
Akshita Jhalani
Crypto Analyst
TL;DR
Mizuho analysts say AI agents could put up to 20% of Circle's revenue at risk. The figure is a ceiling in one scenario, not a forecast. Circle earns mostly on interest from USDC reserves and shares a slice with partners such as Coinbase, so agents optimizing for cost could squeeze that margin.
Key takeaways
- Mizuho says AI agents could put up to 20% of Circle's revenue at risk, a ceiling in one scenario rather than a forecast.
- Most of Circle's revenue is interest on the reserve assets backing USDC, mainly short-term government debt and cash equivalents.
- Circle shares a meaningful slice of that income with distribution partners, Coinbase being the best known.
- Possible pressure points include substitution between stablecoins, yield pass-through to agent wallets and platforms, and disintermediation by rival tokens.
- Circle has not publicly rebutted the figure, and agentic payments today are mostly demos and pilots.
I want to talk about a number that deserves more attention than it's getting. Mizuho's analysts say AI agents could put as much as a fifth of Circle's revenue at risk. Twenty percent is a lot to pin on a company whose business looks, on paper, almost dull.
The bank's note is the latest sign that Wall Street is trying to price a question nobody could answer two years ago: what happens to payments companies when software, not people, picks the rails?
First, the limits. I can't walk through Mizuho's full reasoning line by line, and the "up to" in the headline matters. It's a ceiling in one scenario, not a forecast. What I can do is lay out why the concern is plausible and where it might fall apart.
Mizuho Says Agents Could Shrink Circle's Take
The core claim is simple. Autonomous AI agents, the programs that book, buy and pay on a user's behalf, could change how stablecoin demand is generated and who captures the economics. If that happens, Circle's take could shrink by up to 20%.
I found it notable that a bank would say this about this company. Circle has leaned hard on the story that stablecoins are the natural money for machines. It and its peers have spent months arguing that agents need instant, programmable dollars, and that USDC is the obvious candidate. Mizuho is asking whether the same trend could cut the other way.
Fair question. A thesis that helps you can also expose you.
Circle Earns Its Money on Reserves
To see the exposure, start with how Circle earns. The bulk of its revenue is interest on the reserve assets backing USDC, mostly short‑term government debt and cash equivalents. Holders of USDC don't get that yield. Circle keeps it, then shares a meaningful slice with distribution partners, Coinbase being the best known.
Two consequences follow.
First, revenue tracks the size of USDC in circulation multiplied by prevailing short‑term interest rates. It is a balance‑sheet business wearing a payments costume. Rate cuts hurt it, and so does shrinking supply.
Second, whoever controls distribution controls the margin. Partners that bring in the users get paid for it, and they can bargain for more.
Now add agents. A software agent choosing a stablecoin for a transaction isn't loyal to a brand. It optimizes: cheapest route, best liquidity, lowest friction. That kind of customer tends to push margins down.
Agents Could Squeeze the Model Three Ways
There are a few ways this could play out. None is confirmed, and I'm not attributing any specific one to Mizuho.
One is substitution. If agents treat dollar stablecoins as interchangeable, the incumbent's brand advantage fades. Competitors can offer better terms to whoever builds the agent tooling, and Circle ends up paying more to stay in the flow.
Another is yield pass‑through. A machine‑driven economy might demand that idle balances earn something. If agent wallets or the platforms around them expect a share of reserve income, the spread Circle keeps narrows.
A third is disintermediation. Payment networks, banks and large tech platforms could issue or embed their own tokens, or route through tokenized deposits, leaving USDC as one option among many. Regulators in the US have been building a framework for payment stablecoins, and that framework invites more entrants, not fewer.
The yield angle is the one I find most convincing. Machines are ruthless about idle money. Humans leave cash in a wallet and forget about it; an agent configured to minimize cost won't.
Still, this is the bear case. It assumes agents become a major source of stablecoin volume, and that the volume is large enough to matter against Circle's reserve base. Neither is settled. Agentic payments today are mostly demos, pilots and press releases.
Circle Has Real Answers to Make
USDC is among the most widely integrated dollar stablecoins, and the company has courted regulators and large financial institutions in a way many rivals haven't. Compliance and redemption trust matter to businesses that don't want surprises. An agent acting for a corporate treasury is unlikely to pick the cheapest unknown token over a familiar, regulated one. Dull, it turns out, can be an asset.
Agents might also expand the pie. If software starts moving money at volumes humans never would, total stablecoin demand could grow faster than any margin compression. A smaller slice of a much bigger balance can still be a bigger number.
That's the honest ambiguity in Mizuho's call. A 20% hit in one version of the future is compatible with a revenue surge in another. The analysts are marking a risk, not writing an obituary.
As far as I can tell, Circle hasn't publicly rebutted the figure. Companies rarely respond to a single analyst scenario, and I wouldn't read silence as agreement.
Three Things Will Tell Us More Than the Note
The first is Circle's next earnings report. Watch the distribution‑cost line and how management describes agent‑related partnerships. If partners are taking a bigger share of reserve income, that's the squeeze showing up in real numbers.
The second is USDC supply relative to rivals. Market share shifts in stablecoins tend to be gradual until they aren't.
The third is whether any major agent platform picks a default stablecoin, and what it gets in return. That choice, more than the technology, will decide who earns the spread.
Mizuho has handed investors a number to argue about. My guess is the argument gets louder before it gets settled, and that the first real answer shows up in distribution costs rather than in anyone's price target.
