Stripe warns blockchain must hit 1 billion TPS to support the future wave of AI agents

Stripe says Blockchains may need to Process 1B TPS to Support AI Agents

Financial technology firm Stripe says blockchains may need to process up to 1 billion transactions per second to support the future of artificial intelligence agents. In an annual letter posted to X on Tuesday, Stripe CEO and co-founder Patrick Collison and co-founder John Collison gave a rundown of the firm's performance over 2025, while also making some predictions for the near future. One of the key talking points was the adoption of AI agents and what widespread use could look like in the future. The duo argued that blockchain transaction activity will soon skyrocket as AI agents gradually become the primary conductors of online transactions. However, the Stripe co-founders said there is a significant infrastructure gap in blockchain and said immense scaling is required to meet this incoming demand. "Last year, a memecoin trading frenzy on one of the major blockchains delayed payouts for one Bridge user by over 12 hours and spiked per-transaction prices 35x. While such operational issues are already significant, they will only intensify, for we expect the appetite for transactions to grow a great deal," they wrote, adding: "In our view, agents will most likely soon be responsible for most internet transactions, and we will likely need blockchains that support more than one million -- or even one billion -- transactions per second." According to data from Chainspect, Internet Computer Protocol and Solana are currently the top two blockchains by transaction speed, with roughly 1,196 and 1,140 transactions per second (TPS), respectively. They are the only two on the market currently processing more than 1,000 TPS, and at their peak, they have processed 25,621 TPS and 5,289 TPS, respectively. As it stands, both networks have a theoretical maximum of only 209,708 TPS and 65,000 TPS, respectively. Alongside their predictions on the incoming demand, the Stripe execs also outlined what the main kinds of use cases AI agents will be serving online. Currently, they said that AI agents have moved beyond a phase of "pure hype" into a time of building and "real-world experimentation." Related: Stripe considers acquiring some or all of PayPal: Report They outlined five levels of AI agent capabilities. The first two levels include: completing web forms and descriptive search -- being able to find results for users based on descriptions of situations rather than specific attributes. According to the Stripe execs, AI agents are "hovering on the edge" of levels one and two. The next three levels include: persistence -- remembering user information, requirements, and preferences; delegation -- handling tasks like grocery shopping on your behalf; and anticipation -- being able to provide solutions to problems or schedules without being prompted. "As with the early internet, the future success of agentic commerce is contingent on universal interoperability. Our ascent through the five levels depends on our ability to work together," they wrote, adding: "If all goes well, the little critters won't be cooped up in walled gardens, but will be zooming down the wide-open protocol highways."

Crypto Market Update: Memecoin Surges Show Fragility, AI Agents Could Break Blockchain

The future of Stripe may not just be payments. Founders Patrick Collison and John Collison argue that blockchain infrastructure must scale to 1 billion transactions per second to support the upcoming wave of AI agents. Their warning is clear: as automation becomes autonomous, today's cryptocurrency networks risk becoming bottlenecks rather than backbones. Even the fastest networks have limits. currently peaks around 1,100 TPS in real-world performance. While its theoretical capacity is far higher, it remains nowhere near the billion-TPS threshold Stripe envisions. The gap is not hypothetical. During a trading surge last year, congestion on a major chain led to transaction delays exceeding 12 hours and fees spiking more than 35 times. The episode exposed how fragile are under sudden stress. If speculative trading alone can cause network congestion, AI agents executing millions of automated micro-transactions could overwhelm them entirely.