Stripe founders Patrick Collison and John Collison used their latest annual letter to make a blunt point: if AI agents end up handling most online transactions, blockchain infrastructure will need to scale to 1 billion transactions per second. By their framing, no existing chain is even close to that threshold.
Current throughput is nowhere near the level Stripe is describing
The letter cited Chainspect data showing that only two blockchains currently record real-world throughput above 1,000 TPS. Internet Computer Protocol, or ICP, is running at about 1,196 TPS, with a historical peak of 25,621 TPS and a theoretical limit of 209,708 TPS. Solana is at about 1,140 TPS, with a historical peak of 5,289 TPS and a theoretical cap of 65,000 TPS.
Even if Solana could operate at its stated theoretical maximum, it would still sit more than 15,000 times below a 1 billion TPS target. Using current live throughput, the gap stretches into the million-fold range. That is the scale Stripe is pointing at as AI-driven commerce moves from experimentation into execution.
A memecoin surge already showed how fragile capacity can be
The Collison brothers pointed to a concrete example from last year, when a memecoin trading wave pushed a major blockchain close to failure. According to the letter, the congestion caused payment delays of more than 12 hours for users of Bridge, Stripe’s stablecoin platform, while per-transaction costs jumped by 35 times.
The warning is simple. Bottlenecks like this are not temporary edge cases if AI agents begin sending large volumes of automated payment and settlement requests on-chain. The letter says the industry will need blockchains capable of supporting more than 1 million, and potentially 1 billion, transactions per second.
Stripe maps AI agents into five capability levels
The annual letter also outlined how Stripe sees the commercial path for AI agents. It breaks the progression into five levels. The first two are closest to deployment today: Level 1 covers form filling, such as handling checkout and sign-up flows automatically; Level 2 covers descriptive search, where an agent can find results from natural-language prompts instead of exact keywords.
Levels 3 through 5 remain ahead. Level 3 is persistent memory, allowing agents to retain user preferences and history. Level 4 is delegated task execution, such as completing recurring shopping or planning work. Level 5 is proactive anticipation, where an agent acts before being explicitly asked. Stripe said this model depends on broad interoperability, with agents operating across open protocols rather than inside closed ecosystems.
Stripe has been expanding its crypto and payment stack
The message carries extra weight because Stripe has been making moves in this direction. The source material says Stripe acquired stablecoin infrastructure company Bridge for $1.1 billion in October 2024, then bought crypto wallet developer Privy in mid-2025. It also notes market reports that Stripe and Paradigm have been working on a blockchain called Tempo, designed as a high-performance chain for payment use cases.
Another step already disclosed is ACP, the Agentic Commerce Protocol, which Stripe launched with OpenAI last year. Taken together, those actions place Stripe’s infrastructure warning in a broader strategy: AI agents may increase transaction demand far beyond what public chains can process today, and payment companies are positioning around that gap now.