Biggest by what measure?
The claim only becomes assessable once you fix a unit, and the answer flips depending on which one you choose.
Measure | Can agents lead? | Why |
|---|---|---|
Transaction count | Plausibly, and possibly soon | An agent calling APIs, buying data and paying for compute generates thousands of payments where a human generates a handful. Agentic payment rails have already processed over 167 million transactions. |
Number of wallets | Plausibly | Roughly 40,000–69,000 agents were operating on-chain by early 2026, with around 24,000 registered under a single identity standard within months of its launch. Wallet creation is nearly free. |
Value transacted | No, not remotely | Roughly $50 million across all agentic activity — approximately 0.0001% of stablecoin volume, against $226 billion in B2B stablecoin payments. |
Average payment size | No — structurally the opposite | Around $0.31 per transaction. The architecture is designed for micropayments. |
Agents could become the most numerous crypto users long before they become the most economically significant. Those are different claims and only one of them is near.
For a payments network, count matters. It determines infrastructure load, fee revenue at scale, and whether a rail is genuinely used or merely available. So the headline claim is defensible — provided everyone is clear it is a claim about frequency, not about money.
The structural reason it could happen
The case rests on an exclusion rather than a preference, which is what makes it unusually durable.
Coinbase CEO Brian Armstrong put it directly in March 2026: soon there will be more AI agents than humans making transactions, because AI agents cannot open bank accounts. Banks require identity verification that software cannot provide. A wallet address is generated from a private key and requires no identity verification at all.
The economics reinforce it. Payment processors reject applicants they cannot underwrite, and a software tool with no legal entity, website or trading history is close to impossible to underwrite. Conventional processing fees also exceed the value of the payments agents want to make — no card network is economic at $0.31.
So this is not a case of crypto being marginally better for agents. It is a case of crypto being the only rail currently available to them, and that asymmetry does not resolve itself as the technology matures.
The bottleneck is on the other side
Here is the part almost every piece on this subject omits, and it is where the outcome will actually be decided.
Agents cannot spend money unless someone accepts it. And the supply side is thin.
THE RATIO THAT MATTERS Across roughly $24 million of agentic payment volume over a 30-day period: approximately 94,000 buyers and 22,000 sellers. That is more than four buyers for every seller. OKX Ventures diagnosed the constraint plainly: very few API sellers are willing to accept per-use stablecoin payment, and the vast majority of agents still access services through API keys and subscriptions rather than paying per call. The road is built. The cars have not been produced — but neither have the destinations. |
This reframes the question entirely. The barrier to agents becoming the biggest crypto users is not agent capability, wallet infrastructure or regulatory permission. All three exist. The barrier is that most of what an agent might want to buy is not for sale on a per-call basis.
What a seller gives up by accepting agent payments
It is worth being honest about why sellers have not converted, because the reasons are commercial rather than technical.
What per-call payment costs a seller | Why it matters |
|---|---|
Predictable recurring revenue | A subscription is a forecastable revenue line. Per-call payment converts it into variable usage income that is harder to plan around and harder to value. |
The customer relationship | No account, no email address, no renewal conversation, no upsell path. An agent that pays and leaves is a transaction, not a customer. |
Brand and differentiation | An agent selecting a service programmatically compares price, latency and reliability. It is not persuaded by positioning, design or marketing. |
Price discrimination by tier | Enterprise pricing depends on knowing who the buyer is. An anonymous agent paying per call collapses tiered pricing towards a single rate. |
Lock-in | Switching costs largely disappear when integration is a standard protocol rather than a bespoke contract. |
That list explains the 4:1 ratio better than any technical account does. Sellers are not failing to adopt because integration is hard. They are declining because the business model is worse for them.
And what they gain
No chargebacks. On-chain settlement is final, which removes a category of loss that costs conventional merchants real money.
No onboarding cost. No KYC, no account provisioning, no support burden for a buyer who exists for four seconds.
Global reach by default. No card network geography, no currency conversion, no acquirer relationship.
Access to demand that cannot otherwise pay. This is the real prize — a buyer population that has no alternative payment method at all.
Granular monetisation. Charging precisely for what is consumed rather than approximating it with tiers.
The last two are what will eventually flip the ratio. A seller who accepts agent payments captures demand that literally cannot reach competitors who do not.
The scale the supply side is being built for
The infrastructure investment makes sense only against a large expected market, and the projections are substantial.
Gartner has projected that machine customers could account for up to 20% of revenue by 2030. The x402 Foundation, launched operationally under the Linux Foundation in July 2026, counts Visa, Mastercard, Ripple, American Express, Stripe, Adyen, Shopify, Google, Amazon Web Services and Cloudflare among its founding members. x402 was also selected as the protocol layer for Amazon Bedrock AgentCore Payments.
Those are not organisations that build payment standards speculatively. Shopify and Adyen in particular are merchant-side businesses — their participation is the clearest signal that the seller conversion is being worked on deliberately rather than left to emerge.
Three things that have to be true
Sellers have to price per call. Until the supply side offers metered access as a default rather than an exception, agents have nothing to buy and wallets sit idle.
Agents have to spend without per-transaction human approval. Today the major platforms require confirmation for trades and transfers, which caps frequency at human attention. That is the correct design for now, and it is also the ceiling.
Liability has to be resolved. If an agent makes an erroneous payment, responsibility across user, developer, model provider and platform remains unsettled everywhere. No enterprise seller scales into a channel whose failure mode has no named owner.
The honest current position
Set against the projections, the measured reality is sobering and should be stated.
Agentic payment volume has been roughly $50 million in total — against $226 billion in B2B stablecoin payments over a comparable framing.
Daily transactions on the leading rail fell from roughly 731,000 in December 2025 to around 57,000 by March 2026, a decline of about 92% from peak.
One analysis put the ratio of genuine to artificially generated transactions close to 1:1.
Ecosystem valuations near $7 billion have diverged sharply from actual usage, with several infrastructure projects seeing usage decline by more than 80%.
So the claim in this article’s title is a projection, not a description. Agents are currently among the smallest crypto users by every measure that matters.
The bottom line
AI agents could become the biggest crypto users by transaction count, and the structural argument for it is stronger than most crypto theses: agents cannot get bank accounts, cannot be underwritten by payment processors, and cannot economically use card rails for sub-dollar payments. Crypto is not the better option for them, it is the only one.
But the constraint is not on the agent side. Wallets exist, standards exist, and the largest exchanges and payment networks have shipped the infrastructure. What does not yet exist in sufficient quantity is sellers — four buyers for every one, because per-call payment is a worse business model for the seller than a subscription.
Which means the thing to watch is not agent counts, wallet creation or protocol announcements. It is whether API providers, data vendors and service businesses start offering metered access as standard. When the seller ratio inverts, the rest follows quickly. Until it does, the agents are ready and there is nothing to buy.
Important
This article is general information and analysis about a developing technology sector. It is NOT investment advice and is not a recommendation to buy any asset or use any platform. Figures are drawn from third-party analytics and research, reflect the periods stated, and several carry acknowledged data-quality caveats including difficulty distinguishing genuine from artificially generated transactions. Projections cited are published forecasts by their authors, not outcomes. Systems that allow software to hold and spend funds carry substantial risk including total loss.
Sources
Research and reporting from OKX Ventures, Artemis, Gartner, the Linux Foundation and x402 Foundation, Coinbase and Brian Armstrong’s public statements, plus analysis from CoinDesk, BeInCrypto Institutional Research, Nevermined and KuCoin Research.
Bitnxt tracks wallets, payment infrastructure and crypto technology providers across the US, UK, EU and UAE. Explore the directory at bitnxt.io.
