Built by Grok 4.6 High

NVDA | Earnings Preview

Q2 FY2027 (ended July 26) has not printed. NVIDIA reports Wednesday, August 26, 2026, after the close; written CFO commentary posts ~1:20 p.m. PT and the call is 2:00 p.m. PT. Last-reported baseline is Q1 FY2027 (May 20): revenue $81,615 million (+85.2% YoY vs $44,062 million), Data Center $75,246 million (+92.4%), GAAP / non-GAAP GM 74.9% / 75.0%, non-GAAP EPS $1.87 (SBC now in non-GAAP). That print cleared the $78.0 billion midpoint by +4.6% and the high end of the ±2% band.

Growth trajectory — still accelerating, into a larger base. Reported YoY revenue has re-accelerated for three printed quarters, and the live guide itself extends that acceleration:

Even the bottom of the ±2% band accelerates versus Q1’s +85%. Street’s Q3 number is a 1,280 bp deceleration from the Q2 guide’s +94.7%. That is the first fade the models have written in after four accelerating prints — and it is not something management has guided. Colette said Vera Rubin starts in Q3 on top of a still-constrained Blackwell book.

Watch items into Wednesday

• Guidance / the real bar. NVIDIA guides one quarter of revenue, GM, and OpEx — never EPS, never the year. Live Q2 band is $91.0 billion ±2% ($89.2–$92.8B), no China Data Center compute assumed. FMP street $91.93B / $2.08 sits inside the top quarter of that band. A 12-for-12 beat of the high end and a last-four average +4.7% versus midpoint put the whisper near $95.3B. An in-line-to-guide print is a de facto miss versus history. The Q3 guide (issued on this call) is the event: a sub-$102B number against $103.68B street would be the first outlook-vs-Street miss of the cycle even if Q2 clears $91.9B.
• Tariffs / China. The Q2 number zeroes China DC compute. Licensed H200 exists (25% tariff, US inspection); Commerce U/S Kessler told Congress on July 14 that shipments have been “very few” / “trivial.” Q1 China HQ revenue was already only $4,550 million (−52.9% YoY). Any recognized H200 dollar is unmodelled upside, not the thesis. A BIS Blackwell-diversion inquiry and a Taiwan smuggling detention raise the odds it stays economically zero.
• Gross margin is the honest line. Revenue is the sandbag. Q2 nGAAP GM 75.0% ±50 bps is a hold, not a floor they have been beating. Q1 printed exactly on 75.0%. Below 74.5% into a Blackwell-heavy mix is a cost/mix signal. FY is still “mid-seventies.”
• Quality of demand. Post-print news locked 2027–28 offtake and financing ($500B compute MOUs; $105B residual-value cap on 4.25 IT-GW at PORTS-Pike) without revising Q2. That is the circular-financing critique the Street will test. It does not change whether they print $91B.

Peer tape into 8/26 is uniformly raised (TSMC capex $60–64B, AMAT packaging >70% and a same-July-quarter $10.25B guide, GOOGL/MSFT/AMZN/META all lifted or held AI capex). There is no demand-side excuse. “Demand is strong” is no longer differentiating.

Issued May 20 on the Q1 FY27 call / 8-K EX-99.1. Never revised. NVIDIA does not guide EPS, EBITDA, operating margin, or any segment. Classification: CONSERVATIVE on revenue (China zeroed; 9-for-9 / 12-for-12 above the high end); IN-LINE on margins; TIGHT versus today’s Street.

Verbatim (Colette Kress, Q1 FY27 / transcript): “Total revenue is expected to be $91 billion, plus or minus 2%. We expect sequential growth to be driven primarily by data center. … We are not assuming any Data Center compute revenue from China in our outlook.”

Implied P&L (not company guidance). GAAP GM 74.9% on $91.0B minus GAAP OpEx $8.5B → ~$59.7B GAAP operating income (~65.6% OM) vs Q2 FY26 60.8% (operating income $28,440 million / revenue $46,743 million) — ~+480 bps YoY.

Street catch-up. At the May 20 print, then-Street sat near $86.8B (CNBC / LSEG). The $91.0B guide was a ~$4B raise. Street has since crawled to the top of the company band. That is sandbag-then-catch-up, not an aggressive-guide / Street-still-below setup. The leftover cushion versus Street is the smallest of the last five quarters.

Annual trajectory (not a quarterly print).

FY2026 decelerated 4,870 bps through the Hopper-to-Blackwell handoff and the H20/China hole. Street’s FY2027 +82% is a re-acceleration off that trough. Q1 already printed +85%; the Q2 guide at +95% is consistent with that re-acceleration, not a peak-and-fade. FY2028 street $563.60B is +43% — the first sharp annual deceleration the models have booked.

YoY only. Internal estimates are n/a. NVIDIA is a semiconductor / AI-infrastructure name — the driver stack sits above consolidated revenue (Data Center Compute + Networking; new Hyperscale vs ACIE cut; China DC compute explicitly zeroed).

(a) Current quarter — Q2 FY2027 (upcoming)

FMP /analyst-estimates has no 2026-07-26 quarterly row; the $91.93B / $2.08 pair is print-level /earnings (epsActual / revenueActual still null as of 2026-08-17). Segment-level street is unavailable (no VA / Bloomberg).

Last-reported mix (Q1 FY27) — the baseline this print will be read against

Q1 ACIE $37,377 million already matched Hyperscale $37,869 million on the new cut. If that mix persists, a Q2 beat (if the sandbag holds) is more likely an ACIE / networking story than a hyperscaler-only story. Jensen’s “we should grow faster than hyperscaler CapEx” claim lives or dies on a second quarter of that split.

Non-GAAP definition change. Beginning Q1 FY27, non-GAAP no longer excludes SBC. Use GAAP OpEx for YoY. Anyone still bridging to the old $1.62 Q4 nGAAP EPS (restated $1.59) will invent a miss.

(b) FQ+1 — Q3 FY2027 (expected; will be issued Aug 26)

The Q3 guide does not exist. Italic = expected from the conservative-guide policy, not issued. NVIDIA has never skipped a next-quarter revenue / GM / OpEx guide.

Implied Q3 YoY: $103.68B vs $57,006 million = +81.9% — a 1,280 bp deceleration from the Q2 guide’s +94.7%. Why ~$102B ±2% is the expected guide, not $103.7B: the last seven next-quarter guides were set 2.5–8.0% below the eventual print and, recently, inside or just under then-current street. Repeating that policy into a street number that already requires YoY deceleration is the risk. Vera Rubin “production shipments … starting in Q3” (Kress; GTC Taipei “this fall”) is the one fundamental that could justify guiding to or through street.

(c) FY+1 — FY2028 (unguided) and the standing FY2027 framework

NVIDIA does not issue FY revenue or FY EPS guidance — none, in any year.

FY2028 vs FY2027 street: revenue +43.2%, EPS +42.0% — a sharp deceleration from FY2027’s +82.3% / +88.7% versus FY2026 actuals ($215,938 million / nGAAP $4.77 old SBC-out basis). Arithmetic check: Q1 actual $81,615 million + Q2 street $91.93B + Q3 $103.68B + Q4 $117.10B = $394.3B, which foots to the $393.7B annual average. A Q3 guide that holds +90% would force FY2027 revisions higher.

Tone into the quarter

NVDA guides one quarter ahead, not the year. The live guide is Q2 FY2027, given May 20, never revised. Daloopa discover_company_documents for calendar 2026Q2 returns nothing that touches the number. Street $91.93B is +1.0% above the midpoint and only $0.89B below the high end.

Conviction read: The $91B band is a supply-constrained floor, not a demand forecast. They have now issued four successive next-quarter guides at $54B, $65B, $78B and $91B — each a same-quarter YoY acceleration — and beaten the last one by $3.6B versus the $78.0B midpoint. An in-line print to the midpoint would be a deceleration versus the Q1 beat magnitude and would sit below street. A China-inclusive beat is not in the script.

Assumptions still live

• No China DC compute. Kessler’s “very few” shipments (Jul 14) is not a guide-breaker at $91B scale; it is possible low-single-digit $B upside they will be asked to quantify.
• Sequential growth is Data Center. Q1 Edge was $6.37B (+29% YoY); consumer/gaming is a memory-price headwind, not a DC offset.
• Blackwell is still the Q2 shipment mix. A Q2 Rubin mix surprise would be the first GM risk.
• Supply can deliver ~$91B. Q1 raised total supply (inventory + POs + prepaids) to $145B. GTC Taipei (Jun 1) then said Vera Rubin is in full production and the Rubin supply chain is 2× Grace Blackwell — a 2H statement, not a Q2 miss risk.
• ACIE grows faster than Hyperscale. Q1: Hyperscale $37.87B (+115% YoY); ACIE $37.38B (+74% YoY) — already a coin flip of DC. Wednesday’s split is the tell.
• DSO returns to the mid-50s. Q1 was 45 days on “favorable timing of collections.” A mid-50s DSO on a $91B quarter is a large AR step-up, not a demand miss. FCF $48.55B will not repeat 1:1 if AR normalizes.

Tone arc — adjectives up, China box unchanged

Demand language has escalated every print for four quarters, the visibility number has doubled ($500B through CY26 → $1T through CY27), and China language has not moved at all.

Net shift Q4 → Q1: more aggressive on demand adjectives and the visibility number, more generous on capital return ($20.0 billion returned; new $80B buyback; dividend $0.01 → $0.25), slightly more expensive on OpEx, unchanged on China = $0 and GM = mid-70s. They will not use Wednesday to introduce a demand-slowdown narrative they have spent three calls killing. The thing management is pounding that the Street still under-owns is ACIE + standalone Vera, not “will hyperscalers spend.”

Post-guidance updates (May 20 → Aug 17)

Nothing in the post-print tape cuts the $91B. Several items confirm the China box or pull Rubin to the left (a 2H / Q3+ story).

What Wednesday has to do

1. Clear the $91.0B midpoint and talk to $91.93B / the high end. An in-line-to-guide print is a miss versus whisper after four years of +3–5% beats.
2. Give a Q3 number that does not require the Street to fade Rubin. A Q3 guide that does not hold the +90% YoY slope (a Blackwell-only shape versus a Rubin start) would be read as a ramp slip, not conservatism.
3. Split Hyperscale vs ACIE. A Hyperscale-only beat with ACIE flattening would validate the “six customers, one book” bear.
4. Quantify China or re-zero it. Kessler opened the door to “very few.”
5. Defend GM at 75.0% ±50 bps into a Rubin transition. Any “temporary mix” language is the first crack.
6. Do not let August’s financing headlines substitute for a Q3 guide. The $500B MOUs and Ohio GW are 2027–28 architecture.

Ranked by information value on August 26. “Consensus” = FMP / published sell-side. Bloomberg and Visible Alpha are not connected — treat every street figure as lower confidence than a terminal pull.

Scorecard on the 26th

Where management and the Street actually disagree (the only contrarian setup is a credible team repeating something bullish the Street will not underwrite):

1. The acceleration is not finished. Management guided +94.7% YoY. Street’s Q2 accepts most of that. Street’s Q3 (+81.9%) does not. Colette did not guide a deceleration; she guided a Rubin start.
2. ACIE is a second company the Street does not model. Hyperscale $37.9B and ACIE $37.4B are already a coin flip. FMP has no ACIE line.
3. Standalone Vera is incremental, and it is this year. $20B excluded from the $1T. Street models treat Vera as BOM.

China is not a real disagreement — both sides have it at zero in Q2.

Window: 20 May 2026 (Q1 print) through 17 Aug 2026. Company and brand news excluding the Q1 print itself and the 26 Aug call notice. Post-print news is not a product-cycle surprise. It is a demand-lock plus financing-risk stack: contracted land/power/shell, HBM, packaging, and third-party capital, while China stays economically zero.

What this does / does not change

• Does not change Q2 FY27 guide math. Company already guided $91.0 billion ±2% with no China DC compute. Nothing above is a Q2 shipment beat in hand.
• Does change the FQ+1 / FY+1 conversation. Named Rubin offtake, contracted HBM4, and a Wall Street compute-financing wrapper are the exhibits management will use to defend the trajectory.
• Does raise the quality-of-demand bar. Equity into customers plus a $105B residual-value cap is the setup investing-principles flags: management is repeating a bullish demand claim (“compute is revenue”) that part of the Street now interprets as NVIDIA financing its own end-market.
• China is a residual, not a swing. Licensed H200 flow is trivial.

Explicitly excluded: Q1 print and 26 Aug call notice; Corning optical partnership (6 May, pre-print); Glassdoor / GeForce NOW / Nemotron product-blog; Indonesia and Armenia sovereign-AI ribbon-cuttings; analyst reiterates and any price/valuation tape.

NVDA is a Consistent beater. Versus FMP street it is 12 / 12 on revenue and 12 / 12 on non-GAAP EPS over the last 12 printed quarters, and 4 / 4 on both lines over the last 4. Versus its own revenue guide (midpoint, +1-quarter offset) it is also 12 / 12, and it printed above the high end of the ±2% band in every one of those 12 quarters. Beat magnitude is deteriorating in percent terms (Street has caught up) and improving in dollars (L4 average revenue surprise +$2.0B vs +$1.6B in the prior eight). Guidance is a floor, not a forecast.

The only live-tape blemish is FY2026Q1 EPS: contemporaneous Street was ~$0.85 and the $0.81 print was scored a miss after a $4.5 billion H20 charge. Current FMP has rewritten that estimate down to $0.737 and now shows a beat. Revenue still cleared both Street and the top of the company’s own band that quarter.

Versus own guide (the more informative bar). Offset: guide issued in Q(N) applies to Q(N+1). Last four:

Gross margin is the honest guide. nGAAP GM vs prior-quarter guide: in-line ±50 bps for four of the last five printed quarters. The only miss is FY26Q1 H20 (61.0% vs 71.0%, −1,000 bps). Q1 FY27 printed exactly 75.0% on a 75.0% guide. OpEx lands on the dollar — it is not sandbagged.

Bar into Wednesday

Street has moved inside the guide band. That is new versus FY27Q1, when FMP $78.42B sat only +0.5% above the $78.0B mid and the print still cleared $81.6B. An in-line-to-guide result at $91.0B would now be a ~1% Street miss and would break a 12-quarter “above the high” streak — that is the red-flag watch, not whether they “beat.” The residual that actually moves the setup is the Q3 guide.

Do not treat FY2027Q2 as printed.