Stripe co-founders Patrick and John Collison said this month that blockchain networks may need to reach roughly 1,000,000,000 transactions per second to support an economy shaped by autonomous AI agents. Their argument is that current systems simply are not built for the throughput, fee levels, and latency that machine-speed commerce would demand.
The reason this matters is Stripe’s view of “agentic commerce,” where agents negotiate, pay, and settle at machine speed, often via stablecoins. If transactions become continuous micro-payments between machines, today’s throughput and fee economics quickly become a hard ceiling rather than a minor constraint. That gap is what Stripe is using to frame the next set of infrastructure priorities for payments and crypto rails.
— Stripe (@stripe) February 24, 2026
Why Stripe thinks demand could explode
Stripe pointed to internal signals to make the scale jump feel less theoretical. The Collisons said Stripe’s AI coding agents, referred to as “minions,” already merge more than 1,000 pull requests per week. The message is that when software agents are allowed to operate autonomously, activity multiplies fast and stays consistently high.
Stripe also anchored the forecast in its payments footprint and recent stablecoin momentum. It said it processed about $1.4 trillion in payment volume in 2024 and reported a year-on-year doubling of stablecoin volume in 2025, estimating that stablecoins could reach roughly $4.6 trillion across 1,000,000,000 transactions in the first half of 2025. Those figures are presented as early demand signals for what machine-to-machine settlement could look like when the “user” becomes an agent.
Against that backdrop, Stripe contrasted the target with today’s public chain performance. Internet Computer Protocol and Solana were cited at roughly 1,196 and 1,140 TPS respectively, while NEAR has demonstrated 1,000,000 TPS in tests, and BNB Chain has been discussed with ambitions around 20,000 TPS. Even the most optimistic numbers in that set are described as far short of what continuous machine settlement would require across throughput, latency, and fees.
Product positioning and the “payments chain” thesis
Stripe is not positioning this as a purely academic warning. The Collisons referenced an “Agentic Commerce Suite” and a “Tempo” initiative, framing them as steps toward a vertically integrated AI payments stack that spans settlement through wallets. In Stripe’s framing, the winning stack is likely the one that treats agent payments as a first-class workload, not as an edge case layered onto consumer payments.
On the technical path forward, Stripe argued that getting anywhere near the projected capacity likely requires a mix of Layer-2 scaling, sharding, and purpose-built “Payments Chains” designed specifically for high-volume, low-latency payments. The underlying thesis is that general-purpose chains may not be able to simultaneously optimize for extreme throughput, near-zero fees, and consistent subsecond settlement. The company also referenced emerging building blocks such as x402 and identity constructs described as “Know Your Agent” (KYA) as ways to support machine identities and compliance.
Operationally, Stripe’s outlook implies a different risk profile for liquidity and market infrastructure. High-frequency, low-value agent transactions would need near-zero fees and subsecond settlement to avoid friction that compounds at scale. If fees spike or latency drifts upward, agents do not “wait patiently,” they fail the transaction path and degrade the entire utility of the model. Stripe said this dynamic has been visible during memecoin congestion episodes, when delays and fees rose sharply.
What Stripe is really putting on the table is a build-versus-rebuild choice for the ecosystem. If agentic commerce emerges at the volumes it sketches, settlement and custody providers may need to redesign for throughput, latency, and interoperability rather than merely tuning existing networks. In that scenario, engineering for scale, rethinking fee economics, and integrating robust identity and compliance layers for machine actors becomes less of a roadmap item and more of a baseline requirement.
