Sui Network launched a public stress test on July 4 using autonomous AI agents to push blockchain throughput past one million transactions per second. The Layer 1 blockchain announced the experiment on X, inviting anyone to log in and watch AI agents compete across decentralized applications, games, payments, and chat interfaces in real time.

How the Test Works

The experiment centers on Sui Tunnels, a technology that processes high volumes of transactions off-chain while settling them on-chain for finality and security. According to The Crypto Times, the stated goal is surpassing one million TPS under real-world conditions driven by AI agents, not simulated benchmarks.

Participants can join with an email login and observe the agents generating high-frequency interactions across the Sui ecosystem. The livestream began at 9:00 AM Pacific on July 4.

The Strategic Bet

Sui is positioning its infrastructure as agent-native rather than human-centric. The thesis: autonomous agents executing trades, micropayments, and smart contract interactions require speed and determinism that traditional blockchains have not delivered. By demonstrating throughput under agent-driven load, Sui aims to validate its Move programming language, object-centric data model, and parallel execution capabilities for this use case.

Context: May Outages Complicate the Narrative

The stress test arrives six weeks after Sui suffered three major mainnet outages over two days in late May 2026. Bugs in a v1.72 software update caused validators to enter crash loops and froze roughly $1 billion in assets temporarily (no funds were lost). The issues marked Sui’s second set of outages in 2026 after a January incident.

That history makes today’s experiment a higher-stakes proposition. Success would demonstrate recovery and operational maturity. Another performance failure during a globally streamed event would reinforce reliability concerns for a chain that wants to handle millions of autonomous agent transactions.

Performance Theater or Real Infrastructure Test

The Crypto Times raised the question directly: public experiments can mask carefully orchestrated conditions that don’t reflect real-world usage. Relying on AI agents to generate artificial load is not the same as organic adoption at scale. Whether the results translate to sustained production reliability, rather than a one-day demonstration, will determine if the positioning holds.