The Sui network recorded a peak throughput of 6,086,766 transactions per second (TPS) on July 4, 2026, during a public scalability experiment involving autonomous AI agents. The performance milestone, which occurred on the Sui mainnet, focused on demonstrating the network’s capacity to handle high-frequency activity across gaming, payments, and messaging workflows.
According to an official report from the Sui Foundation, the experiment significantly surpassed its initial target of 1 million TPS. The registered peak is approximately 20 times higher than the network’s previous maximum benchmark of 297,000 TPS, which had been established in a controlled testing environment.
Programmable Tunnels and Operational Implementation
The high-throughput result was facilitated by the use of “programmable tunnels,” which function as off-chain payment and state channels. These channels aggregate activity and settle to the Sui mainnet upon closing, allowing the infrastructure to manage massive transaction volumes without immediate on-chain congestion. During the live event, participants utilized an explorer built specifically for the experiment to observe AI agents interacting in real-time battles and financial exchanges.
This implementation highlights a shift toward supporting “agentic” applications, where autonomous software entities generate high-velocity transaction data. While other networks are also scaling for similar use cases—such as the Fetch.ai Agent Marketplace which recently scaled to 3 million agents—Sui’s test was designed to verify how its specific architectural tunnels behave under extreme public demand.
Infrastructure Reliability
The 6 million TPS figure represents a specific operational peak achieved through specialized state channels rather than sustained global mainnet activity. For users and developers, the result provides a benchmark for how the network might handle future institutional rails or high-frequency retail applications that utilize off-chain settlement layers.
The successful test follows previous operational challenges for the network. Earlier in the year, the Sui mainnet experienced a stall due to a validator consensus issue, which briefly impacted block production. The July 4 experiment appears to serve as a practical demonstration of infrastructure recovery and scaling capability. Further development of AI-specific infrastructure remains a competitive sector, with other Layer 1 projects like Aptos committing $50 million to similar AI agent trading and storage tools.
The Sui Foundation indicated that the data gathered during the livestream will be used to further refine the programmable tunnel architecture and its integration with the main execution layer.
