Sui for AI: Blockchain infrastructure for multi-agent systems

By the end of 2026, up to 75% of companies are expected to invest in agentic systems, powering autonomous agents that plan, act, and transact with limited human oversight. These AI agents will increasingly control real assets and execute financial decisions in production environments where the consequences are real and financially significant.

Still today only a fraction of organizations report mature controls over what these agents are permitted to do, once they begin operating at scale.

As agents move from analytics to execution, infrastructure constraints become visible. Trades, payments, rebalancing logic, and cross-system coordination all require deterministic settlement and low-latency responsiveness.

Most blockchains were designed for sequential, human-initiated interaction, where a single user submits one transaction at a time, and the network processes those actions in order. Multi-agent systems, however, require many autonomous programs to act simultaneously.

Sui is gaining attention because its object-centric architecture aligns more naturally with these requirements. Rather than forcing many transactions to run at the same time onto an account-based model, it introduces structural primitives that better support independent, simultaneous execution.

Below, we examine why Sui is structurally suited for multi-agent coordination and how pairing it with secure signing infrastructure enables production-grade agent systems.

Why Sui is structurally aligned with multi-agent execution

Combined with ecosystem tooling tailored to agentic finance, Sui is increasingly positioned as an execution layer for autonomous economic systems rather than a general-purpose chain retrofitted for AI.

Sui’s object-centric design and parallel execution model align directly with these requirements. For AI systems managing multiple workflows and assets at once, this architecture reduces friction between offchain decision-making and onchain settlement. It gives each asset or workflow a clearly defined state, ensuring transactions follow predictable execution paths, and allowing independent actions to be processed at the same time without interfering with one another.

Sui’s architecture departs from traditional account-based models by treating digital assets and contracts as first-class objects. This object-centric design changes how state is accessed, modified, and secured.

On many traditional blockchains, including account-based systems like Ethereum, application state lives in a shared global ledger tied to accounts and contract storage. Transactions read from and write to this shared state, and because it is difficult to pre-determine all state interactions, execution is typically sequential. Even unrelated transactions must wait their turn, creating contention under load.

With Sui, each object has explicit ownership and can be modified independently. As a result, unrelated transactions can execute in parallel without competing for the same global state.

For multi-agent systems, this distinction is foundational. If dozens or hundreds of agents are executing trades, managing treasuries, or coordinating workflows simultaneously, they should not block one another unless they are interacting with the same object.

Parallel execution reduces bottlenecks and improves throughput under load. In practice, this makes Sui structurally aligned with machine-speed coordination rather than human-paced transaction flow.

A data model that mirrors multi-agent design

Multi-agent systems are modular by nature. Each agent typically maintains its own state, authority boundaries, and task scope.

Sui’s object-based model mirrors this architecture at the protocol layer. This reduces reliance on fragile shared state and lowers the risk of cascading failures. An issue affecting one object does not automatically compromise unrelated agents or processes.

For fintech builders designing hierarchical agent systems with parent and sub-agents, this structure enables clearer separation of responsibility. Authority can be scoped precisely, supporting safer delegation and more controlled execution paths.

Other features positioning Sui for agentic systems

Beyond its object model and execution engine, Sui’s ecosystem includes additional primitives that are structurally relevant for agentic systems. 

From encrypted data access and decentralized storage to high-throughput settlement and stablecoin integration, these components extend Sui from a transaction layer into a more complete infrastructure for multi-agent financial and AI-driven workflows.

SEAL and programmable encrypted data access

Sui’s ecosystem includes SEAL, a framework for managing encrypted data with onchain-controlled access. Rather than storing sensitive data directly onchain, SEAL ties decryption rights to object ownership and programmable permissions.

For AI agents, this is critical. Agents often require persistent memory, access to private datasets, or controlled data sharing across collaborators.

By anchoring access control to Sui’s object model, SEAL allows encrypted state to be transferred, delegated, or revoked atomically. This supports agent-owned memory, conditional data exchange, and monetized outputs without exposing plaintext onchain.

In multi-agent financial systems, combining programmable execution with programmable data access strengthens both coordination and security. 

High throughput and low latency for real-time coordination

Agent systems frequently operate in environments where latency directly impacts outcomes. Trading agents, arbitrage systems, and automated payment routers cannot tolerate multi-minute settlement windows or unpredictable confirmation times.

Because of its object-centric design, Sui’s execution engine is made for high throughput and fast finality. In controlled validator performance tests, Sui has demonstrated throughput ranging from ~10,800 to as high as ~297,000 transactions per second. 

As agent populations scale from dozens to thousands of autonomous actors, these performance characteristics become structural constraints. Parallel execution capacity and sub-second finality are not incremental optimizations for multi-agent systems. They are baseline requirements for reliable, real-time coordination.

Composability without fragile coupling

Composability is essential in DeFi and fintech systems. Agents need to interact with shared liquidity pools, contracts, and protocols, but they also need clear control over their own funds and decision logic. In multi-agent environments, these interactions happen continuously and often at the same time.

On Sui, assets and contracts are structured as distinct objects with clearly defined ownership and access rules. This means an agent can use or interact with a shared pool or contract without becoming tightly entangled with every other participant in the system. 

For multi-agent financial systems, this separation limits the blast radius of failures while still allowing agents to build on shared infrastructure like pools, vaults, and settlement rails. Agents can cooperate and compose across protocols without turning the entire system into a single fragile dependency chain. As the number of agents and transactions grows, this clear boundary between collaboration and isolation becomes critical for safe, scalable operation.

‍

Walrus and decentralized data infrastructure for AI

Sui’s ecosystem includes Walrus, a decentralized data layer built to support large-scale data storage alongside onchain execution. While Sui manages transactional state through its object model, Walrus is designed to handle high-volume data such as model artifacts, agent logs, datasets, and long-form memory.

AI agents require more than token transfers and contract calls. They generate context, intermediate reasoning traces, embeddings, and historical records that cannot realistically live inside the base-layer state.

Walrus allows agents to anchor references to stored data onchain while keeping the underlying data in a scalable, decentralized storage layer. This preserves auditability and provenance without burdening the execution engine.

For multi-agent systems, this becomes a shared memory substrate. Agents can reference common datasets, publish outputs, or coordinate around structured data objects while maintaining separation between execution logic and data persistence.

Stablecoins and economic primitives for agent workflows

Agent-based financial systems require reliable units of account and predictable settlement rails. Sui’s ecosystem includes established stablecoins and core DeFi primitives such as onchain liquidity pools, lending protocols, and automated market makers that agents can programmatically access.

Using stablecoins as the settlement layer allows agents to denominate contracts, subscriptions, and cross-agent payments in assets designed to track fiat value. This reduces balance sheet volatility and simplifies accounting logic for recurring payments, revenue sharing, escrow flows, and automated treasury rebalancing. Low transaction costs and fast finality further make it feasible for agents to execute high-frequency or small-value transactions without eroding margins.

For fintech and AI-driven financial applications where predictable settlement value is mandatory, these economic primitives are not optional features. They form the base layer that allows agents to coordinate economically, manage capital programmatically, and operate continuously without constant human intervention.

Sui support on Turnkey: Secure signing for production agent systems

Turnkey provides production-grade infrastructure for teams building on Sui, including secure key generation, address derivation, and transaction signing. Turnkey’s infrastructure derives Sui addresses, enabling Agentic Wallets to sign raw Sui transactions through secure APIs without ever exposing private keys. 

For multi-agent systems, this separation is critical. Each autonomous agent can control its own wallet or operate with delegated authority, and Turnkey ensures that transactions are signed according to policies you define rather than ad hoc application logic.

Turnkey’s policy engine enables programmable controls over signing behavior. Organizations can enforce transaction constraints, approval flows, and scoped permissions for agents,  preventing unauthorized interactions even in complex multi-agent environments.

By combining Sui’s parallel, object-centric execution with Turnkey’s secure signing, policy controls, and multitenant key isolation, AI and fintech teams can build scalable multi-agent systems that are not only fast and composable but also production-ready and secure. 

Turnkey for Sui: Multi-agent systems and the new economy

AI systems are moving from recommendation to execution. As agents begin coordinating capital, interacting with protocols, and settling value autonomously, infrastructure becomes a gating factor for safety and scale.

Multi-agent systems require more than throughput. They demand parallel execution without contention, explicit ownership boundaries, composability without cascading risk, encrypted data access, and scalable storage for agent memory. Sui’s object-centric architecture, combined with ecosystem components like SEAL and Walrus, aligns directly with these structural requirements.

Execution, however, is only one half of production readiness. Autonomous systems must also be governed. Secure key isolation, programmable signing policies, and enforceable transaction constraints ensure that agent activity remains bounded and auditable.

By pairing Sui’s execution model with hardened signing infrastructure such as Turnkey, AI and fintech teams can move from experimentation to deployment. The result is not just faster agents, but coordinated, policy-enforced multi-agent systems capable of operating safely in real financial environments.

Get started with Turnkey today. 

‍

‍

‍

‍

‍