Friday, 16 October 2026
SubmitMy events
Technology

Yat Siu says blockchain will be the settlement layer for AI agents

The Animoca Brands chairman says banks cannot serve AI agents, but card networks are already building payments for machines.

Key points
  • Yat Siu said on Oct. 7, 2026, that blockchains, not banks, will settle payments made by AI agents.
  • AI agents made more than 138 million payments over the x402 protocol, most under 10 cents, Chainalysis found.
  • Mastercard launched Agent Pay for Machines in June 2026, putting card rails into the same race.

Yat Siu, co-founder and executive chairman of Animoca Brands, says AI agents will settle their payments on blockchains because banks are not built to serve them.

He makes the case as AI agents have made more than 138 million payments over the x402 protocol, according to Chainalysis data reported by Blockhead on Oct. 15, 2026.

Siu set out his view on stage at TOKEN2049 in Singapore on Oct. 7, 2026, in a keynote titled "When Games Become Reality: The Return of the Metaverse". Animoca Brands builds and invests in blockchain games and digital-property projects.

Machines start to pay each other

Agent payments are tiny and frequent. More than 87% of x402 payments fell between $0.0001 and $0.10, according to Chainalysis, and most ran on Base, Coinbase's network.

The dollars that settle them are already on-chain. Stablecoins worth $303.1 billion circulated across blockchains on Oct. 16, according to DefiLlama.

The Federal Reserve is watching the same shift.

Governor Christopher Waller said on Sept. 29 that agents shopping around the clock will need many small payments, and that micropayments favor rails with low flat fees, in a speech on payments in the age of AI agents.

'Designed for machine speed'

Siu's core claim is that blockchains fit machines better than people. "We think that blockchain is the native settlement layer for the AI economy," he said.

"The blockchain was not designed for human speed. It was designed for machine speed," Siu said.

He described crypto as "the perfect financial system for machines designed for 24/7 operations," because tokens are bearer assets that an agent can hold and send without a branch or a bank account.

The check: the activity is real but small in dollars. Chainalysis did not publish a total dollar value for the 138 million payments, and most of them were worth less than 10 cents.

Can banks serve an agent?

"Traditional finance cannot bank an AI," Siu said. Crypto wallets, in his view, let agents hold money, earn yield and pay without a bank account.

The check: card networks are moving into the same space. Mastercard launched Agent Pay for Machines on June 10, 2026, a service that lets AI agents and machines pay each other, including fractions of a cent, the company said.

More than 30 industry leaders are among the first to adopt or support the service, Mastercard said.

Waller described the same pattern in bank terms: a buyer could give an agent a pre-funded virtual card to shop for groceries. Banks may not open accounts for agents, but card rails already have a way in.

'Who owns the agency of the agent'

Siu's second argument is about control. He said he runs 400 agents himself and expects people to use several, one for family, one for friends, one for work. "But actually, do you really want one agent managing everything for you?" he asked.

He warned that platforms decide what happens to the agents built on them. "They have the right to remove your agent at any time," Siu said of large AI platforms' terms of service.

"The question is no longer who owns your data. It's who owns the agency of the agent," he said, and his answer was blunt: "we need to own our agents."

His estimates are large. Siu said 6 billion to 7 billion people online, with about 10 agents each, could mean 50 billion to 100 billion agents, his own projection.

By the numbers

MetricValueAs ofSource
x402 payments by AI agentsmore than 138 millionOct. 15, 2026Chainalysis via Blockhead
Share of x402 payments from $0.0001 to $0.10more than 87%Oct. 15, 2026Chainalysis via Blockhead
Stablecoins on all chains$303.1 billionOct. 16, 2026DefiLlama
Mastercard Agent Pay for Machines launchJune 10, 2026June 10, 2026Mastercard
Agents Siu says he runs400Oct. 7, 2026Yat Siu

The other side: trust and liability

The Fed sees trust, not technology, as the main barrier. "Who is on the hook if an agent makes the wrong purchase?" Waller asked in his Sept. 29 speech.

He drew a line between two models. In agent-assisted commerce the buyer stays in control and pays. In agent-delegated commerce the agent shops and pays alone, which Waller said needs far more trust mechanisms and guardrails.

He named authentication, liability and fraud as the challenges, and said fraud models built for human behavior will need recalibrating for agents. Those problems do not disappear on a blockchain, where a mistaken transfer is hard to reverse.

Siu's 50 billion to 100 billion agents is his own projection. We found no independent count of agents in use today.

What to watch: standards and the first agent wallets

The first signal is standards. Waller asked whether agentic commerce will settle on platform-specific or interoperable standards. Open standards would favor Siu's view that people should own their agents.

The second is dollar volume. If x402 and similar protocols report rising dollar totals, not only payment counts, the settlement-layer argument gets stronger.

The third is Mastercard's machine payments service. Its adoption will show whether agents choose card rails or blockchains when both are available.

What we do not know is how many agents hold their own wallets today and how much money they control.

How we reported this: Siu's remarks come from his Oct. 7, 2026, TOKEN2049 Singapore keynote, checked against the video's auto-generated captions. Payment counts come from Chainalysis as reported by Blockhead; stablecoin supply from DefiLlama as of Oct. 16; Waller's remarks from his Sept. 29 speech on the Federal Reserve website.

Editorial standards · Report a correction

01Recommended.

esc

Enter for all results