AI Agents Settle $73M in Crypto Payments Across 176 Million Transactions
Keyrock, the cryptocurrency market-making and research firm, published findings on May 25 showing that AI-powered software agents processed $73 million across 176 million blockchain-based transactions in the 12 months ending April 2026. The average transaction size was just 31 cents. USDC captured 98.6% of all payment volume.
The data marks the first large-scale, independently documented record of autonomous agents using crypto as an operational payment layer rather than a speculative asset.
What the Keyrock Data Shows
The Keyrock report covers transactions where AI agents autonomously paid for computational inputs including API calls, cloud compute time, and data subscriptions. Transaction counts reached 176 million over the measurement period.
That works out to roughly 14.7 million transactions per month, or about 490,000 per day. The average value of 31 cents per transaction is consistent with micropayment use cases where traditional card rails impose minimum fees that make sub-dollar charges uneconomical.
Keyrock said that transaction counts in the tracked dataset grew month-over-month across the full period, with acceleration visible in the final quarter.
Also Read: Iran Appears to Be Backing Down Over Strait of Hormuz, Petraeus Says
Why Stablecoins Dominate Machine Payments
The 98.6% USDC share in the Keyrock data is the single most consequential number in the report. It confirms that volatility is a structural barrier to autonomous agent adoption of non-pegged assets.
An AI agent paying for an API call in Bitcoin (BTC) or Ethereum (ETH) would face unpredictable cost-per-query swings that make budget planning and cost-accounting impossible without a conversion layer. A stablecoin, a cryptocurrency designed to maintain a fixed value against a reference asset such as the U.S. dollar, removes that variable entirely.
The 31-cent average transaction value also sits below the effective minimum for Visa or Mastercard settlement, which typically carries interchange costs of around 1.5% to 3% plus a flat fee. On-chain stablecoin transfers on networks like Solana (SOL) or Ethereum (ETH) Layer-2s can settle for fractions of a cent, making the economics viable at sub-dollar scale.
Also Read: Bitcoin Holds Above $77,000 as Oil Slide Lifts Crypto Sentiment
How We Got Here
The idea of machines paying machines using cryptocurrency circulated as a design concept for years before any deployment-scale evidence existed.
The shift accelerated in 2024 and 2025 as large language models gained the ability to call external tools autonomously, creating demand for programmatic, non-human-authorized payments. Several infrastructure layers converged to make this practical. Coinbase (COIN) launched its AgentKit developer toolkit in early 2025, giving developers a standard interface for embedding wallet functionality into agent workflows.
Stripe re-enabled cryptocurrency payouts for global platforms in the same period, and Google began testing cryptocurrency-denominated API billing in limited programs. The Keyrock research covers the period immediately following those infrastructure releases, which is why the 12-month window captures early but real adoption rather than a projection or pilot.
Also Read: Oil Dips Below $100 on U.S.-Iran Peace Deal Hopes
What Comes Next for Agentic Payments
The Keyrock findings put a concrete number on a market that most institutional analysis had treated as speculative.
Three dynamics will determine whether this scales. First, wallet custody: most agent wallets today are either developer-controlled hot wallets or custodied through exchanges, which reintroduces counterparty risk that on-chain settlement is supposed to eliminate.
Second, regulatory classification: the SEC and FinCEN have not issued formal guidance on whether AI agent payment flows trigger money-transmission obligations for the developers who deploy them. Third, competing rails: Stripe’s stablecoin payment product and PayPal’s PYUSD infrastructure both target programmatic payment use cases, and traditional processors are adding sub-dollar capability to compete with blockchain settlement.
The 31-cent average transaction value suggests the current use case is narrow and concentrated in compute-cost payments. A broader agentic economy, covering things like content licensing, sensor data, or model-to-model inference fees, would require order-of-magnitude growth in both transaction count and total value to become systemically significant.
Read Next: Oil Markets Near Critical Lows as Carlyle’s Currie Warns of July U.S.
Crunch
