A $50 billion rush: AI Agent wallets—a game of “hard to make money right in front of you”

Source: Tiger Research

Compiled and edited by: BitpushNews

Continuous headline reports have been circulating about AI Agents (artificial intelligence agents) autonomously trading and processing payments. However, the crypto wallet industry has already been quietly paving the way for this. At present, more than ten companies are building custom wallets specifically for AI Agents. What exactly are they going after? And how big could the potential final payoff be?

Key takeaways

• When AI Agents replace humans to browse the internet and buy goods or information, they will ultimately generate thousands of micro-payments, with each payment worth only a few cents. Existing card payment networks (Card Rails) cannot support payment scales that small, making it crucial to have wallets that can automatically split and send funds based on predefined conditions.
• Although there is currently a lack of near-term profitability, companies such as Coinbase and Binance are still actively building AI wallet infrastructure. The reason is that doing so allows them to lock in future customer segments before AI Agents begin large-scale trading. At this stage, the core focus is building a user base before actual demand explodes.
• Calculations based on Coinbase data show that rising AI Agent usage could bring up to a 7x increase over its current revenue.
• Payment records accumulated inside wallets can demonstrate whether AI Agents are profitable, opening the door to loans backed by future earnings—similar to providing credit loans to small businesses based on their card swipe sales history.
• This is still at the stage of “possibility,” not “proof.” AI Agents still make operational mistakes and execute incorrect payments, and the relevant rules vary by country and company, while the legal status of AI Agents has not yet been clarified. Therefore, the competition is not to capture today’s revenue, but to secure a favorable position years in advance in the large market expected to take shape soon.

1. AI Agents are spreading fast

Earlier this year, a widely discussed experiment on Polymarket provided an AI Agent with $50 in startup capital and let it trade autonomously. The experiment conditions were: if it could not pay for its own API and server costs through autonomous self-profit, it would cease to exist. The Agent then successfully completed the trades, and since then, a series of other Agents have started trading in a similar way.

Although AI Agents have not fully integrated into everyday life yet, there is no doubt that they will be widely adopted in the near future.

2. Every Agent transaction starts with a wallet

AI Agents have not entered the mainstream payments space yet. Their most active application right now is running crypto trading bots inside the crypto ecosystem. These bots operate independently of traditional payment networks and focus exclusively on crypto trading.

However, in the future, payments will expand into areas that people today can hardly imagine. As introduced in earlier reports, AI is changing the nature of payments themselves. Once Agents (not humans) directly interact and navigate on networks, the amount of each payment will drop sharply. The cost of a single API call or one data query could be as low as $0.001, and in extremely extreme cases, even just $0.00001.

To move beyond today’s wallet use cases—automatically splitting and sending such tiny payments based on predefined conditions, with no human intervention throughout the process—requires a programmable payment system (Programmable Payment System). This is exactly the context in which the x402 payment network emerged, and wallets are the foundation for running that network.

However, existing payment networks are designed with “people” as the trading主体.

Credit cards are issued to individual cardholders and run on a “chargeback/dispute” structure—meaning when something goes wrong with a transaction, humans initiate the dispute and reverse the transaction, and each transaction has a fixed fee of up to several tens of cents. When a person makes an occasional $20 purchase, this is perfectly fine. But once Agents start sending payments at thousands of transactions per second—even if each API call costs only $0.001 or each data record costs only $0.00001—this payment model simply becomes economically untenable.

The core question is: can the funds themselves be “programmable”?

Card networks can enable the automatic input of payment details, but they cannot be programmed to split cash flows, stream payments in real time, or settle instantly based on specific conditions. On the network that wallets run, this capability is built in by default. Storing payment details on a card, at best, only lets humans perform transactions at a human scale on someone else’s behalf. Once the economic form shifts to direct machine-to-machine transactions, the wallet becomes the only viable starting point.

3. Agents are a $50 billion business

As shown in the image above, the scope of wallet providers is very broad, covering everything from exchanges to stablecoin issuers and other entities. So why are so many diverse participants moving into the Agent wallet infrastructure space, which currently has no clear short-term profitability?

The answer is that these companies are building for future revenue and future business lines—not for today.

Embedding Agent functionality into wallets is not an initiative that generates income immediately. What it builds is a fundamental capability to absorb that transaction volume when Agents begin generating large-scale activity.

The key is that AI Agents will ultimately run all day long in a browserless environment without human intervention. Imagine a user asks an Agent to produce a research report. As the Agent gathers information, every time it extracts data from different paid platforms, it will execute a micro-payment. A single simple user instruction could trigger 20 to 30 or even more payments in an instant.

What seems like a simple single action to a human, once processed through an AI Agent, turns into an enormous volume of payment transactions.

How this shift in the payment environment will affect company profitability can be inferred from publicly available Coinbase data. The calculation uses Coinbase’s 9.2 million monthly transaction users (MTU) rather than its roughly 120 million total registered user base.

Combining three variables—adoption rate, number of Agents per user, and daily call frequency—yields the following scenario projections:

• Conservative scenario (10% adoption rate, 1 Agent per user, 50 calls per day): adds about $84 million in annual revenue, up 1.2%.
• Neutral scenario (50% adoption rate, 2 Agents per user, 200 calls per day): added revenue jumps to about $3.36 billion, up 46.8%.
• Aggressive scenario (100% adoption rate, 3 Agents per user, 1000 calls per day): annual revenue reaches about $50.37 billion, about 7x Coinbase’s current total revenue.

Most striking in this comparison is that the gap between these three scenarios expands geometrically rather than arithmetically. Adoption rate itself increases by only 10x (from 10% to 100%), but the resulting revenue gap expands by about 600x (from $84 million to $50.37 billion).

Because “adoption rate,” “number of Agents per user,” and “daily call volume” are all multiplicative variables, even a small increase in any one factor leads to an exponential rise in the total. Therefore, once Agents achieve large-scale adoption and user numbers surge, the resulting revenue stream could reach up to about 7x of current total revenue.

That is why Coinbase is pushing hard for Agent wallet infrastructure even today, despite having no explicit revenue at present. It’s doing so to lock in its market share when the AI Agent-driven era arrives.

4. Moving toward a new kind of Agent neobanking (Neobanking)

The transaction data accumulated through wallet infrastructure is not just simple records. It lays the groundwork for an entirely new business model—because payment history stored in wallets can serve as a credit assessment standard, proving the financial condition and operating performance of AI Agents.

Once this data-based credit assessment system is established, wallet providers can naturally expand into the next generation of financial services—for example, revenue-based financing (Revenue-Based Financing, RBF) specifically targeted at Agents.

Stripe Capital is a prime example of building new financial business on top of existing payment data. When Stripe launched its lending service Stripe Capital in September 2019, it did not rely on external credit bureaus or cumbersome loan documents. It used only real-time merchant sales data flowing through its own payment network to assess loan eligibility and loan amounts.

Stripe’s case shows that a company can build high-value financial business on top of its existing data pipelines without needing to set up a separate sales network or do additional marketing expansion.

Agent wallet providers are likely to take the same expansion path. By continuously accumulating Agent income data through wallets, they can create the basis to advance operating funds through RBF, and transform into an Agent-focused financial platform to profit from it.

However, building this new business line depends on a prerequisite: AI Agents must evolve beyond simple payment execution tools, becoming revenue-generating asset holders on their own and earning enough real income to repay loans.

5. This growth is still not proven

The projected 7x revenue growth for Coinbase and the expansion into RBF discussed earlier are all based on optimistic scenarios built on the assumption that Agent payments will become widespread. To establish this system in the real economy still faces major obstacles.

First, there are still significant doubts about AI Agents’ actual purchase conversion rates and payment reliability. During autonomous ordering, Agents can still make incorrect moves that lead to erroneous payments caused by “hallucination.” Sometimes transactions are also directly blocked because the card issuer’s fraud detection system (FDS) rejects them. Therefore, the current real payment completion rate remains low.

In addition, payment protocols such as x402, AP2, and MPP are still fragmented and have not unified into a single standard. At the same time, the lack of clear KYC (identity verification) and financial regulatory policies for AI Agents that are not legal entities is another major barrier to further market expansion.

Therefore, for wallet providers right now, the goal is not short-term fee revenue. Apple’s App Store took 15 years to build a fee revenue market of $10 billion per year, and WeChat Pay took 7 years to build a large mini-program ecosystem. Agent wallets are following a similar long-term timeline, focusing on building an ecosystem rather than fighting for near-term marginal returns.

This competition is not about capturing today’s small incremental revenue. It’s about who can first gain control over the data that determines where funds flow in the fully formed agent economy expected five to ten years from now.