Sui Showcases Atomic Transactions for AI Agents at Basecamp, Cites 6 Million TPS Peak
Key Takeaways
- •Sui showcased Programmable Transaction Blocks at Basecamp as a way to support autonomous AI agents with atomic transaction execution.
- •The PTB design can execute up to 1,024 Move function calls in one transaction and reverts fully if any step fails.
- •Sui says transactions finalize in about 400 milliseconds, and livestream experiments reached a peak of 6,086,766 transactions per second.
- •Sui is a launch partner for Google’s Agentic Payments Protocol, which aims to let AI agents make secure and verifiable transactions.
- •The network’s atomic transaction model is also presented as useful for multi-party settlements, conditional lending, and other complex workflows.

Blockchain networks have spent years promising to support the next wave of AI applications. At Basecamp, its flagship developer conference, the Sui Network moved that promise past the pitch-deck stage with a working demonstration of atomic execution built for autonomous agents.
The showcase centered on Sui's Programmable Transaction Blocks (PTBs), which can execute up to 1,024 Move function calls inside a single atomic transaction. Move is a smart-contract language originally developed at Meta for the shelved Diem stablecoin project, and Sui itself was built by Mysten Labs, a team founded by former Meta blockchain engineers. In practice, that allows an AI agent to authenticate, pull data, execute financial logic, and settle in one action that either completes in full or rolls back entirely if anything breaks.
The behavior is comparable to a vending machine that refuses to take a customer's money unless it can also guarantee delivery: no partial states, no stuck funds, no half-executed workflows.
Why atomic execution matters for AI agents
AI agents are increasingly designed to operate autonomously, making decisions and executing multi-step processes without a human signing off at each stage. The problem is that most blockchains were not built for that kind of sequential, interdependent logic.
Transactions on Sui finalize in approximately 400 milliseconds. For reference, that is fast enough to fit inside the gap between two heartbeats — a margin that matters considerably when autonomous agents are making real-time decisions in financial or data-sensitive environments.
The network's scalability numbers are harder to ignore. During livestream experiments, Sui recorded a peak throughput of 6,086,766 transactions per second, driven specifically by AI agent operations.
Google partnership and the agentic payments layer
Sui has also secured a strategic position in what may become a defining infrastructure layer for AI commerce: the network is a launch partner for Google's Agentic Payments Protocol (AP2), a framework designed to enable AI agents to conduct transactions securely and verifiably, alongside other launch partners drawn from the payments and blockchain industries.
Google is not the only major player working on this problem. Coinbase's x402 protocol, introduced in 2025, repurposes the long-unused HTTP 402 "Payment Required" status code for machine-to-machine payments, one of several parallel efforts now forming around agent commerce.
The core of that argument rests on verifiability. Autonomous AI systems require execution environments where any outside party can confirm what happened and when. Sui's on-chain state model and atomic transaction structure provide exactly that kind of shared, auditable record.
Basecamp as ecosystem signal
Upcoming Basecamp events are scheduled for Dubai in May 2025 and Singapore in October 2026, pointing to a deliberate push into markets with significant institutional and regulatory interest in both AI and blockchain infrastructure.
For developers, the PTB architecture is meaningful beyond AI applications. Any workflow that requires complex, conditional logic executed reliably — multi-party settlements, conditional lending, or composable DeFi strategies — benefits from the same atomic guarantees.
The 6-million-plus TPS figure, achieved through AI agent-driven load during a controlled experiment rather than live public mainnet traffic, sets a benchmark that most competing networks would struggle to match under equivalent conditions. The pace at which third-party developers adopt PTBs for AI agent use cases, and whether the Google AP2 partnership translates into measurable transaction volume, remain key questions for market participants tracking the network.