Sui’s programmable offchain tunnels reached a peak of 40,614,180 transactions per second during a live stress test designed to simulate high-frequency activity between AI agents, the layer-1 blockchain said. The demonstration, held on Wednesday at Sui Basecamp in Singapore, marked a sharp increase from a similar test in July that reached about 6.1 million TPS.

The result puts a spotlight on a question that is becoming central to blockchain infrastructure: how networks will cope if autonomous software agents, rather than humans, become the dominant source of transactions. Sui’s answer is not to force every interaction onto the base chain. Instead, its tunnels are programmable channels that open on Sui mainnet, process transactions offchain, and settle back to the blockchain when closed.

Key facts

• Headline: Sui tunnels hit 40.6M TPS in a live AI agent stress test.
• Peak throughput: 40,614,180 transactions per second.
• Location: Sui Basecamp in Singapore.
• Previous benchmark: About 6.1 million TPS in July.
• Mainnet activity: More than 10,000 tunnels were opened during the test.
• Use cases: Payments, games and chat applications.
• Audit: CertiK served as an independent auditor; a report is expected in the coming days.
• Key caveat: The figure measures offchain tunnel throughput, not base-layer onchain TPS.

During the test, more than 10,000 tunnels were opened on Sui mainnet, with activity spanning payments, games and chat applications. That detail matters because it suggests the benchmark was not limited to a single narrow workload. Payments, gaming actions and chat messages have different transaction profiles, and a stress test that combines them is closer to the mixed traffic a real agent economy might produce.

Still, the headline number requires context. The 40.6 million TPS figure refers to activity inside Sui’s offchain tunnels, not to the base-layer throughput of the Sui blockchain itself. Offchain channels can aggregate and process many interactions before settling a smaller number of transactions onchain. That design can be powerful for high-frequency use cases, but it is not the same as saying Sui’s layer 1 processed 40.6 million onchain transactions per second.

Sui described the tunnels as programmable channels designed to handle high-frequency activity, including interactions between AI agents, without requiring every transaction to be processed directly onchain. In practice, that means a tunnel can be opened onchain, used for rapid offchain transactions, and then closed with a settlement back to Sui. The model resembles payment channels and state channels, but with a greater emphasis on programmability and on serving automated, high-volume applications.

The distinction between offchain execution and onchain settlement is important for evaluating the benchmark. Offchain systems can avoid the latency and cost associated with broadcasting every transaction to a global blockchain. They can also batch or net activity before final settlement. The trade-off is that security, dispute resolution and settlement guarantees depend on how the tunnel system is designed and how effectively it is enforced onchain when the channel closes.

Sui’s broader pitch is that the network is built for a future in which millions of AI agents transact continuously. Adeniyi Abiodun, co-founder and chief product officer at Mysten Labs, the original contributor to Sui, framed the test around that thesis. “Sui is built for the moment when millions of AI agents are transacting every day,” he said. He added: “No market in the world needs 6 million transactions per second, let alone 40 million: people just don’t move that fast. Agents do.”

That argument reflects a growing focus across the blockchain industry on machine-to-machine commerce. AI agents could eventually pay for data, compute, application programming interface calls, content, gaming actions and other digital services. Many of those interactions would be small and frequent, and some would occur in bursts that human-driven payment systems were never designed to handle. If agents are to operate continuously, they will need infrastructure that can process a high volume of low-value transactions without congesting the underlying chain.

At the same time, AI-agent transactions are not automatically meaningful just because they are fast. The value of a high-throughput system depends on whether it can provide reliable settlement, clear security guarantees and a workable economic model. A stress test can demonstrate raw capacity, but it cannot by itself prove that a network will perform reliably under sustained real-world demand. That is why independent verification of the test data is likely to be closely watched.

CertiK served as an independent auditor during the demonstration and is reviewing the test data, with a report expected in the coming days. The audit is significant because TPS claims are notoriously difficult to compare across networks. Throughput depends on transaction size, batching methods, hardware, software configuration and whether the transactions are simple transfers or complex smart contract calls. An offchain tunnel test also measures a different kind of activity from a base-layer benchmark, so the numbers should not be read as direct equivalents.

Sui itself is a layer-1 blockchain whose original contributor is Mysten Labs. The network has emphasized scalability and its object-centric data model, which is intended to support parallel execution of certain transactions. Tunnels fit into that broader strategy by offering an offchain path for workloads that do not need every interaction settled immediately on the base chain. For high-frequency agent activity, the appeal is straightforward: keep the fast interactions offchain and use Sui for settlement and security when channels close.

The jump from about 6.1 million TPS in July to 40,614,180 TPS in the latest test is notable, but it also raises questions about what changed. The network said the test was live and involved more than 10,000 tunnels on mainnet, with activity across payments, games and chat. Those details suggest a broader workload than a simple transfer benchmark, but the pending audit report should provide more clarity on the test setup, the transaction mix and the conditions under which the peak was achieved.

For the AI-agent narrative, the test is a signal that blockchain teams are preparing for a world in which software programs, not people, generate the bulk of transactions. That world is still speculative. Agents today are often limited to narrow tasks, and most consumer and enterprise payment flows still depend on human authorization and traditional financial rails. But the direction of travel is clear: as agents become more capable, the infrastructure they use will need to handle volume, speed and programmability at a scale that existing payment systems were not designed for.

Sui’s tunnel test does not settle the debate over whether offchain channels are the best way to serve that future. Competing approaches include higher-throughput base layers, rollups, specialized execution environments and other forms of scaling infrastructure. What the test does show is that Sui is trying to position its tunnel architecture as a practical option for high-frequency agent activity, and it is submitting the results to independent review rather than relying solely on its own claims.

The next milestone is the publication of the audit report, which should provide a clearer view of how the test was conducted and what its results mean for Sui’s AI-agent ambitions. Until then, the 40.6 million TPS figure stands as a headline benchmark with an important qualifier: it was achieved through programmable offchain tunnels, not through direct onchain processing on Sui’s base layer.

Source:Cointelegraph News