AI Agent Wallet Infrastructure Quietly Heats Up: Coinbase Estimates Show Revenue Could Surge Up to 700%

Written by: Ekko an, Ryan Yoon, Tiger Research

Compiled by: AididiaoJP, Foresight News

News headlines are always reporting on AI agents trading autonomously and handling payments themselves. In fact, the crypto wallet industry has already been paving the way quietly. Currently, more than ten companies are building wallets specifically for agents. What do they really want? And how large is the potential return?

Key Points

When AI agents browse the web, purchase goods, or buy information on behalf of humans, they initiate hundreds or thousands of micro-payments worth only a few cents or even less. The existing bank card payment system simply cannot support this volume, hence the need for wallets that can automatically split and send funds based on preset conditions.

Despite seeing almost no revenue in the short term, companies like Coinbase and Binance are still doubling down on AI wallet infrastructure. The reason is simple: secure the future user base before agents engage in large-scale trading. This current phase is about seizing the initiative before the explosion of actual demand.

Based on Coinbase's data calculations, after the usage of AI agents rises, its revenue could reach up to approximately 7 times its current level.

Payment records accumulated in the wallet can intuitively show whether an AI agent is generating revenue, opening the door for loans based on future earnings—just like granting credit to a small business based on its card swipe transaction flow.

However, all of this currently remains at the stage of possibility, not verified fact. AI agents still make mistakes and execute erroneous payments; rules vary across countries and companies; and the legal status of agents is not yet clear. Therefore, the focus of current competition is not on making money today, but on positioning themselves years in advance for a market expected to take shape.

AI Agents Are Becoming Fully Active

Earlier this year, there was a widely watched experiment on the prediction market Polymarket: an AI agent was given $50 in startup capital to trade autonomously, with the condition that it would "disappear" if it did not earn enough to cover API and server costs. As a result, this agent traded successfully. Since then, a batch of similar agents has started trading in the same way.

AI agents have not yet entered daily life, but it is already clear that they will be used on a large scale in the near future.

Every Agent Transaction Starts with a Wallet

Currently, AI agents have not yet entered daily payment scenarios. Their most active application remains trading bots within the crypto ecosystem—independent of traditional payment rails, focusing on cryptocurrency trading.

In the future, payments will extend into areas unimaginable today. As we pointed out in previous reports, AI is changing the nature of payments. Once agents act directly on the web instead of humans, the amount per single payment will drop sharply. The cost of one API call or one data query could be as low as $0.001, or even only $0.00001 in extreme cases.

To surpass current wallet usage methods and achieve such micro-payments, automatically split based on preset conditions, and without human intervention, reliance on a programmable payment system is necessary. This is the background behind the emergence of the x402 payment rail, and the wallet is the foundation upon which this rail operates.

However, existing payment rails are designed around "humans" as the transaction subject.

Bank cards are issued to specific cardholders and adopt a chargeback mechanism—when problems arise, humans dispute and reverse transactions, and each transaction charges a fixed fee of several dozen cents. These are not issues when humans occasionally spend $20 to buy something. But once agents start issuing thousands of payments per second, with each API call at $0.001 and each data record at $0.00001, this payment model becomes economically completely unworkable.

The core question is: Is money itself programmable?

Bank cards can automate payment information input, but cannot be programmed to split funds by condition, stream payments, or settle fund flows instantly. The rails on which wallets operate, however, default to possessing these capabilities. Storing payment information on a card can at most execute a "human-scale" transaction on behalf of a person. Once the economy shifts to direct machine-to-machine transactions, the wallet becomes the only possible starting point.

Agents Could Be a $50 Billion Business

From the relevant table, it can be seen that wallet providers cover a wide range, from exchanges to stablecoin issuers. Why would so many different types of players enter the agent wallet infrastructure, which shows almost no profitability in the short term?

The answer is: they are laying out future revenue and business, not today's. Adding agent features to wallets now is not to make money immediately, but to build capacity in advance to absorb massive transaction volume when agents become active on a large scale.

The key is that AI agents will eventually run 24/7 in browserless environments, completely without human intervention. Imagine a user asking an agent to complete a research report. When the agent collects information, every time it pulls paid data from different platforms, it executes a micro-payment. One instruction from a user could trigger 20 to 30 or even more payments in an instant.

A request that looks like a simple operation to a human becomes a large number of payment transactions once processed by an AI agent.

How will this change in the payment environment affect corporate revenue? It can be estimated using data disclosed by Coinbase. The calculation is based on Coinbase's 9.2 million Monthly Transacting Users (MTU), rather than its total registered user count of approximately 120 million.

Combining three variables—adoption rate, number of agents per user, and daily call frequency—the following scenarios are derived:

• Conservative scenario (10% adoption rate, 1 agent per user, 50 calls per day): Annual revenue increase of approximately $84 million, a 1.2% increase.
• Neutral scenario (50% adoption rate, 2 agents per user, 200 calls per day): Additional revenue rises significantly to approximately $3.36 billion, a 46.8% increase.
• Aggressive scenario (100% adoption rate, 3 agents per user, 1000 calls per day): Annual revenue of approximately $50.37 billion, about 7 times Coinbase's current total revenue.

The most prominent point in this comparison is: the gap between the three scenarios is geometrically amplified, not simply added. The adoption rate only rises from 10% to 100% (10 times), yet the revenue gap expands by about 600 times—from $84 million to $50.37 billion.

Because the three variables of adoption rate, number of agents per user, and daily call volume are multiplicative, any small increase will bring exponential growth to the total. Therefore, once agents achieve large-scale adoption and the user count surges, the resulting revenue stream could reach up to approximately 7 times the current total revenue.

This is also the reason why Coinbase is vigorously promoting agent wallet infrastructure even though it has almost no related revenue today—it wants to lock in that share of income expected to come in the agent era in advance.

Moving Towards a New Bank for Agents

Transaction data accumulated through wallet infrastructure is far more than simple records. It provides a foundation for new business models: the payment history stored in the wallet can become a credit standard for evaluating the financial status and performance of AI agents.

Once this data-based credit evaluation system is established, wallet providers can naturally extend to next-generation financial services, such as Revenue Based Financing (RBF) specifically for agents.

Stripe Capital is a typical case. It successfully built new financial business on top of existing payment data. When Stripe launched the lending service Stripe Capital in September 2019, it did not rely on external credit agencies, nor did it require cumbersome loan materials, but directly used the real-time sales data of each merchant in its own payment network to evaluate loan eligibility and amounts.

Stripe's case shows that a company can build high-value financial business on top of existing operational data pipelines, without additional sales networks or marketing investment.

Agent wallet providers are likely to follow the same expansion path. By continuously accumulating agent revenue data through the wallet, there is a foundation for providing operating capital through RBF and gaining revenue as a financial platform focused on agents.

However, to truly build this new business line, there is a prerequisite: AI agents must evolve from simple payment executors into asset-holding entities that can generate revenue themselves, earning enough real income to repay loans.

This Growth Remains Unverified

The descriptions above regarding Coinbase's revenue growing up to 7 times and expanding towards RBF are all optimistic scenarios assuming widespread adoption of agent payments. To truly implement these in the real economy, the obstacles remain significant.

First, there are still major questions regarding the actual purchase conversion rate and payment reliability of AI agents. Agents still "hallucinate" and make mistakes when placing orders autonomously, leading to erroneous payments; sometimes they are directly intercepted by the card issuer's Fraud Detection System (FDS). The actual payment completion rate is therefore still low.

In addition, payment protocols such as x402, AP2, and MPP are still in a fragmented state and have not converged to a unified standard; AI agents are not legal entities, lacking clear KYC (identity verification) and financial regulation, which further hinders market expansion.

Therefore, the current goal of wallet providers is not short-term fee income. The Apple App Store took 15 years to build a $10 billion annual fee market, and WeChat Pay took 7 years to establish a massive mini-program ecosystem. Agent wallets are similarly on a long timeline—they are building an ecosystem, not competing for immediate returns.

The current competition is not for that marginal income today, but to see which company can first control the fund flow data after the agent economy fully takes shape in the next five to ten years.