Stripe co-founder just stated that blockchain may need 1 billion TPS to support an AI agent-driven future.

This highlights the importance of this number—and the current reality of the industry.
Patrick and John Collison wrote in their 2025 annual letter that AI agents could soon handle most internet transactions, while current blockchain infrastructure is far from ready.
They provided a real-world example: on a major blockchain, a memecoin surge caused payment delays of over 12 hours and transaction fees to spike 35 times.
Chainspect shows the current state:
→ Solana: ~1,140 TPS (theoretical max: 65K)
→ ICP: ~1,196 TPS (theoretical max: ~210K)
→ Most networks: throughput around 1,000 TPS
This is several orders of magnitude below 1 billion.
Now consider this: in April 2025, when Qubic officially launches its mainnet on Layer 1, transaction volume reached 15.52 million TPS, independently verified and published by CertiK. No rollups, no Layer 2 dependencies, zero fees, instant settlement.
This is not a testnet simulation but a real mainnet stress test, generating 1.518 billion transactions and verified by one of the industry’s most respected auditing firms.
Just 15.52 million TPS is not enough to close the gap with 1 billion TPS, but it’s the closest any active network has achieved so far—and the gap is significant. The architecture behind it (tick-based consensus mechanism with atomic execution and finality) is specifically designed to meet the high capacity and real-time computing demands of AI workloads.
Discussions about infrastructure are shifting. Stripe has just told us what the future standard should be. The question now is: which networks are working toward this goal.