The Stablecoin Narrative Peaked Three Quarters Ago

I spend my working hours inside a database of 253,129 earnings call transcripts, and one of the things it is genuinely good for is dating a hype cycle. Not feeling the hype — dating it. You run one full-text query across every call in the archive, group the hits by quarter, and the entire narrative arc of a theme falls out as a single column of numbers. No sampling, no watchlist bias, no vibes.

This week the query was "stablecoin". It is early August 2026, the big payment networks reported last week, and the numbers surprised me twice. First surprise: the stablecoin narrative peaked in Q3 2025 — three quarters ago — and it has not come back. Second surprise: that is not the bearish signal it looks like. The count fell while the substance rose, and the gap between those two things is the whole story.

From rounding error to 115 companies

For all of 2024, "stablecoin" was a rounding error on earnings calls. 4 companies mentioned it in Q1, 5 in Q2, 3 in Q3, 9 in Q4. That is 21 company-quarters across a full year, in an archive that captures roughly 10,000 calls per quarter. Statistically, the theme did not exist in the earnings-call conversation.

Then the GENIUS Act happened — the US stablecoin framework, signed into law in July 2025. Here is the full quarterly series, counting companies that mention "stablecoin" on at least one earnings call:

One methodology note, once: each number is the count of companies whose earnings calls match a full-text search for "stablecoin" in that quarter. Keyword matching makes these upper bounds — a CFO describing settlement infrastructure and an analyst asking a throwaway question count exactly the same, and the occasional incidental co-occurrence slips in. The absolute level is fuzzy. The shape is not.

And the shape is textbook. Mentions were already climbing through the first half of 2025 as the legislation moved — 25 companies in Q1, 39 in Q2 — and then the count nearly tripled in a single quarter once the framework became law: 115 companies in Q3 2025. From Q4 2024 to the peak, that is roughly a 13x jump in three quarters.

Regulatory catalysts do this to earnings calls, and it is worth being precise about why. Executives do not add a theme to prepared remarks because the technology matured that quarter. They add it because it became safe to say and expensive to omit. A signing ceremony flips both switches on the same day — legal clarity removes the compliance hedge, and every analyst on every financials call suddenly has a mandatory question.

Three quarters off the peak

Then it cooled. 95 companies in Q4 2025. A bounce back to 104 in Q1 2026. Then 89 in Q2 2026 — down 23% from the peak and the lowest reading since Q2 2025, before the spike fully hit. It is not a straight line down; Q1 2026 interrupted the slide. But all three quarters since the peak sit well below it, and the most recent one is the lowest of the three.

The reflexive read is that the narrative is dying. That is the read I want to argue with. Mention counts measure attention, not substance — and when a theme cools, the interesting question is never "how many fewer companies" but "which companies stopped talking".

The tourists left. The rails stayed.

Cut Q2 2026 by sector and the answer is unambiguous: Financials, 77 companies — over 85% of the list. Information Technology contributes 13, and after that it is a tail you can count on one hand per sector: Communication Services 4, Health Care 4, Industrials 3, Consumer Staples 2.

A 115-company quarter immediately after a headline law is full of what I think of as narrative tourists: companies with no stablecoin product and no stablecoin roadmap, mentioning the theme because an analyst asked or because it made the macro section of prepared remarks sound current. Tourists churn out fast. Once the question stops being novel, the sell side stops asking it of companies where the answer is obviously "not applicable", and the mention evaporates from the transcript. What is left in Q2 2026 is a list dominated by companies for which stablecoins are a line of business, not a talking point.

The top mentioners since April 2026 make the point better than any classifier could: Mastercard, Circle, Coinbase, Visa, Fiserv, Strategy. An issuer, an exchange, two card networks, a payments processor, and a bitcoin treasury company. Nobody on that list is touring. The theme consolidated to the companies actually building.

What the survivors are actually saying

The count fell and the content got denser. Three excerpts from the late-July 2026 calls — all within two days of each other — sketch the entire competitive structure of stablecoin payments better than most industry reports I have read.

Start with Mastercard's July 30 call: "We believe stablecoins have great potential but to work, there are a few essential principles for it to scale: reliability, security and interoperability, and that's what Mastercard delivers." On the same call, management said the launch of its stablecoin initiative had "more than 30 industry leaders participating" — naming checkout.com, Cloudflare and Coinbase among the partners. That is a network doing what networks do: positioning itself as the neutral layer everyone else has to plug into, and listing the companies already plugged in.

Visa's July 28 call is even more explicit about scope: "We are active and investing in each layer of the stablecoin stack from blockchain to issuance, wallets, infrastructure and orchestration and applications." Not a pilot, not an experiment, not a partnership press release — a stated claim on every layer of the stack.

And on Coinbase's July 30 call, CEO Brian Armstrong was asked why the company launched an additional stablecoin alongside its Circle partnership: "we're a multi stablecoin platform. We want to provide the stablecoins that all of our customers want to use. And where possible, we want to strike good economic arrangements with that."

Read the three together and you are watching an industry negotiate its structure in public. The card networks are claiming the interoperability layer — the exact position they already own in card payments. The exchange is refusing issuer lock-in and openly optimizing stablecoin economics across suppliers. These are the same firms whose executives spent recent quarters fielding agentic-commerce questions — the payment giants have decided the next platform shifts run through their rails, and they are saying so on the record, in transcripts anyone can search.

That is what "fewer mentions, more substance" looks like at the single-call level. In Q3 2025 you had 115 companies saying the word; in mid-2026 you have a smaller group describing infrastructure, partners, and economic arrangements. If you only tracked the count, you would have the story exactly backwards.

Diffusion vs. consolidation — two narrative shapes

I ran this same exercise on "agentic AI" across 2,400 earnings calls, and the contrast is instructive, because the two themes traced completely different shapes through the same corpus.

Agentic AI diffused. It went from 45 companies in H2 2024 to nearly 500 per quarter by early 2026, spreading from software into banks, retailers and industrials — all eleven GICS sectors — before flattening at high altitude. There, the raw mention count stayed a decent proxy for the theme's importance, because breadth was the story.

Stablecoins consolidated. A regulatory catalyst, a vertical spike, roughly two quarters of tourists, then a retreat into the one sector that actually builds payment infrastructure. Here the raw count is a misleading proxy — it fell 23% during the same stretch in which Mastercard was announcing 30-plus participants and Visa was investing in every layer of the stack.

Same tool, same corpus, opposite interpretations. Which is precisely why counting mentions is the beginning of the work, not the end. A falling count with a stable core of builders is a maturing theme. A falling count where the builders go quiet too — that would be the bearish version. The sector split is how you tell the two apart, and it takes one extra grouping step.

Track it yourself

Everything above comes out of one endpoint. The query behind the table:

curl 'https://earningscalls.dev/api/v1/search?q=stablecoin&type=transcripts&date_from=2025-01-01' \
  -H "X-API-Key: $KEY"

The q parameter supports exact phrases in quotes, AND/OR operators, and -term negation, so you can tighten the net — "stablecoin" AND settlement — or cut noise out of it. The full parameter reference is in the docs, and the same search is exposed as an MCP tool if you would rather ask Claude than write the loop yourself.

Two follow-ups are worth automating:

Track the sector split over time. Join the matched calls to sector data, one row per quarter. Consolidation shows up as financials' share climbing while the total falls — it is over 85% now. If you want the scaffolding, I wrote up a theme tracker in Python that does exactly this loop: query, group, plot, repeat next quarter.

Use the non-financial remainder as a contrarian tourist detector. Scope "stablecoin" matches to everything outside Financials and watch that slice specifically. Today it is a handful of companies per sector. If it starts re-inflating — consumer names and industrials suddenly working stablecoins back into prepared remarks — the tourist phase is restarting, and you will see it in the counts a quarter before you read it in a headline.

The lifecycle read

"Peaked three quarters ago" is a statement about attention, not adoption. Attention peaked in Q3 2025, three quarters before I am writing this, and it peaked for the most mechanical reason imaginable: a law made the word mandatory for one earnings season. What remains in 2026 is smaller, more concentrated, and far more informative — payment networks describing rails, an exchange describing multi-issuer economics, an issuer and processors filling in the rest.

The general rule I keep coming back to: mention spikes tell you when a theme becomes safe to say. The decline afterwards tells you who actually meant it.

The archive behind these numbers — 253,129 transcripts, 12,799 companies, full-text search across all of them — is queryable at earningscalls.dev.