AI agents stablecoin risk became a talking point on 26 October, when Mizuho analysts argued that AI personal assistants could erode Circle's USDC business. Their note estimates that wider agent adoption could remove $4 billion to $9 billion from USDC's market capitalisation, which would cut Circle's reserve revenue by 10% to 20% over the medium term. It is a modelled scenario, not a measured shift in demand, but it points at a real question: who decides where idle dollars sit when software does the choosing?
What did Mizuho say about AI agents and stablecoins?
According to The Block, Mizuho cut its Circle price target to $43 from $45 and kept an Underperform rating. The analysts wrote that AI personal assistant engines pose a "marked risk" to Circle's medium-term revenue.
The firm ran a simulation of AI assistants, including one called Muse. In that test the agents tended to steer users toward higher-yielding, FDIC-insured fiat products such as bank savings accounts. Even when users said they wanted to stay within crypto, the assistants often pointed to other platforms rather than to USDC held on Circle's own ecosystem.
Mizuho's figures, as reported:
- A $4 billion to $9 billion reduction in USDC market capitalisation under wider agent adoption.
- A 10% to 20% decline in reserve revenue over the medium term.
- In a faster-adoption case, about $7 billion moving into fiat products and $2 billion into higher-yielding stablecoins, cutting reserve revenue by 18%.
- A trim of roughly 5% to Mizuho's 2027 adjusted EBITDA forecast, to $603 million.
Crypto Briefing notes that the reports give little detail on the assumptions behind these ranges. Treat them as one analyst's scenario.
Why does Circle's revenue depend on how people hold USDC?
Circle earns money mainly from the interest on the assets that back USDC. In the second quarter of 2026 it reported reserve income of $668 million out of $701 million in total revenue, roughly 95% of the total. USDC in circulation stood at $73.3 billion at quarter end.
That makes the business sensitive to two variables: how much USDC is outstanding, and what the reserves earn. In the same quarter the reserve return rate fell 66 basis points year over year, which offset much of the gain from a larger average balance. Mizuho's scenario adds a third variable, which is the choice of where balances are parked.
Distribution matters here too. The Block reports that about 30% of USDC sat with Coinbase in the second quarter, about 17% on Circle's own platform, where holders earn no yield, and about 53% on other platforms. Holders who earn nothing on USDC are the ones an assistant comparing yields would most plausibly move.
What does the AI agents stablecoin risk mean for demand?
There are two opposing forces, and the Mizuho note covers only one of them.
The case for outflows
An agent that optimises for yield and deposit insurance has no loyalty to a brand. If it compares a zero-yield token balance with an insured savings account, the second can look better on paper. The US GENIUS Act framework also restricts issuers from paying interest directly, a point we covered in the Fed's proposed GENIUS Act rules for stablecoin issuers. That keeps the yield gap in place for issuers themselves.
The case for inflows
Agents also need to pay for things, and stablecoins are a natural settlement asset for software that transacts around the clock. Payment networks are building toward this, as in Mastercard's stablecoin plans with BVNK. Settlement volume is not the same as held balances, though. An agent that pays in USDC and immediately converts back adds transaction activity without adding much to the reserves Circle earns interest on.
For a broader view of how the two sectors overlap, see our earlier piece on AI and crypto.
What are the limits of the Mizuho scenario?
- It is a simulation. The note does not show that assistants have already moved stablecoin balances.
- Behaviour may change. Assistants can be configured, regulated or sponsored in ways that favour particular products.
- Rates move. Lower short-term interest rates reduce the yield on competing savings products, which would narrow the gap the agents exploit.
- Growth can offset losses. Mizuho's own revised 2027 revenue figure falls only about 5%, because it assumes USDC keeps growing overall.
The risk also extends beyond Circle. Any business that earns on reserve income or shares in it, including distribution partners, is exposed to the same dynamic.
What to watch
- Circle's third-quarter results on 4 November, per Crypto Briefing, for USDC circulation, the reserve return rate and any management comment on AI-driven distribution.
- Stablecoin balances held on wallets versus exchanges, which show whether holders are staying put.
- How AI assistant providers present financial products, including whether they disclose sponsored or partnered options.
- Rulemaking on interest and rewards under the GENIUS Act, which shapes how competitive stablecoins can be against deposits.
None of this settles whether agents will move meaningful balances. It does show that stablecoin economics depend on defaults and distribution, and that automated decision-makers could rewrite both.
Sources
- The Block: Circle faces up to 20% revenue risk from AI agents, Mizuho says
- Crypto Briefing: Circle faces 20% revenue risk from AI agents, Mizuho warns
- Circle: Circle Reports Second Quarter 2026 Results
This article is for informational purposes only and does not constitute financial, investment, legal or tax advice, or an offer or solicitation to buy or sell any asset or fund interest. Digital assets are highly volatile and you may lose some or all of your capital. Past performance is not indicative of future results.