Key takeaways
- NVIDIA (NASDAQ: NVDA) reports second-quarter fiscal 2027 results on Wednesday, Aug. 26. Written CFO commentary posts about 1:20 p.m. PT. The call is 2 p.m. PT / 5 p.m. ET on investor.nvidia.com. The quarter ended July 26, 2026. That is company IR, not a rumor calendar.
- Company outlook from the Q1 release: Q2 revenue $91.0 billion ±2%. GAAP / non-GAAP gross margin 74.9% / 75.0% ±50 bp. GAAP / non-GAAP opex about $8.5 billion / $8.3 billion. Explicit: “not assuming any Data Center compute revenue from China.”
- Q1 (ended April 26): revenue $81.6 billion (+85% year over year), data center $75.2 billion (+92%). GAAP diluted EPS $2.39; non-GAAP $1.87. The gap is mostly equity-mark noise. Board added an $80 billion repurchase authorization and lifted the quarterly dividend from $0.01 to $0.25. Authorization is a ceiling: stock buybacks.
NVIDIA earnings is the search because the company’s own number is already public and Wall Street is arguing about the number above it. The guide is $91 billion. Some desks are writing $93–$95 billion as “the real beat.” A print at $91 billion is a make. A print at $94 billion is what those desks already paid for. That is the setup. It is not a reason to buy or sell the stock.
The calendar
Do not mix three clocks:
- Aug. 26, ~1:20 p.m. PT — results and Colette Kress’s written commentary.
- Aug. 26, 2 p.m. PT — webcast. Analyst Q&A only.
- Aug. 27–29 — Jackson Hole. Payments theme. Not an NVIDIA slide. It still moves the cost of capital the hyperscalers use to pay for GPUs: Jackson Hole 2026.
The July FOMC minutes are already out. Nine held 3.50–3.75%; three presidents wanted a hike. That tape is lagging this print: FOMC minutes.
What Q1 actually printed
Source: NVIDIA Q1 FY27 release.
- Revenue $81.615 billion. Prior quarter $68.127 billion. Year-ago $44.062 billion.
- Data center $75.2 billion. That is the AI-factory line, not gaming.
- Edge computing $6.4 billion (+29% year over year) — autos, RTX, industrial. It is not the multiple.
- GAAP gross margin 74.9%. Non-GAAP 75.0%.
- GAAP net income $58.321 billion; non-GAAP $45.548 billion. Read both. GAAP ate a large equity-securities gain. Using only $2.39 EPS as “the run-rate” is how people overstate cash earnings.
The same release named Vera Rubin (Vera CPU + BlueField-4 STX), Dynamo 1.0 on Blackwell, and multi-year optics agreements with Coherent, Corning, and Lumentum. Packaging and fiber are the other bills on the same factory: AI semiconductor and optical stocks.
The $91 billion guide
$91.0 billion ±2% is a band of roughly $89.2–$92.8 billion. Street notes talking $94–$95 billion are not the company. If the print lands at $91.5 billion and Q3 guide is only a modest step-up, that can still sell off. If the print is $94 billion and Q3 is raised into the high $100 billions, that is a different tape. Do not treat “beat” as a binary.
Gross-margin guide: 74.9% GAAP / 75.0% non-GAAP, ±50 bp. Memory as a larger slice of bill of materials is the bear footnote in sell-side notes. NVIDIA did not put a memory-cost paragraph in the Q1 outlook bullets. Listen for it on Aug. 26. A 50 bp miss on 75% is a lot of dollars on $91 billion of sales.
Opex guided to about $8.5 billion GAAP. That is still a small tax on a $50-plus billion operating-income run rate. It is not the debate. Supply of Rubin/Blackwell and whether customers can power and finance the racks is the debate.
China, and the financing overlay
Official sentence: Q2 outlook assumes no data-center compute revenue from China. Any H20/H200 licensing news is upside to that model, or a footnote that does not hit the P&L this quarter. Do not add a China line to $91 billion and call it “the guide.”
Wires in mid-August described NVIDIA backing large OpenAI / data-center lease numbers (figures moved around $105 billion vs earlier higher talk) and a Wall Street financing platform aimed at hundreds of billions of GPU infrastructure. Those are not in the Q1 GAAP income statement as product revenue. They are balance-sheet and customer-finance questions. Ask: does the 10-Q later show guarantees, residual-value promises, or equity in the power developer? A press blog is not a 10-Q.
The July FOMC minutes already flagged AI-infrastructure borrowing and an equity premium last seen near the dot-com years. That is the Fed’s staff, not a Twitter thread.
What the print is not
Not a Fed decision. Not proof the 30-year at 5.3% “killed AI capex.” Not a China export license. Not a buyback executed — the $80 billion is room on the authorization. Not optical-vendor earnings. Those companies report their own quarters.
Withholding and estimated-tax calendars do not move because NVDA printed. Sept. 15 is still the third 1040-ES installment for people who owe estimates. Different machine.
After the call
- Revenue vs $91 billion ±2%, and whether China DC compute stayed at zero in the actuals.
- Q3 guide vs the street’s step-up, not vs a round number on TV.
- Gross margin vs 74.9% / 75.0% ±50 bp. Memory commentary.
- One sentence on financing / OpenAI / residual value, then go read the 10-Q when it files.
- Vera Rubin production language vs “late 2026” already in the Q1 story.
A 4.7% 10-year still sits under hyperscaler WACC. Duration math and GPU shipments are the same week, not the same formula: 10-year Treasury yield.
Education, not a recommendation to buy, sell, or short NVDA or any supplier. Guidance is the company’s. Street ranges are other people’s models. Spreads, taxes, and 10-Q footnotes are real. Check investor.nvidia.com for the live release.
