Sui recorded a peak of 40,614,180 transactions per second during a live Basecamp demonstration in Singapore, but the figure reflects activity inside programmable offchain tunnels rather than conventional Layer 1 throughput. According to the Sui Foundation’s official results, users and AI agents generated the 40.6 million TPS peak inside Sui Tunnels, with final channel states settling back to Sui mainnet. The result exceeded the demonstration’s 20 million TPS target and the previous tunnel benchmark of 6,086,766 TPS recorded in July.
More than 10,000 tunnels were opened on Sui mainnet within seconds during the October 7 event, according to the Foundation. Participants then used those channels for payments, games and chat interactions without submitting every intermediate action to the base chain. The benchmark therefore measures the aggregate execution capacity of the tunnel architecture, not the number of ordinary Sui transactions validators processed each second.
Tunnels Move High-Frequency Execution Offchain
Sui Tunnels operate broadly like state or payment channels. Participants commit to the channel onchain, exchange mutually authorized state updates offchain and eventually submit the resulting state for settlement. Sui says a basic tunnel requires two blockchain transactions, one to open and another to close, while potentially allowing a much larger number of interactions between those events. That architecture shifts repeated execution away from consensus while retaining an onchain settlement and verification layer.
Unlike payment-only channels, Sui describes its tunnels as programmable. The Basecamp demonstration included chat, gaming and payment applications, while every closed tunnel was mutually cosigned and designed to be independently verifiable onchain. The architecture is being positioned alongside Sui’s wider machine-commerce work, including pay-per-call AI services planned with Alibaba Cloud. The underlying thesis is that software agents may generate interaction frequencies that are uneconomical to settle individually on a Layer 1.
The same distinction also separates tunnel throughput from Sui’s previous base-layer performance benchmarks. Before the July tunnel experiment, Sui cited roughly 297,000 TPS from a controlled benchmark environment. The July 6.08 million result and October 40.6 million result were both produced inside programmable tunnels. Neither figure establishes that Sui mainnet can execute millions of normal onchain transactions per second.
CertiK Review Remains an Important Qualification
CertiK participated in the Basecamp attempt as an independent auditor. Sui says every counted transaction was written into a cryptographic transcript and that CertiK observed the demonstration in real time. However, the Foundation also said CertiK would subsequently examine the proofs, execution logs, journals and other supporting evidence before publishing a report. The detailed independent verification is therefore still forthcoming even though Sui describes the live result as independently analyzed.
The experiment fits Sui’s broader effort to separate user-facing activity from unnecessary blockchain friction. The network is already supporting gasless stablecoin settlement for Daya and has been incorporated into regulated stablecoin payment infrastructure through Remi. Tunnels extend that approach by reducing how frequently high-volume applications need to touch consensus at all, rather than simply reducing the cost of each individual onchain transaction.
The Basecamp result consequently demonstrates the capacity of a specific scaling architecture more than raw Layer 1 performance. Sui successfully coordinated more than 10,000 mainnet-opened tunnels while the offchain system reached 40.6 million state transitions per second under demonstration conditions. The meaningful adoption test will be whether production applications generate sustained high-frequency activity through tunnels and settle economically meaningful state back to Sui, not whether the benchmark can be presented as 40 million TPS of ordinary mainnet traffic.