The Race to Build Financial Rails for AI Agents

Ant Digital Technologies, the blockchain arm of Chinese financial giant Ant Group, unveiled Anvita at the company's Real Up summit in Cannes this month. It is a platform designed for one specific purpose: giving AI agents a place to hold assets, find each other, coordinate tasks, and settle payments in real time, without waiting for a human to approve the transaction.

The platform ships in two parts. Anvita TaaS (tokenization-as-a-service) focuses on institutional infrastructure, covering custody and treasury tools for tokenizing real-world assets. Anvita Flow is where the agentic piece lives. Agents on the platform can register, discover other agents, delegate subtasks, and pay for them using USDC stablecoins via HTTP, down to fractions of a cent.

"Pure RWA is just the 'static infrastructure' of digital assets. The real transformation lies in moving toward an onchain agentic economy, where autonomous agents will not just analyze data, they will hold assets, execute trades, and optimize portfolios." — Zhuoqun Bian, president of blockchain business at Ant Digital Technologies

Anvita Flow uses the x402 protocol, developed by Coinbase and Cloudflare, which routes stablecoin payments directly over HTTP. The protocol has already processed more than $600 million in transaction volume and supports roughly 500,000 active AI wallets. Those numbers will look small if Coinbase CEO Brian Armstrong's prediction holds: he expects AI agents to surpass humans in transaction volume before long.

Ant Group is not alone in seeing this as the next infrastructure fight. Visa launched its Trusted Agent Protocol earlier this year, routing AI purchases through card rails. Google debuted an Agent Payments Protocol last September. And Mastercard acquired BVNK, a stablecoin firm, for $1.8 billion in January. That last move was a fairly direct statement about where it thinks the money is going.

The numbers from Solana suggest the trend is already underway. AI agents accounted for $31 billion in payment volume on the network in 2025, and the Solana Foundation has counted more than 15 million onchain agent transactions. McKinsey estimates agentic commerce payments could reach $1.7 trillion by 2030.

There are real governance questions nobody has answered yet. When an AI agent makes a bad financial decision autonomously, who is liable? Nava, a startup that raised $8.3 million in seed funding in April, is building what its CEO calls the "trust side" of the equation: tooling that keeps agents from going off the rails without requiring a human in the loop for every payment. It is a narrow niche today. In a few years, it may be the most important piece of the whole stack.