Animoca’s Yat Siu says 50B–100B AI agents will need crypto wallets to pay
OKX, Citi and Coinbase executives agree on crypto rails, but not on the settlement asset.
Animoca Brands chairman Yat Siu put a number on the agentic-commerce narrative: 50 billion to 100 billion autonomous AI agents transacting online “over the next few years.” The executives debating what those agents will use to pay each other converged on crypto wallets and rails, but split on whether the end state is stablecoins, thousands of tokens, or a geopolitically neutral “super currency.”
50B–100B Agents: Why Crypto Wallets Are Being Pitched as the Default
Siu’s core claim is mechanical, not philosophical. If software agents are going to book, schedule, and pay without a human approving each action, they need an account they can control programmatically, and today’s banking stack is built around human identity verification.
In a video interview, Siu said AI agents cannot open bank accounts “yet, at least,” because traditional payment systems require KYC and identity checks designed for humans. His proposed workaround is crypto-native custody: “A crypto wallet would seem the most obvious way,” he said, calling crypto “the perfect machine banking system.” In this framing, the wallet is the primitive that lets an agent hold value, sign transactions, and settle with other agents without waiting for a bank to decide what an “AI customer” is.
Animoca is positioning itself as an agent platform builder rather than a metaverse world-builder. The company pivoted from virtual worlds to an AI-driven metaverse in May 2026 and is building Minds, a persistent AI agent platform. Alongside that, Animoca announced an investment program of up to $10 million for developers building applications on Minds, a signal that it expects third parties to ship the agent behaviors and commerce flows that make the wallet question real.
The market implication is straightforward: if the constraint is identity and account ownership, crypto rails are being pitched as the default path for non-human actors to move money. The harder part is what “money” means when the payer and payee are both machines.
Stablecoins vs Thousands of Tokens vs a Neutral ‘Super Currency’
The debate splits into three settlement-asset theses.
Erald Ghoos, CEO of OKX Europe, argued for convergence. “There will be an AI currency coming,” he said, adding: “This is not going to be fiat, for sure.” He framed the candidate set as “a stablecoin or some other crypto token,” and predicted a single AI-focused “super currency” could become “by far, by far the largest currency that this world has ever seen.”
Siu’s view is closer to routing than convergence. He said agents could transact across “thousands of tokens” while shielding the human owner from the complexity. “The agent knows what to do,” he said. “The human never has to focus his attention on a thousand tokens.” In that world, the settlement layer is a basket of assets and liquidity venues, and the agent is the abstraction layer that makes it usable.
Ghoos’ pushback is geopolitical. He argued a single country-linked stablecoin may not work as a global standard, even if it works locally. A U.S.-dollar stablecoin might fit U.S. businesses and U.S.-based agents, but he questioned whether China’s agents would use a USD-pegged token, and whether the U.S. would tolerate a yuan-pegged stablecoin as the world’s machine settlement asset. His conclusion was that the winning instrument may need to be neutral and not obviously tied to one jurisdiction: “It might be something more abstract.”
None of the executives provided a concrete design, issuer, chain choice, collateral model, or launch timeline for an “AI currency” or “super currency.” That leaves the market structure unresolved, with narratives likely to rotate between stablecoin dominance, multi-token agent routing, and a new neutral unit that does not exist yet.
Trust, Identity and Rail Ownership: The Bottlenecks Before Scale
Even if wallets and a settlement asset are available, Citi’s Debo Sen framed the gating items as controls, not throughput. Sen, Citi’s head of payments, said autonomous agent payments need stronger safeguards around “trust, identity and authorization” before they can operate at scale. She also sketched an adoption path: early use cases in small merchant transactions, then a move into higher-stakes business-to-business payments.
The prize being dangled is large. Juniper Research forecast agentic commerce will generate $1.5 trillion globally by 2030, but the forecast does not specify which currency or network captures those flows. That uncertainty matters because the “winner” may be less about the unit of account and more about who gets embedded into the payment path.
Coinbase’s Keith Grose put the strategic question on the rails themselves. “The question is: who’s going to own those rails, or will they be shared rails or permissionless rails?” he said, adding: “It’s too early to tell.” In practice, the forward signals are the unglamorous ones: whether anyone publishes a spec for an AI-focused currency, whether agent payments graduate from small-merchant experiments into B2B settlement, and whether major platforms commit to shared, permissionless rails or build controlled networks that capture volume.
My Take: Traders Should Treat ‘Machine Money’ as a Rails-and-Standards Race, Not a Single-Ticker Bet
The part that matters right now is the constraint Siu and others are pointing at: identity verification is built for humans, and that makes bank accounts a bad fit for autonomous software. If that remains true, crypto wallets become less a narrative and more a default interface for agents to hold and move value.
The real test is whether anyone turns the “super currency” idea into a spec with an issuer, collateral model, governance, and a clear rail choice, or whether the market drifts toward agent-driven routing across many tokens because it is easier to ship. If those standards harden and start pulling flows from small-merchant payments into B2B settlement, that’s when “machine money” stops being a theme and starts being infrastructure that captures volume.
