Date: 2026-02-23 Quarter: Q3 FY26 (three months ended October 26, 2025; reported November 19, 2025) Market cap: ~$3.4T | EV/TTM Rev: ~18x | Revenue growth: +62.5% YoY Note: Q4 FY26 earnings report due tomorrow, February 25, 2026.
NVDA Q3 FY26 was one of the strongest quarters any company has ever reported — 57.0Brevenue, largestQoQdollaraddinthecompany′shistory(+10.3B), first YoY re-acceleration in seven quarters, and all-time highs on every P&L line. The Blackwell ramp is not just working, it's compressing: GB300 crossed over GB200 to ~67% of Blackwell mix in a single quarter. The one genuine uncertainty heading into FY27 is gross margin sustainability in the mid-70s against rising input costs — management has committed but hedged. With Q4 earnings tomorrow, this is effectively a pre-Q4 positioning analysis framed around Q3 results. Conviction: 4/5. Would be 5 if FY27 margin guidance were more concrete than "working to hold mid-70s."
| Criterion | Threshold | Actual | Status |
|---|---|---|---|
| Revenue YoY growth | >30% (>40% preferred) | +62.5% | Pass |
| Gross margin | >60% (>70% preferred) | 73.4% GAAP / 73.6% Non-GAAP | Pass |
| Revenue per quarter | >$50M | $57.0B | Pass |
| Data availability | 4+ quarters | 12 quarters (35 in DB) | Pass |
| Share dilution | <10% annual | Declining — buybacks reducing count | Pass |
| GAAP profitability trajectory | Improving or positive | $31.9B net income, 56% net margin | Pass |
Gate result: PASS (6/6)
| Factor | Rating | Detail |
|---|---|---|
| Growth | Strong | +62.5% YoY; +22.0% QoQ. First YoY re-acceleration in 7 quarters. Largest QoQ dollar add in company history (+$10.3B). |
| Trajectory | Accelerating | YoY: 55.6% → 62.5% (+6.9pp). QoQ: +6.1% → +22.0% (+15.9pp). Reversal of two consecutive QoQ decelerations. Q4 guide implies +65% YoY. |
| Margins | High | GAAP GM 73.4%, recovering from Q1 trough of 60.5%. Non-GAAP op margin 66.2%. GAAP op margin 63.2% (+0.9pp YoY). FCF margin 38.7% (volatile Q-to-Q). FY27 mid-70s commitment hedged with "input costs rising." |
| Dominance | Dominant | 92% discrete GPU share; 86% data center GPU share. Every frontier AI model runs on NVIDIA. CUDA ecosystem moat deepening — A100 from 2019 still at full utilization. NVLink Fusion expanding to Fujitsu, Intel, ARM. |
| Valuation | Rich | ~18x EV/TTM revenue; ~44x TTM FCF; ~35x GAAP P/E. But earnings growing ~45% YoY; PEG converging toward 0.5x. Multiple compressing as earnings grow faster than stock price. |
| Special | Present | $500B Blackwell+Rubin pipeline visibility through CY2026; Rubin silicon received; Anthropic adopting NVIDIA for first time; OpenAI 10GW partnership in formalization; physical AI / robotics inflection; Q4 earnings tomorrow. |
| | Q4_FY23 | Q1_FY24 | Q2_FY24 | Q3_FY24 | Q4_FY24 | Q1_FY25 | Q2_FY25 | Q3_FY25 | Q4_FY25 | Q1_FY26 | Q2_FY26 | Q3_FY26 | | | Jan-2023 | Apr-2023 | Jul-2023 | Oct-2023 | Jan-2024 | Apr-2024 | Jul-2024 | Oct-2024 | Jan-2025 | Apr-2025 | Jul-2025 | Oct-2025 | |---|---|---|---|---|---|---|---|---|---|---|---|---| | Revenue ($B) | 6.1 | 7.2 | 13.5 | 18.1 | 22.1 | 26.0 | 30.0 | 35.1 | 39.3 | 44.1 | 46.7 | 57.0 | | YoY % | -20.8% | -13.2% | +101.5% | +205.5% | +265.3% | +262.1% | +122.4% | +93.6% | +77.9% | +69.2% | +55.6% | +62.5% | | QoQ % | +2.0% | +18.9% | +87.8% | +34.2% | +22.0% | +17.8% | +15.3% | +16.8% | +12.1% | +12.0% | +6.1% | +22.0% | | GM [GAAP] | 63.3% | 64.6% | 70.1% | 74.0% | 76.0% | 78.4% | 75.1% | 74.6% | 73.0% | 60.5% | 72.4% | 73.4% | | GM [Non-GAAP] | — | — | — | 75.0% | 76.7% | 78.9% | 75.7% | 75.0% | 73.5% | 61.0% | 72.7% | 73.6% | | Op Margin [GAAP] | 20.8% | 29.8% | 50.3% | 57.5% | 61.6% | 64.9% | 62.1% | 62.3% | 61.1% | 49.1% | 60.8% | 63.2% | | Op Margin [Non-GAAP] | — | — | — | 63.8% | 66.7% | 69.3% | 66.4% | 66.3% | 64.9% | 52.8% | 64.5% | 66.2% | | Net Margin [GAAP] | 23.4% | 28.4% | 45.8% | 51.0% | 55.6% | 57.1% | 55.3% | 55.0% | 56.2% | 42.6% | 56.5% | 56.0% | | EPS [GAAP] | — | $0.82¹ | $0.25 | $0.37 | $0.493 | $0.60 | $0.67 | $0.78 | $0.89 | $0.76 | $1.08 | $1.30 | | EPS [Non-GAAP] | — | — | — | $0.402 | $0.516 | $0.612 | $0.68 | $0.81 | $0.89 | $0.81 | $1.05 | 1.30||FCF(B) | 1.7 | 2.6 | 6.0 | 7.0 | 11.2 | 14.9 | 13.5 | 16.8 | 15.5 | 26.1 | 13.5 | 22.1 | | FCF Margin | 28.7% | 36.7% | 44.8% | 38.8% | 50.9% | 57.3% | 44.9% | 47.8% | 39.5% | 59.3% | 28.8% | 38.7% |
¹ Pre-split value; post-split equivalent ~$0.082.
TTM (Q4 FY25–Q3 FY26): Revenue $187.1B | FCF $77.2B | Net Income (GAAP) $98.2B
Segment breakdown (Q3 FY26):
Prior beliefs based on Q2 FY26 guidance ($54.0B ± 2%) and management commentary.
| Metric | Expected | Actual | Verdict |
|---|---|---|---|
| Revenue | $54.0B (midpoint; range $52.9-55.1B) | $57.0B | Beat — exceeded top of range by ~$1.9B |
| Non-GAAP Gross Margin | 73.5% ± 50bps | 73.6% | Met — in-line |
| H20 China revenue | 0–5B (upside possible if geopolitical resolves) | ~$50M | Missed — geopolitical headwinds persisted |
| GB300 vs GB200 mix | GB300 ramping, possibly approaching parity | GB300 ~67% of Blackwell revenue | Positive surprise — faster than implied |
| Data Center revenue | ~$46-48B (implied) | $51.2B | Beat significantly |
| GAAP Op Margin | ~60-62% range | 63.2% | Upside — operating leverage stronger |
| FCF | ~$18-22B | $22.1B | High end achieved |
| GAAP EPS | ~$1.05-1.15 | $1.30 | Beat materially |
Delta assessment: The revenue beat magnitude is the primary signal — $57.0B vs. 54.0Bmidpointis5.51.9B above the top of the stated range. This is enormous at this scale. The primary driver was DC compute jumping 9.2BQoQasGB300becamethedominantproduct.Critically, theH20near − miss( 50M vs. possible $2-5B upside) means the beat is cleaner than headline — Blackwell demand absorbed the China vacuum entirely. Supply commitments +63% QoQ to $50.3B and multi-year cloud agreements +106% QoQ to $26.0B are the forward demand signals management is managing to. The clean nature of the beat (organic Blackwell demand, no China help) is bullish for sustainability.
| Indicator | Q1 FY26 | Q2 FY26 | Q3 FY26 | Direction |
|---|---|---|---|---|
| Data Center Revenue ($B) | 39.1 | 41.1 | 51.2 | Re-accelerating strongly |
| DC Compute Revenue ($B) | 34.2 | 33.8 | 43.0 | +$9.2B QoQ after Q2 pause |
| Networking Revenue ($B) | n/a | 7.3 | 8.2 | Sustained; +162% YoY |
| Supply commitments ($B) | n/a | ~31 | 50.3 | +63% QoQ — strong forward demand |
| Multi-year cloud agreements ($B) | n/a | ~12.7 | 26.0 | +106% QoQ — locked-in demand |
| Hyperscaler 2026 CapEx est. ($B) | ~400 | ~400 | ~600 | +$200B in-year expansion |
| Blackwell/Rubin pipeline | — | — | $500B through CY2026 | Forward revenue anchored |
Assessment: Bullish divergence — leading indicators running well ahead of any single quarter's revenue. Supply commitments at 50.3Brepresent 2212.7B → 26.0B)isthemostimportantleadingindicator : hyperscalersarenotmakingspotbuys, they′remakingmulti − yearcommitments.Thispatternhasconsistentlyprecededquarterlybeats.TheDCcomputeQoQpauseinQ2(–400M) followed by the $9.2B surge in Q3 confirms it was a timing/supply issue, not demand softness.
Blackwell performance is third-party validated. MLPerf Training v5.1: NVDA swept every benchmark; GB200 NVL72 3.2x faster than H100 FP8 at same GPU count. SemiAnalysis InferenceMAX v1: Blackwell highest performance and best TCO across every model. 15x cost-per-token reduction vs. prior generation. These are independent benchmarks, not NVIDIA's own claims. [MLPerf, SemiAnalysis]
Market share: dominant, slightly eroding at margins. 92% discrete GPU share in Q3 2025 (down from 94% in Q2). 86% data center GPU share. AMD MI300X has real deployments (Meta Llama 405B live traffic at 100%; Microsoft ND MI300X v5; Oracle 50K-GPU cluster), but CUDA ecosystem delivers 10-30% better real-world performance on most workloads. Hardware spec advantage doesn't fully translate to deployment advantage. [WCCFTech, Tom's Hardware]
DeepSeek was noise. $593B single-day loss in January 2025 was market overreaction — the quarter confirmed demand unaffected. Ironically, DeepSeek-R1 benchmarks ran on Blackwell Ultra and showed 5x throughput per GPU vs. Hopper. The long-reasoning model transition structurally increases compute per token, making it a net NVDA positive. [Bloomberg, TechBrew]
Employee quality is a genuine differentiator. 98% CEO approval (Glassdoor, 3,809 reviews). 2.5% voluntary turnover vs. 16.4% semiconductor industry average. 95% of employees believe positive business outlook. Flat org with 60 direct reports to Jensen is unusual but produces speed. Risk: team-dependent experience quality. [Glassdoor, Blind]
NIM microservices showing real enterprise traction. Concrete deployments: Lowe's, Siemens, Box, Dropbox, Cohesity, Glean, Hippocratic AI. 2.6x throughput vs. off-the-shelf H100. IBM, Red Hat, Nutanix, VMware, Canonical all publishing NIM integration guidance — ecosystem breadth signals genuine adoption rather than lab interest. [NVIDIA NIM page, Glean case study, Introl]
Physical AI thesis: early but coherent. CES 2026 (Jan 2026): Jensen introduced Cosmos (world foundation model, open), Gr00t (humanoid robot VLA model), Alpamayo (AV reasoning model), Rubin platform confirmation. Framework mirrors the 2022-23 data center GPU inflection pitch — same conviction, same specificity. Robotics, AV, industrial automation all require NVIDIA's full stack. Current revenue contribution minimal ($592M Auto). [NVIDIA CES 2026, Axios]
Analyst consensus bullish pre-Q4. Citi (Atif Malik, Buy, $270 PT) expects Q4 to exceed $65.6B consensus. "Blackwell demand off the charts." Bear cases: HBM cost inflation + 25% government fee on China H200 sales could compress FY27 margins. [Markets Financial Content, Seeking Alpha]
| Metric | Current | 1Y Ago (Q3 FY25) | Assessment |
|---|---|---|---|
| Market cap | ~$3.4T | ~$3.6T (peak Oct-24) | ~5% below peak despite earnings +45% — valuation is derating |
| EV/TTM Revenue | ~18x | ~37x peak Oct-24 | Multiple compressing rapidly as earnings grow |
| EV/TTM FCF | ~44x | ~55x+ | Coming in; FCF growing faster than market cap |
| GAAP P/E (TTM) | ~35x | ~60x | Significant compression; approaching reasonable for growth rate |
| Non-GAAP P/E (est. fwd FY27) | ~27x | — | Reasonable if growth sustains 30%+ |
| FCF yield | ~2.3% | ~1.5% | Improving |
| TTM EPS [GAAP] | ~$4.03 | ~$2.79 | +44% YoY earnings growth |
Interpretation: NVDA has been in a quiet valuation reset for 16 months — stock flat since Oct-24 while earnings grew 44%. This is multiple compression, not earnings deterioration. At 35x GAAP TTM and 18x TTM revenue, NVDA is still not cheap. But the PEG ratio is approaching 0.5x (35x P/E / ~65% growth), which is historically cheap for a category-defining platform company. The argument against: at $187B TTM revenue, sustaining 65% growth becomes mathematically harder. The argument for: the demand signals (supply commitments, multi-year agreements, $600B 2026 hyperscaler capex) suggest management visibility into 30-50% growth through FY27.
Shareholder returns: $37.0B returned in first 9 months of FY26 (buybacks + dividends). $62.2B buyback authorization remaining. Share count declining at -1.2% YoY — accretive but modest at this market cap.
NVIDIA operates at the intersection of three simultaneous platform shifts:
CPU → GPU accelerated computing. Moore's Law ended; GPU acceleration is the only path to continued compute efficiency improvement. This transition is ~5-15% penetrated globally — a decade-long runway.
Generative AI infrastructure. Hyperscaler 2026 CapEx estimated at 600B(+200B vs. start of year). Foundation model builders (OpenAI, Anthropic, xAI, Mistral) competing to scale compute. Both Anthropic (1GW commitment, first time on NVIDIA) and OpenAI (10GW strategic partnership in formation) are effectively becoming NVIDIA customers directly, not just through cloud intermediaries.
Agentic + Physical AI. CES 2026 framing: "ChatGPT moment for physical AI is nearly here." 200+ robotics companies in NVIDIA's ecosystem. DRIVE Orin in production vehicles. Omniverse for industrial digital twins. Revenue contribution minimal today; could be the dominant demand driver by FY28-29.
NVDA is a platform, not a product: CUDA (6,000+ applications, 5.4M developers, 25 years of optimization), NVLink Fusion (being adopted by Intel, Fujitsu, ARM — becoming industry interconnect standard), NVIDIA Dynamo (open-source inference framework adopted by every major CSP). The installed base creates a recurring upgrade cycle. A100 (2019) is still at full utilization in 2025 because software optimization improved it continuously — this is the CUDA moat made concrete.
TAM: Management cites 3 − 4trillioninannualAIinfrastructurespendby2030.NVIDIAcontentpergigawatt: 30B (Blackwell), rising with Rubin. At $1T annual spend and 30B/GW : NVDA′sshare 30-35B per 33GW deployed. Current TTM revenue $187B. The path from $187B to $300-350B by FY28 is visible if these assumptions hold.
FY27 gross margin. Management committed to "mid-70s" but flagged rising input costs without specificity. HBM (LPDDR5X / HBM3E) pricing, TSMC CoWoS-L packaging, NVLink switch complexity — all are cost headwinds. If GM settles at 70-72% rather than 75-76%, at $250B revenue scale that's $5-10B less annual gross profit. This is the single most material financial risk.
ASIC displacement by hyperscaler custom silicon. Google TPU v7 (Ironwood, 4,614 TFLOPS FP8), Amazon Trainium2, Meta MTIA — all real products in production. Google's Anthropic deal for 1M TPUs shows the model; Anthropic simultaneously committed 1GW to NVIDIA. The same customer can use both. CUDA moat is real but not impenetrable; 2027+ is the realistic window for meaningful displacement.
Hyperscaler CapEx cyclicality. $600B 2026 aggregate CapEx is a lot of demand pre-placed. If inference capacity comes online faster than incremental demand through 2027-28, GPU orders slow. Historical precedent for enterprise tech CapEx cycles is concerning. The counter: AI inference demand (reasoning models, agentic workflows) may scale to absorb capacity as fast as it's built.
Export control escalation. H20 restrictions already zeroed out ~$10-15B annualized China revenue. New rules on Blackwell to additional markets (India, Middle East, Southeast Asia) would be material. The sovereign AI narrative and KSA/HUMAIN deal (400-600K GPUs over 3 years) increases geopolitical exposure in both directions.
Rubin execution. Seven-chip system; materially more complex than Blackwell. Blackwell itself had a design issue that delayed ramp ~3-4 months. A similar slip with Rubin in H2 2026 would reset FY27 estimates. Silicon received per Q3 commentary, but "executing to bring up beautifully" is Jensen being positive — not definitive confirmation of zero issues.
Q4 FY26 earnings tomorrow (Feb 25). Consensus $65.7B. NVDA has beaten guidance every quarter in this window. A beat-and-raise consistent with Q3's pattern would validate the forward demand narrative. The key variable: whether Q1 FY27 guidance implies acceleration or deceleration.
Rubin ramp (H2 CY2026). Promised "X-factor" improvement over Blackwell. NVIDIA content per GW expected to move from ~$30B (Blackwell) to higher (Rubin). A clean Rubin ramp would establish FY27 as a revenue step-change year similar to FY25 (Blackwell cycle start).
Physical AI inflection. 200+ robotics companies in NVIDIA's ecosystem. If the "ChatGPT moment for physical AI" arrives in 2026-27, it adds a demand driver uncorrelated to hyperscaler CapEx cycles. Current contribution is ~$600M/quarter (Auto). Even a 5x increase to $3B/quarter would be meaningful at scale.
Anthropic and OpenAI direct partnerships. First time Anthropic is on NVIDIA hardware (1GW commitment). OpenAI 10GW partnership in formalization. These remove the CSP intermediary layer and create direct hyperscale demand signals. Jensen: "NVIDIA's platform is the singular platform in the world that runs every AI model."
Multiple re-rating from earnings growth. Stock flat since Oct-24; earnings up 44%. If FY27 sustains 30-40% revenue growth with stable margins, forward P/E continues compressing toward market multiples while earnings compound. Requires no re-rating event — just time and continued execution.
No position held by this persona.
Sources: EDGAR 8-K EX-99.1 (Q3 FY26 press release, Nov 19, 2025), Q3 FY26 earnings transcript (Nov 19, 2025), CFO Commentary, scuttlebutt stage 2026-02-23, quant-prep stage 2026-02-23, transcript-digest stage 2026-02-23.