Tiger Research says wallet infrastructure for AI agents is turning into a contested area for exchanges and stablecoin companies. Citing Coinbase data, the report estimates that if AI agents reach broad adoption, related wallet revenue could climb to as much as seven times today’s level, roughly $50.37 billion. In a conservative case, the increase would be just 1.2%.
The report says the gap is driven by geometric growth rather than a straight-line rise. Three variables sit at the center of the model: adoption rate, the number of agents used by each customer, and the number of daily calls each agent makes. BlockTempo translated and summarized the report.
Three multiplying variables create a roughly 600-fold revenue gap
For now, AI agents are not yet part of everyday payment flows. The report says their most active role remains inside crypto, mainly as trading bots operating outside traditional payment rails and focused on digital-asset markets.
That could change if AI begins acting directly on the internet at scale. In that setting, the size of each payment may fall sharply. A single API call or one data query could cost as little as $0.001, and in extreme cases just $0.00001.
The report argues that payments this small, split automatically by preset rules and executed without human intervention, require programmable payment systems. It places the x402 payment rail in that context, with wallets serving as the basic execution layer.
Existing payment rails, by contrast, were built around humans. Cards are issued to a named holder, rely on chargebacks when something goes wrong, and collect fixed fees worth tens of cents per transaction. That works when a person occasionally spends $20. It becomes uneconomic if agents start sending thousands of payments every second while each API call is worth $0.001 and each data record is worth $0.00001.
From the Polymarket experiment to machine-native payments
The report points to an experiment on Polymarket earlier this year. An AI agent was given $50 in startup capital and allowed to trade on its own, under one condition: if it failed to earn enough to cover API and server costs, it would “disappear.” According to the report, the agent traded successfully, and similar agents later began operating in the same way.
The underlying question, it says, is whether money itself can be programmed. A bank card can automate payment entry, but it cannot split payments by condition, stream payments, or settle flows in real time. Wallet rails are built with those functions in mind. If the economy shifts toward direct machine-to-machine transactions, the report says wallets become the only realistic starting point.
It also notes that wallet providers span a wide field, from exchanges to stablecoin issuers. Many of them are entering AI-agent wallet infrastructure even though short-term profits are hard to see. The reason, the report says, is not current revenue but future capacity. Firms are building the rails now so they can absorb large transaction volume later if agent activity scales.
Coinbase’s three scenarios: 1.2%, 46.8%, and a 7x revenue outcome
The report says AI agents may eventually run around the clock in browserless environments without human intervention. If a user asks an agent to produce a research report, for example, the agent may need to pull paid data from several platforms. Each retrieval could trigger a micropayment, and one user request could instantly turn into 20 to 30 payments or more.
What looks like one simple task from the user side becomes a batch of transactions once handled by an AI agent. To estimate the revenue effect, the report uses Coinbase’s disclosed figures and takes the company’s 9.2 million monthly transacting users, or MTUs, as the base instead of its roughly 120 million registered users.
It then applies three assumptions — adoption rate, agents per user, and daily call frequency — across three scenarios:
- Conservative case: 10% adoption, one agent per user, and 50 calls a day. Estimated annual incremental revenue: about $84 million, or a 1.2% increase.
- Base case: 50% adoption, two agents per user, and 200 calls a day. Estimated additional revenue: about $3.36 billion, or a 46.8% increase.
- Aggressive case: 100% adoption, three agents per user, and 1,000 calls a day. Estimated annual revenue: about $50.37 billion, around seven times Coinbase’s current total revenue.
The report highlights the scale of the spread. Adoption rises by a factor of 10 between the conservative and aggressive cases, from 10% to 100%, but estimated revenue expands by about 600 times, from $84 million to $50.37 billion.
Payment data could support RBF products for AI agents
Because those three inputs multiply together, even small gains in any one of them can drive a sharp increase in total volume. That is one reason, the report says, Coinbase is pushing AI-agent wallet infrastructure despite having little direct revenue from it today.
The value of the wallet layer would not stop at transaction records. Payment history stored in wallets could become a credit signal for judging an AI agent’s financial condition and operating performance. On that basis, wallet providers could move into the next stage of financial services, including revenue-based financing, or RBF, tailored to AI agents.
As an example, the report points to Stripe Capital. When Stripe launched the lending service in September 2019, it did not rely on outside credit bureaus or extensive paperwork. Instead, it used real-time sales data from merchants on its own payment network to assess eligibility and loan size.
That case suggests a company can build a high-value financial business on top of existing operational data channels without creating a separate sales network or adding fresh marketing spend. The report says AI-agent wallet providers could follow a similar path by collecting agent income data over time and then using that data to offer working capital through RBF.
Hallucinations, fraud blocks and regulation remain open problems
The report also makes clear that the 7x Coinbase revenue scenario and any expansion into RBF depend on wide adoption of AI-agent payments. That remains a best-case framework rather than a near-term outcome.
Several obstacles stand in the way. The report says AI agents still face major questions around conversion and payment reliability. They can hallucinate during autonomous purchasing and trigger incorrect payments. Transactions can also be blocked by card issuers’ fraud detection systems, or FDS, leaving payment completion rates low.
Protocol fragmentation is another issue. The report names x402, AP2 and MPP as examples of payment protocols that have yet to converge on one standard. It also notes that AI agents are not legal persons, and there is still no clear KYC or financial regulatory framework for them.
For that reason, wallet providers are not chasing short-term fee income, the report says. It compares the current stage to long buildouts in other ecosystems: Apple’s App Store took 15 years to build a $10 billion annual fee market, and WeChat Pay took seven years to establish its mini-program ecosystem. AI-agent wallets, in the report’s view, are on a long timeline as well. The near-term contest is not about marginal revenue today, but about which company secures the flow data of an AI-agent economy that could take shape over the next five to 10 years.