The September Setup: A Month of Tests for the Scoreboard

Where the Scoreboard Stands

Readers new to this site should start with The Conditions · Coverage, our public scoreboard. Every framework we publish carries explicit break conditions: observable developments that would prove the framework wrong, written down in advance so that we cannot quietly redefine success later. Six frameworks are live as of August 31, 2026, from The Three Waves of AI (set July 27) through The Displacement Cascade (set August 31), and all 29 conditions across them are marked holding. September is the first full month those conditions face live data. Every date below carries its source, and where a date is expected but not company-confirmed, we say so.

The Calendar, Sourced

Two items we looked for and could not pin to September dates: a specific NRC decision date on the first commercial SMR construction permits (legal-industry commentary from spring 2026 anticipated decisions during 2026 but named no month), and any NVIDIA-hosted event this month (GTC Washington runs November 30 to December 3, per NVIDIA's site). We publish what we can date and leave out what we cannot.

What September Could Trigger

Here is each catalyst mapped to the published condition it tests. The condition language below is quoted from our own pages; the full text and current status of every condition lives at /conditions.

September 2: Broadcom, the demand ratio's first witness

Broadcom reports fiscal third-quarter results on September 2, per its investor-relations announcement, with consensus near $29.4 billion in revenue per press previews as of late August. This is the same catalyst that stress-tested everything in June: on June 3, Broadcom's AI revenue guide of $16.0 billion against a consensus near $17.2 billion preceded the Nasdaq's 4.2% June 5 decline, per press reports of the call as documented in our Fourth Leg note. The print bears on our NVIDIA framework's second condition, "The demand ratio turning": "If volume growth slows below price deflation on a sustained basis, AI compute revenue shrinks while capex is still landing, and every infrastructure thesis breaks." One quarter cannot trigger a condition defined as sustained, but Broadcom's custom-accelerator trajectory is September's first dated evidence on whether AI compute demand is still outrunning its own price deflation.

September 4 and 11: the macro gate

The August jobs report arrives September 4 and August CPI on September 11, both at 8:30 a.m. ET per the BLS schedule, and the FOMC meets September 15-16, with the statement and an updated Summary of Economic Projections due September 16 at 2:00 p.m. ET, per the Federal Reserve's published calendar. These three dates test the Fourth Leg's fifth condition, "The macro overwhelming everything": "If a sustained rate-hiking cycle or an AI-financing credit event arrives, every leg of this framework de-rates together regardless of fundamentals." In June, consensus expected CPI near 4.2% and December-hike odds stood near 70% after a hot payrolls report, per market pricing at the time; September's dot plot is the first scheduled chance since our frameworks were set for the rate regime to confirm or relax that pressure. The jobs report additionally bears on the Displacement Cascade's first condition, "The unemployment crossover reversing," which tracks whether white-collar unemployment falls back below blue-collar.

September 8-11: Communacopia, where capex guidance gets previewed

The Goldman Sachs Communacopia + Technology conference runs September 8-11 in San Francisco, per the event's own page and corporate investor calendars including Microsoft's. Conference commentary is soft evidence, but it is the month's best scheduled venue for hyperscaler capital-spending language between earnings cycles, and that bears on two conditions at once: the NVIDIA framework's "Profitable-buyer capex rolling over" ("Two consecutive quarters of hyperscaler capex guide-downs would signal the buildout's foundation is cracking before any NVIDIA print would") and the Token Tollbooth's "The capex throttle failing," under which AI revenue decelerating materially while spending guidance keeps rising for several consecutive quarters breaks that framework. Note the asymmetry: one breaks on spending discipline arriving too fast, the other on it never arriving.

September 9 (expected): Oracle, the backlog referendum

Oracle is expected to report fiscal first-quarter 2027 results on Wednesday, September 9, per earnings-calendar aggregators as of August 31; the company had not formally announced the date as we published, and we flag that plainly. Oracle sits on both sides of the tollbooth: it is among the largest committed AI-infrastructure spenders, and its remaining performance obligations have been the market's favorite proxy for contracted AI demand. A strong backlog print is evidence that token demand is still being contracted ahead of supply; a soft one feeds "The capex throttle failing." We treat it as the month's cleanest read on whether the order book still supports the roughly $725 billion of 2026 hyperscaler capex, up about 77% year over year per company guidance and sell-side aggregations we cited in August.

September 10: Adobe, the seat-compression bellwether

Adobe reports fiscal third-quarter results September 10 after the close, per its scheduling announcement as carried by press coverage. Adobe is the market's chosen poster child for the "agents kill per-seat software" fear behind the first quarter's software drawdown, which makes it a live test of the Fourth Leg's second condition, "Seat loss showing up in retention": "If net revenue retention at the anchor platforms begins visibly reflecting seat compression across multiple quarters, the bear narrative the market priced in the first quarter of 2026 was correct, and we were the ones fighting the tape." The condition requires multiple quarters; September 10 supplies one dated point, in either direction.

September 15-17: AI Infra Summit, the 800VDC roll call

The AI Infra Summit runs September 15-17 in Santa Clara with an NVIDIA keynote by Ian Buck, per NVIDIA's events page. Infrastructure conferences are where power-architecture adoption gets discussed by the people writing the purchase orders, and that bears on the Fourth Leg's fourth condition, "800VDC failing to leave the reference design": "If major hyperscalers decline to adopt 800-volt DC distribution outside NVIDIA's own rack architectures, or the next rack generations slip materially, the power-silicon content story loses its slope." We will watch the public materials for named non-NVIDIA deployments of 800-volt DC distribution, or their conspicuous absence.

September 29-30: the nuclear calendar's quiet month

The World Nuclear SMR & Advanced Reactor Congress meets September 29-30 in Nashville, per the organizer's event page. We found no NRC decision formally scheduled for September; legal-industry commentary from spring 2026 expected decisions on the first two commercial SMR construction permits sometime during 2026, per an American Bar Association trends article. Our Energy + Nuclear + AI framework's conditions "First-power slippage past the procurement window" and "Regulatory reversal" are therefore on passive watch: no scheduled event can trigger them, but any slippage announcement or shift in NRC posture that surfaces, including at the Congress, would be dated evidence. Silence, in this case, is the condition holding.

September 30: Micron, the margin floor's best witness

Micron reports fiscal fourth-quarter results September 30, per its August 26 announcement. This is, in our opinion, the month's most consequential print for our scoreboard. NVIDIA's August 26 guidance put its gross-margin trough at 71-72%, attributed partly to memory costs, per the company's call, and our NVIDIA framework's fourth condition, "The margin floor breaking," reads: "A print below 71%, or a push-out of the recovery, would signal memory costs are structural, making earnings paths wrong." NVIDIA itself does not report in September, but Micron's pricing commentary and guidance are the best dated evidence on the input driving that condition. The tape into the print: TrendForce forecast third-quarter conventional DRAM contract prices up 13-18% quarter over quarter and NAND up 10-15% (July 3, 2026), noted PC DRAM contract prices still rising in its August 31 update, and separately expects HBM contract prices to move sharply higher in 2027 (June 2, 2026). If Micron's guide shows memory pricing still accelerating into 2027, the case that memory costs are structural rather than cyclical strengthens, and with it the pressure on the margin-floor condition. If pricing momentum moderates, the condition breathes easier.

The Tape Going In

Two late-August prints frame the month. Salesforce reported second-quarter fiscal 2027 results on August 26, 2026, disclosing Agentforce annualized revenue above $1.5 billion, up 240% year over year, per its press release; when the Fourth Leg framework was drafted in June, the comparable figure was $800 million growing 169%, from the February 25 report. Agent revenue accelerating from an already-inflecting base is the opposite of the Fourth Leg's first condition, "Agent revenue decelerating hard," and we log it as such. And NVIDIA's August 26 report, covered in our NVIDIA note, guided fiscal 2028 revenue growth near 70% on a supply-capped basis while guiding margins down to the 71-72% trough. September inherits both facts: the demand story intact by the company's own telling, and the cost story unresolved.

What Could Go Wrong With This Setup

The steelman against this note is not against any single framework; it is against the exercise. Stated as forcefully as we can state it:

Calendars measure what is scheduled, and breaks are usually not. The worst stretch of 2026 for the AI complex came from a guidance line inside an earnings call (June 3-5) that no calendar would have flagged in advance. A financing accident, a customer-guarantee disclosure, or a safety incident in the reactor class can arrive on any Tuesday. Mapping the month's scheduled events risks implying the risk is scheduled. It is not.

Single prints cannot trigger multi-quarter conditions. Nearly every condition on our scoreboard is defined over consecutive quarters precisely to avoid overreacting to one number. The honest corollary: September cannot vindicate us either. If we declare victory on October 1, we will have broken our own rules.

Watching for confirmation is not the same as watching. We have published views and may hold positions in the names discussed; assume we are talking our book, per the disclosures below. Mapping events to falsification criteria rather than to hopes is the counterweight, but the bias risk is real and it is ours.

The crowding cuts both ways on event days. Semiconductor positioning was at record crowding in the May survey data we have cited, per press summaries of the Bank of America Global Fund Manager Survey. Crowded tapes overreact to scheduled events in both directions, so September's prints may move prices far more than they move evidence. A violent single-session move on an in-line print would tell us about positioning, not about our conditions.

One date in this note is not confirmed. Oracle's September 9 date is an aggregator expectation, not a company announcement, as of publication. If it moves, the month's cleanest backlog read moves with it.

Our View, Stated as Opinion

Our opinion, and it is opinion: September 2026 is a month where the AI trade's evidence calendar is denser than its narrative calendar, which is the configuration we prefer. The prints that matter most for our published work are Broadcom on the demand ratio, Micron on the margin floor, and Adobe on seat compression, with the FOMC's September 16 projections governing whether any of it is audible over the macro. We expect, but do not promise, that all 29 conditions will still read holding on September 30, and if any does not, the scoreboard at /conditions will say so in public, dated, before we explain ourselves. That is the arrangement, and it is the whole reason to read us.