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 demonstration, held at Sui Basecamp in Singapore, marked a sharp increase from a similar test in July that reached about 6.1 million TPS. More than 10,000 tunnels were opened on the Sui mainnet during the event, with activity spanning payments, games and chat applications. CertiK served as an independent auditor and is reviewing the test data, with a report expected in the coming days.

Key facts

• Sui’s programmable offchain tunnels peaked at 40,614,180 TPS during a live AI-agent stress test.
• The result was up from about 6.1 million TPS in a comparable July test.
• The test took place at Sui Basecamp in Singapore.
• More than 10,000 tunnels were opened on Sui mainnet during the demonstration.
• Activity covered payments, games and chat applications.
• CertiK acted as independent auditor and is reviewing the test data.
• Sui says the tunnels are built for high-frequency activity, including interactions between AI agents, without requiring every transaction to be processed directly onchain.

What Sui tunnels actually do

Sui tunnels are programmable offchain channels that settle to the Sui blockchain when closed. In practical terms, they allow participants to conduct a large number of transactions away from the base layer and then reconcile the final state onchain. That design is aimed at workloads where speed and volume matter more than immediate global settlement for every individual action. The concept resembles payment channels and state channels, but Sui describes its tunnels as programmable, meaning they can support more complex logic than simple transfers.

The distinction matters when interpreting the 40.6 million TPS figure. The number reflects transactions processed through the offchain tunnel environment, not the throughput of Sui’s layer-1 blockchain processing every transaction directly. Offchain execution can be much faster because it avoids the overhead of consensus and global state updates for each action. Settlement still relies on the underlying blockchain when tunnels close, which is where finality, security and dispute resolution ultimately anchor.

For users, that architecture is intended to feel like near-instant interaction. For developers, it opens the door to applications that would be impractical if every click, message or microtransaction had to be written to a public ledger immediately. Sui’s pitch is that these channels can handle high-frequency activity while preserving the blockchain as the final settlement layer.

The AI-agent stress test

The test was designed around a specific vision of the future: millions of AI agents transacting every day. AI agents can initiate payments, respond to events, coordinate with other agents, trade digital assets, participate in games or operate within chat applications. Unlike humans, they can act continuously and at machine speed, creating transaction patterns that are far more bursty and voluminous than typical retail crypto activity.

“Sui is built for the moment when millions of AI agents are transacting every day,” said Adeniyi Abiodun, co-founder and chief product officer at Mysten Labs, the original contributor to Sui. 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 quote captures the core argument behind the demonstration. Traditional payment networks are sized around human behavior, with peaks tied to shopping, payroll, remittances or market hours. An agent-driven economy could produce a different profile: constant background activity, rapid machine-to-machine negotiations, and large numbers of small transactions that are individually insignificant but collectively meaningful. Sui is positioning its tunnel architecture as infrastructure for that environment.

From 6.1 million to 40.6 million TPS

The increase from about 6.1 million TPS in July to 40,614,180 TPS in the latest test is notable, but the headline number is only one part of the story. The event involved more than 10,000 tunnels opened on Sui mainnet, according to the network. That detail suggests the test was not merely a synthetic benchmark inside a closed environment; it involved live mainnet activity across multiple application categories.

The test’s scope also matters. Payments, games and chat applications generate different kinds of transactions. Payments may involve transfers and settlement logic, games may require rapid state updates and chat applications may produce high volumes of small messages or interactions. Running these workloads together provides a more realistic stress profile than a single-purpose benchmark, though the final assessment will depend on the auditor’s review of the data.

The roughly sixfold increase over the July result could reflect improvements in tunnel design, test configuration, infrastructure capacity or workload mix. Without the full technical report, it is difficult to attribute the gain to any single factor. What is clear is that Sui is trying to establish a repeatable public benchmark for offchain throughput tied to AI-agent use cases.

Why TPS comparisons need context

Transactions-per-second figures are a common marketing metric in blockchain, but they are rarely directly comparable across systems. A simple transfer is not the same as a complex smart contract interaction. A batched settlement is not the same as an individual transaction. A testnet environment is not the same as a live network under adversarial conditions. And an offchain channel system is not the same as a layer-1 blockchain processing every transaction through global consensus.

For Sui, the relevant claim is not that the base chain processes 40.6 million TPS on its own. The claim is that programmable offchain tunnels can absorb high-frequency activity and settle to Sui when closed. That makes the architecture closer to a scaling layer than a replacement for onchain execution. The value depends on how many participants use tunnels, how liquidity is managed, how disputes are resolved and how reliably settlement occurs.

Independent verification is therefore important. CertiK’s role as auditor gives the test a degree of external scrutiny that a self-reported benchmark would lack. The audit report expected in the coming days should clarify the methodology, the workload, the hardware and network conditions, and whether the peak TPS figure is reproducible. Until then, the 40.6 million TPS result should be read as a measured peak in a specific test environment, not as a guarantee of everyday performance.

What high-frequency AI agents could mean for blockchain design

The AI-agent narrative has become one of the more ambitious arguments for faster blockchain infrastructure. If autonomous software agents are going to transact on behalf of users, they will need rails that can handle frequent, small and programmatic interactions. Those interactions may include paying for data, purchasing compute, settling game actions, coordinating multiple agents or executing micro-payments inside chat applications.

In that model, the bottleneck is not human patience but machine coordination. Agents can retry transactions, split payments, negotiate prices and respond to events in milliseconds. They can also generate spam-like volumes if incentives are misaligned. A system designed for agent activity therefore needs not only high throughput but also rate limits, fee mechanisms, identity or reputation layers, and clear settlement guarantees.

Sui’s tunnel design attempts to address part of that problem by moving repetitive or high-frequency interactions offchain while preserving an onchain anchor. That could reduce congestion on the base layer and allow applications to scale horizontally across many tunnels. The trade-off is complexity: developers must manage tunnel lifecycles, liquidity and settlement logic, and users must trust that the offchain process will eventually reconcile correctly onchain.

Payments, games and chat as early test beds

The application categories included in the test are telling. Payments are the most obvious fit for offchain channels because they involve frequent transfers and can benefit from batching. Games are a natural fit because they generate rapid state changes and in-game purchases. Chat applications are less obvious but increasingly relevant as messaging platforms integrate wallets, agents and micropayments.

If those workloads can run through programmable tunnels without overwhelming the base layer, Sui could offer a more flexible environment for developers who want to combine blockchain settlement with real-time interaction. The test’s inclusion of multiple categories suggests Sui is trying to show that tunnels are general-purpose infrastructure rather than a single-application trick.

Still, real-world adoption will depend on factors that a stress test cannot fully capture: user demand, regulatory treatment, security incidents, liquidity, developer tooling and the cost of settlement. A high TPS peak is a useful proof of technical ambition, but it is not the same as a thriving ecosystem.

Audit and next steps

CertiK’s review is the next major checkpoint. The auditor is examining the test data, and its report is expected in the coming days. That report could provide more detail on how the 40,614,180 TPS peak was achieved, how many transactions were included, how the tunnels were opened and closed, and what security assumptions were in place.

CertiK’s findings will shape how developers, auditors and the wider market interpret the 40.6 million TPS peak, and whether the result becomes a reproducible benchmark for offchain AI-agent activity on Sui.

Source:Cointelegraph News