August 2026 in Earnings Calls: The Month in Review

I work inside a corpus of 253,000+ earnings call transcripts every day, and the first thing I do on the first of a new month is close out the old one: count the calls, rerun the theme searches, read the outliers. August 2026 was the easiest month of the year to summarize in one sense — it was the biggest reporting month of the quarter by a wide margin, and one company dominated the headlines so completely that you already know its name. It was also the trickiest, because the moves I actually care about happened quietly underneath that noise, in themes small enough that nobody was watching them.

Here is August, in the order the data tells it.

The Numbers First

August delivered 4,809 earnings calls from 4,508 companies. For context: June had 1,613 calls and July had 2,861. August out-produced June and July combined.

Nobody reads 4,809 transcripts. At that volume the only honest way to work is to let full-text search do the first pass, rank what it finds, and then spend your reading hours on the calls where something actually moved. That is the workflow behind everything below: the same fixed set of phrase searches I run every month, against every call in the window, so the numbers are comparable month to month.

That volume also shapes how you have to read the results. Roughly 66% more companies reported in August than in July, which means any theme that grew about 66% month-over-month did nothing special — it just rode the reporting calendar. When I compare themes below, I compare them against that baseline, never against raw July counts. Most month-in-review pieces skip this step, and it makes half their conclusions wrong.

The month started early — Caterpillar was on the tape August 4 — and barely paused until the last week. Here is the full theme board, ranked by companies whose August calls matched each search:

One methodology note, stated once and applying to every number here: these counts are calls matching a full-text search for the phrase, so keyword matching is an upper bound. The search catches the CFO who says a rate cut is not in the plan just as readily as the one banking on it. Treat these figures as a ceiling on attention, not a measure of conviction.

The NVIDIA Moment

NVIDIA reported on August 26, and for about 48 hours it was the only earnings call that existed.

The numbers deserve their reputation. Revenue of $96.2 billion, up 106% year-over-year and 18% sequentially. Data center revenue of $89.0 billion, up 117% year-over-year, against expectations of roughly $85.7 billion. Gross margin held at 75.0%. EPS of $2.22 cleared a consensus around $2.06 to $2.09. The stock rose about 9% the next session, adding roughly $440 billion of market cap in a single day.

But I read transcripts for a living, and the transcript matters more than the print. Management framed demand as structural, not cyclical: "The surge in AI demand is driving a global infrastructure buildout, supported by an expanding and diverse set of growth opportunities, spanning hyperscalers, AI labs, AI natives, enterprises and sovereign customers."

Then they put a number on it: "We expect to grow revenue by approximately 70% in fiscal 2028. This is a supply-constrained outlook." Supply-constrained is the phrase to sit with — in their words, "more compute drives more revenue as new GPU capacity comes online." Demand is not the variable anymore; capacity is. And on the biggest geopolitical question, the answer was blunt: "there is no China data center compute revenue in our forward outlook."

I unpacked the downstream consequences in NVIDIA's $96 billion quarter: what it means for everyone else. The short version: when the largest supplier in the ecosystem calls its own outlook supply-constrained, every company downstream of it inherits that constraint, and analysts on every AI-adjacent call from here on will ask about it.

Power Was the Biggest Theme — Again

Look back at the table. The number one theme of August was not rate cuts, and it was not AI capex. It was "data center power," matched by 1,180 companies — more than a quarter of every company that reported in the month. "Power constraints" showed up on calls from 520 companies, "AI capex" from 852, "inference demand" from 167, and "sovereign AI" from 123.

AMD's August 11 call put it plainly: the binding constraints on the AI buildout are "land and power," with the company pointing to its first full gigawatt with Anthropic coming online in 2027. Read that again — a gigawatt as the unit of account for a chip company's customer relationship. That is where this market operates now.

None of this is new; it is confirmation at higher volume. I traced the trend earlier in Power is the new bottleneck in earnings calls, and August did nothing to weaken the thesis. When the loudest company of the month says supply-constrained and the broadest theme of the month is electricity, those are the same story told at two altitudes.

The Fed Pivot

Jackson Hole delivered the macro moment of the month. Powell signaled openness to a September cut with the sentence traders were waiting for: "the baseline outlook and the shifting balance of risks may warrant adjusting our policy stance." The market reaction was immediate — the Dow jumped roughly 800 points to a record close, the S&P 500 gained about 1.5%, the Nasdaq about 1.9%, and traders priced September-cut odds above 90%.

In the corpus, 985 companies discussed rate cuts on August calls, the second-biggest theme of the month. But here is the detail I find more interesting than the headline count: the theme grew slower than the reporting volume. July had 709 companies on the phrase; 709 to 985 is about +39%, against a baseline of roughly 66% more companies reporting. Adjusted for volume, rate-cut talk actually lost share of the conversation.

Two readings, both probably true. First, the theme was near saturation before Powell said a word — when a fifth of all reporting companies already discuss rate cuts every month, there is not much room left to grow. Second, a large share of August's calls happened before the speech, so the Powell effect will land in September and October transcripts, not August ones. That is a hypothesis the data can test next month, and I intend to.

The Tariff Refund Wave Rolled On

Same pattern, different theme. 313 companies mentioned "tariff refund" in August — and Q3 is already the biggest quarter on record for the phrase, the wave set off by the Supreme Court's IEEPA ruling. Month-over-month, though, 205 to 313 is about +53%, under the volume baseline. The shock is maturing into process.

You can see the maturity in what companies say. The early phase was contingency language — whether refunds would come, how much, when. The August phase is allocation language: what the money is being used for. Walmart's August 20 call is the cleanest example of the genre — tariff refunds deployed straight into price. That is a strategic decision hiding inside an accounting event, and it says a lot about where retail thinks its next fight is.

If you want the full anatomy of this wave — the ruling, the mechanics, the quarter-by-quarter numbers — I wrote the deep dive in The tariff refund quarter.

The Quiet Risers

This is the section I actually built this post for. While everyone watched NVIDIA, three themes re-accelerated against the volume baseline:

• Agentic commerce: 78 → 164 companies, about +110% against the +66% baseline
• Humanoid robot: 26 → 56 companies, about +115%
• Stablecoin: 14 → 29 companies, about +107%

All three roughly doubled while the corpus grew by two-thirds. That is the comparison that matters. In a month where nearly every raw theme count set a quarterly high simply because more companies were talking, the volume-adjusted view separates themes that gained share of the conversation from themes that merely showed up in a bigger sample. Rate cuts and tariff refunds fell behind the baseline. These three cleared it by forty-plus points each.

Yes, they are small bases — 29 companies on stablecoins is a rounding error next to 1,180 on data center power — but small bases beating the baseline is exactly where next year's big themes hide. Every giant theme on the board above was once a two-digit count that most people ignored.

Agentic commerce is the one I would watch most closely, because the speaker set is widening. This started as a payments-industry conversation — I covered that phase in Agentic commerce: what payments giants told investors — but in August it was Walmart, a retailer, closing its earnings call with remarks about "leading in agentic experiences." When a theme jumps from the companies that process transactions to the companies that sell you groceries, it is moving down the stack toward the customer.

Every count in this post is reproducible against the live corpus with one request:

curl 'https://earningscalls.dev/api/v1/search?q="agentic commerce"&type=transcripts&date_from=2026-08-01&date_to=2026-08-31' -H "X-API-Key: $KEY"

Swap the phrase, shift the date window, and you have your own theme monitor. If you want the fully automated version with charts and a baseline adjustment built in, I published the complete build in Build an earnings call theme tracker in Python.

What September Brings

September is a lighter reporting month, which makes it a better listening month — fewer calls, more time per transcript. The setup going in: a rate cut the market treats as near-certain, an NVIDIA supply-constraint framing that every AI-adjacent management team will be asked to respond to, a tariff refund wave shifting from windfall to allocation, and three small themes compounding faster than anyone's coverage list accounts for. I laid out the full watchlist in the September 2026 earnings preview.

My own plan is simple: keep running the same searches, keep applying the same baseline discipline, and watch whether the Powell effect shows up in rate-cut counts once post-Jackson-Hole calls dominate the sample. Everything I used for this post — 253,000+ transcripts, 12,799 companies, 11.99 million speaker segments, coverage from 2020 to today — is the same data you can query yourself through the REST API or wire into your own tools via the MCP server.

Want to run these searches yourself? Grab an API key at earningscalls.dev and query 253,000+ earnings call transcripts in minutes.