AI agents make 138 million crypto payments as U.S. dominates spending on AI models

We have spent years talking about the potential for machines to pay each other. It was always the 'killer app' that was just around the corner. But new data suggests we aren't waiting for the future anymore. We are living in it. According to the latest research from Chainalysis, AI agents have already executed 138 million cryptocurrency payments. This isn't a pilot program or a testnet experiment. It is a massive, quiet economy operating under the surface of the traditional financial system.

For founders, the most striking part of this data isn't just the volume of transactions. It is the geographic concentration. U.S.-linked wallets are responsible for more than 88% of the spending on AI models. Meanwhile, China, a country we are told is a fierce rival in the AI arms race, accounts for a measly 0.1%. This gap is staggering, and it tells us a lot about where the actual building is happening versus where the geopolitical posturing is taking place.

The Rise of the Autonomous Payer

An AI agent is essentially software with a wallet. It has a specific goal, and it has the autonomy to spend resources to achieve that goal. If an agent needs more compute power, it buys it. If it needs to access a specific dataset, it pays for the API call. These 138 million payments represent the friction being removed from the development cycle. In the old world, a developer would have to put a credit card on file, hit a monthly limit, and deal with fraud alerts. In the agentic world, the software just pays for what it needs, exactly when it needs it.

This is why crypto and AI are merging so aggressively. You can't give a LLM a bank account. Traditional banks are built on 'Know Your Customer' rules designed for humans with social security numbers. Crypto is built for addresses. A piece of code can hold a private key just as easily as a person can. What we are seeing in the Chainalysis data is the first real proof that agents prefer the efficiency of on-chain rails over the legacy banking system.

The Great Divergence: U.S. vs China

The 88% to 0.1% split between the U.S. and China is the headline that should make every founder stop and think. For all the talk about China's dominance in hardware or surveillance AI, they are currently non-players in the decentralized AI model market. There are a few reasons for this, and they aren't all related to tech talent.

First, there is the regulatory wall. China has cracked down hard on crypto for years. If your economy makes it nearly impossible to hold or move digital assets, your AI agents can't participate in the global marketplace for compute and inference. Second, the leading models—the ones people actually want to pay for—are still largely clustered in U.S. hubs or developed by U.S.-aligned entities. If you are building an agent today, you are likely plugging into OpenAI, Anthropic, or Meta’s Llama series. All of these require a bridge to the Western financial or crypto ecosystem.

What This Means for Builders

If you are a founder, this data is your green light. It confirms that the infrastructure for an agentic economy is already battle-tested. You don't have to wonder if people will use crypto to power AI; they are doing it to the tune of hundreds of millions of transactions. Here is how you should be looking at this:

• Focus on Agentic Workflows: Stop building tools for humans to use. Start building tools for agents to use. Your next 'customer' might be a script, not a person sitting at a desk.
• Global by Default: The U.S. dominance is a strength, but the underlying tech is borderless. Use crypto to bypass the geographic silos that traditional SaaS companies are stuck in.
• Efficiency Over Hype: These 138 million payments weren't made by people buying NFTs. They were utility payments for model access. Build things that provide actual utility, and the agents will find you.

The skepticism I usually hold for 'crypto-AI' buzzwords is softened by these numbers because they represent actual usage. We are moving away from the era of speculation and into the era of automated utility. The fact that the U.S. is leading so heavily suggests that the next decade of AI innovation will be built on the back of American-led crypto infrastructure, regardless of what the regulators in D.C. might say on any given Tuesday.

The Bottleneck is Software, Not Capital

When you see 138 million payments, you might think there is a massive amount of capital flowing. But agentic payments are often tiny—micropayments that would be eaten alive by credit card fees. This is the 'long tail' of the economy. An agent might pay half a cent to verify a piece of data. Multiply that by millions of agents, and you have a new type of economy that doesn't rely on high-margin, high-ticket sales. It relies on volume and speed.

The real winner in the AI race isn't the company with the biggest model; it's the ecosystem that allows those models to trade value with the least amount of friction.

Right now, that ecosystem is clearly centered in the U.S. and centered on-chain. If China or Europe want to compete, they can't just build better algorithms. They have to fix their financial rails. You cannot have a world-class AI industry if your money moves at the speed of a 1970s wire transfer.

The Takeaway

The agentic economy is no longer a whitepaper theory. The data shows that AI agents are active, they are paying, and they are overwhelmingly concentrated in the U.S. market. For founders, the path is clear: build for the machines. The machines have the wallets, they have the speed, and they don't care about your UI. They only care about the API and the cost per token. If you can provide value to an agent, you are tapping into a market that is already processing 138 million transactions while the rest of the world is still trying to figure out what a wallet is.

Read the original at CoinDesk →