AI infrastructure · Nvidia reports tonight · August 26, 2026

Nvidia Earnings: The Next Guide Sets the Grade

Author Brandon Leon Posted Wednesday, August 26, 2026, before the print (results ~4:20 p.m. ET; call 5:00 p.m.) Coverage The setup · the test, published in advance · the backstop · Friday's podium

TL;DR: Nvidia reports after Wednesday's close. Consensus sits at about $92.1 billion versus Nvidia's $91 billion guide — roughly 97% growth — in a market that sold at least the last two post-earnings reactions. We think the quarter itself is the least interesting number on the page; one number sets the initial grade tonight. We are publishing the framework before results: an FQ3 revenue guide of roughly $100 billion or more would support the acceleration case; a guide below roughly $95 billion would trigger a formal review of the position. Neither outcome alone settles the thesis. Also inside: the residual-value guarantee — capped at $105 billion — that Nvidia filed on August 17 backing OpenAI's Ohio data-center leases, the CEO's public $1 trillion Blackwell-and-Rubin claim, why Fed Chair Kevin Warsh's Jackson Hole speech 36 hours later may matter as much as the print, and a housekeeping note: this piece comes with our first Nvidia disclosure.

1. The setup: 97% growth is the expectation

The setup is unusual. Wall Street expects 97% year-over-year revenue growth — roughly $92.07 billion against the company's own $91 billion guide — and the honest question is whether even that can satisfy a market that has sold recent beats: the February report beat and closed down 5.5% the next day; May beat and closed down about 2%. The options market prices roughly a ±7% move tonight against realized post-print moves that have run well under that.¹ The stock comes into the print off a seven-session slide before Tuesday's bounce — positioning, not news.

That fade pattern is the tell about what tonight actually is. When a company doubles its revenue and the stock sells off, the market is not grading the quarter — it is grading the durability of the growth, and no single quarter can prove durability. The valuation says the same thing from the other side: the stock enters the print at approximately 22 times forward earnings, the bottom third of its own three-year range, after a year in which the multiple compressed while earnings roughly doubled.⁵ The de-rate already happened. What's left is the argument about the E — which is why we read tonight the way we read every governed event: one number, thresholds written in advance, graded in public afterward.

Chart 1 — Tonight's bar

The company guided $91 billion. The Street expects more. The whisper expects more than that.

Guide per NVIDIA's May 20 release; consensus and whisper per fn1. Tonight's actual is graded against this chart, in public, in the Sunday outlook.²

Chart 2 — The revenue path tonight extends (or doesn't)

Three quarters of acceleration on a $70–90 billion base. The fourth bar is what tonight decides.

February and May prints per NVIDIA releases; tonight's expectation per consensus (fn1); the $95B/$100B lines per §2's published test. Bars from zero; the dashed bar is drawn to the $100B threshold, not a prediction.¹

Chart 3 — The multiple already paid its toll

The stock enters the print in the bottom third of its own three-year valuation range.

Range and current multiple per the desk's August 18 underwrite review (approximate by construction — forward-multiple bands shift with estimates); the compression-while-earnings-doubled observation per the same review.⁵

Simply: Nvidia is expected to nearly double its revenue from a year ago, and the market has recently sold the stock even on good news. Tonight is about whether the company convinces investors the growth keeps going.

2. The test, published before the event

Quarterly results describe the past; for a company priced like this one, the only load-bearing number is the next quarter's revenue guide, which arrives at roughly 4:20 p.m. ET today. We wrote our test into the desk's underwrite a week ago, on August 18, and we're publishing it now — before the release — so it can be graded in public either way: an FQ3 revenue guide of roughly $100 billion or more would support the acceleration case; a guide below roughly $95 billion would trigger a formal review of the position — a review, not an automatic sale; the distinction is the framework. Neither outcome alone settles the thesis. Between the two sits the base case: good, decelerating, and — on recent form — possibly sold anyway.

Why those lines and not others: at $100 billion-plus, sequential growth is still compounding on a $90 billion base, which would be the hardest thing any company has done at this scale — the acceleration itself is the thesis. Below $95 billion, the arithmetic is the argument: against roughly $92 billion expected tonight, a sub-$95 billion guide would put sequential growth near 3% or less — the flattest quarter-over-quarter step Nvidia will have guided in this buildout — and the honest question a review must answer is whether that's digestion (power, packaging, timing) or deceleration (demand). One guide cannot distinguish the two, which is why a miss arms a review rather than a sale, and why the October hyperscaler capex guides — not tonight — remain the real fork for everything this desk covers.

Chart 4 — The test, on the record before the release

Tonight's guide lands somewhere on this line. We grade it in public either way.

Thresholds per the desk's August 18 underwrite; both are approximate by design and labeled as estimates there. Tonight's guide per the NVIDIA release, graded Sunday.²

Simply: The number that matters tonight isn't this quarter's revenue — it's what Nvidia says the next quarter will be. We committed to our decision lines a week ago and we're publishing them this morning, before the announcement, so you can hold us to them: above $100 billion means the growth machine is still speeding up; below $95 billion means our rules require a formal re-examination of the position we hold — a re-examination, not an automatic sale. Sunday's note reports the grade.

3. The backstop, the book, and Friday's podium

Housekeeping — a first disclosure This is our first published note on Nvidia, and it comes with a disclosure the regulars haven't seen before: we hold NVDA, accumulated in 2024 — before this site existed. The deployment series you've followed since July tracks every new dollar with published gates; the personal book underneath it also carries older positions that predate the public work. This note is not a trade alert or a price target; it applies a prepublished framework to a position we already own. Positions are disclosed whenever a covered name is discussed.

Two things hardened in the week before the print. First, the CEO put a public number on the order book: per press accounts of his recent remarks, Jensen Huang claims $1 trillion in total Blackwell and Rubin sales between 2025 and the end of calendar 2027.³ Second, the OpenAI relationship turned concrete, and the mechanics matter: in a document filed August 17, first reported by Nikkei and Fortune, Nvidia agreed to residual-value guarantees — capped at $105 billion — backing OpenAI's leases at a SoftBank-backed data-center campus in Ohio. In the base case, where OpenAI pays its rent, Nvidia pays nothing; the guarantee bites only if OpenAI stops paying and the space can't be re-let for enough. On the equity side, Huang says a $100 billion investment is "probably not in the cards" and roughly $30 billion is likely as far as Nvidia goes. This is potentially underwriting a portion of a major customer's infrastructure commitments — and the 10-Q matters because it may clarify how that exposure is carried. If material commitments are disclosed, the useful distinction is between ordinary ecosystem support and arrangements that shift demand risk onto Nvidia's balance sheet. The bull case says $725 billion of budgeted hyperscaler capex makes the OpenAI tranche a rounding error; the cautious case — ours — says marginal dollars are where cycles turn, and this one is guaranteed by the seller.³

Chart 5 — The circle we read the 10-Q for

Money flows both directions between Nvidia and its most important new customer. Both readings of that fact fit — the 10-Q decides which.

Guarantee mechanics and the Huang remarks per fn3 (the guarantee is capped and contingent — in the base case Nvidia pays nothing); hyperscaler capex per the July prints as covered in the deployment series. A two-way flow is a documented fact; the reading of it is the argument.³

The week isn't over when the call ends: Fed Chair Kevin Warsh delivers his first Jackson Hole keynote Friday morning, August 28, per the symposium schedule — 36 hours after this print — into a market where the July minutes showed broader hike sympathy than the vote count and the long bond sits near multi-decade highs. A hawkish Friday would test every long-duration growth multiple in the market, this one included, no matter what tonight's guide said. Broadcom's September 2 print then gives the first clean read on custom-silicon growth against Nvidia's merchant line.⁴

What makes this note wrong

The sold-beat pattern is two clean data points, not a law, and after-hours reactions routinely reverse by the next morning — whatever happens at 4:21 p.m., the grade belongs to Sunday, not to tonight's tape. Our thresholds are estimates and could be wrong in both directions: an acceleration guide can still be undone by October's capex guides (tonight is necessary, not sufficient), and a light guide can be digestion — power, packaging, timing — rather than demand, which is exactly the question a review exists to answer. And the $105 billion guarantee cuts both ways: it is either the strongest possible signal of demand confidence, or the clearest evidence that demand needs underwriting. Both readings fit today's facts.

Simply: We own Nvidia stock and are saying so plainly — this is our first article about it. Beyond tonight's numbers, we're watching Nvidia's new backstop of a major customer's rent bills and the Fed chair's Friday speech. Sunday's note grades everything against the lines published here.

Get the next note by email → All writing

Sources & footnotes

Setup: consensus FQ2 revenue $92.07B / adj EPS $2.09, ~97% y/y growth vs. $46.7B a year ago; the company's own guide $91.0B ±2% (May 20 release) (Motley Fool, Aug 25; Fool, Aug 24). Post-beat reactions: Feb 25 report beat and closed −5.5% the next session (~$260B of market cap, per TradingKey's contemporaneous coverage); May 20 report beat and closed about −2% (desk print-history review, Aug 18). The ±~7% implied move is approximate, per options pricing checked Aug 18 in the desk underwrite; consensus figures as of Aug 24–25 coverage. Seven-session slide into the print, Tuesday rebound (Bloomberg, Aug 24; Yahoo, Aug 24). ↩
Results post tonight at investor.nvidia.com (~4:20 p.m. ET; call 5:00 p.m.). The guide is graded against §2's thresholds — as written, wherever it lands — in Sunday's outlook. ↩
The Ohio guarantee: residual-value guarantees capped at $105B backing OpenAI's leases at the SoftBank-backed SB Energy campus, per a document Nvidia filed Aug 17 (Nikkei Asia; Fortune, Aug 18 — noting the figure came in well below earlier ~$250B reports; mechanics per Motley Fool, Aug 25: in the base case, where OpenAI pays its rent, Nvidia pays nothing). Huang on equity: $100B "probably not in the cards," ~$30B likely the extent (Yahoo Finance). The $1T Blackwell+Rubin claim through CY2027 (Yahoo Finance, Aug 24). Hyperscaler 2026 capex ~$725B budgeted per the July prints as covered in the deployment series opener. ↩
Jackson Hole: Warsh keynote Friday Aug 28 (Kansas City Fed); July minutes breadth per the minutes ("several participants favored an increase"). AVGO FQ3: Sep 2 (company IR). ↩
Valuation basis and timestamps: "forward" = next-twelve-month (NTM) consensus EPS; ~22.5× at the desk's Aug 18 pull ($219.99), roughly 22× at Tuesday Aug 25's $213.05 close (derived from the same estimate base). Own-range band roughly 20× (Apr 2025 trough) to 45×+ (2023–24 peaks), per the Aug 18 underwrite review of stockanalysis.com and MacroTrends data; trailing EPS grew ~110% over the year through mid-August while the trailing multiple compressed from the mid-40s (February peak) to the mid-30s, same sources. Forward-multiple bands are approximate by construction and shift with estimates. ↩

Methodology: tests are graded against their written wording. Quotes are verbatim where primary documents or official transcripts are available; other factual claims are sourced in the notes. Fills and positions are disclosed only when real. Nothing here is a price target.

Nothing on this page is investment advice. We work in scenarios and trip-wires, not predictions. See disclaimer.

Disclosure: I/we have beneficial long positions in NVIDIA, a memory-sector ETF (ticker DRAM), ASML, and a diversified quantum-computing ETF (QTUM) through stock ownership. I/we have no position in Broadcom, Alphabet, or Micron. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. This commentary is for informational and educational purposes only and does not constitute investment, tax, or legal advice. Past performance is not indicative of future results. See disclaimer.