 2 solutions, fees are even less than 1 cent.
Settlement in seconds
Card settlement takes T+1 to T+3 days. While waiting for a single card authorization, an agent can execute thousands of stablecoin transactions. Stablecoin transactions achieve final settlement within seconds on EVM chains, and even sub-second times on Solana.
Global reach without intermediaries
When an AI agent calls a US API, purchases data from Singapore, or uses European computing power, it doesn't need to switch payment networks. Stablecoins are native to the internet and do not require SWIFT or correspondent banking systems.
Machine-readable and programmable
Stablecoins can be directly invoked, verified, and transferred by software. Payment logic can be embedded in contracts — funds can be automatically released when a task reaches a key milestone, or billing can occur in real time based on usage.
USDC's Dominance and the Risks
The report shows that 98.6% of AI agent payments are settled in USDC.
This concentration is itself a risk. Keyrock explicitly warns in its report that the entire emerging AI payment system is highly dependent on the regulatory and infrastructure stability of a single stablecoin issuer. If Circle faces regulatory hurdles or technical disruptions, the entire payment chain for AI agents would be directly affected.
Real-World Implementations
Several key developments in the field of machine payments:
Coinbase's x402 protocol allows AI agents to pay directly in USDC for on-chain analytics, cloud services, and other resources, without requiring accounts or subscription systems. Since May 2025, it has settled over 50 million transactions across all supported chains.
Stripe and Tempo launched the MPP (Machine Payments Protocol), enabling automated payment coordination between AI agents and service providers, completing the entire process of request, authorization, and stablecoin settlement without human intervention.
Google's AP2 delegated payment system and Visa's expanded tokenized payment credential service are also positioning themselves in this space.
Confirm you understand the role of stablecoins in machine payments: the next time you see an AI agent boasting that it can "do things for you automatically," ask yourself — what rail does its payment run on? If it's stablecoins, its cost structure can most likely support high-frequency microtransactions. If it still relies on bank cards or pre-funded accounts, its so-called "autonomy" may just be simple automation backed by manual top-ups from humans behind the scenes.
Next step recommendation: if you are developing AI agent-related products, do the math first when evaluating payment solutions — what is the amount per agent transaction? What is the estimated annual transaction volume? If most transactions are below $0.30, you can rule out card-based solutions directly; stablecoins are the only viable choice at this stage.
