Built by Claude FableNVDA | Q2 FY2027 Earnings Review
What happened. NVIDIA delivered $96.2B in Q2 FY2027 revenue (+106% YoY, +18% QoQ), beating consensus by $3.95B (+4.3%) and its own guide midpoint by $5.2B (+5.7%). Non-GAAP EPS of $2.22 beat consensus by 6.2%. Data Center revenue hit $89B, with the new Hyperscale sub-segment at $49B and ACIE (sovereign, NeoCloud, enterprise) at $40B (+138% YoY). This was the 12th consecutive quarter of beating both consensus and the company's own guide high end.
What is new. Three material developments emerged: (1) NVIDIA issued its first-ever full-year forward guide, projecting FY2028 revenue growth of ~70% YoY (~$672B), explicitly noting this is supply-constrained against ~100% unconstrained demand. (2) Management proactively reset gross margin expectations downward from the 75% peak to a Q4 FY2027 trough of 71-72%, settling at 72-73% in FY2028, driven by HBM memory pricing that "exceeded prior expectations and is headed even higher." (3) A new NeoCloud revenue-sharing model was launched (hardware sale plus recurring rental revenue), alongside $500B+ in third-party capital partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR.
Tone. Markedly more confident than Q1 (first-ever forward guide), but also more transparent on negatives (proactive margin reset, preemptive defense of "circular financing" concerns). Management explicitly addressed the bear case that NVIDIA is financing its own customers: "We see it differently... our risk is limited. NVIDIA compute is fungible and durable."
Contradictions. Four material contradictions vs. prior-quarter statements: gross margin guide reversed from "mid-seventies" to 71-72% trough in 98 days; China DC revenue moved from emphatic "zero generated" to disclosed shipments; revenue per GW cited as both $40B and $60B within the same call; and OpEx growth guide revised up 300-500 bps in one quarter.
Upcoming catalysts. Vera Rubin ramp (~20% of Q3 DC revenue, fastest ramp ever), Goldman Sachs Communacopia keynote (Sep 10), GTC Berlin (Oct 21), Groq 3 LPX volume shipments, and the agentic AI demand inflection (15-100x compute per task vs. prompted AI).
| Metric | Q1 FY25 | Q2 FY25 | Q3 FY25 | Q4 FY25 | Q1 FY26 | Q2 FY26 | Q3 FY26 | Q4 FY26 | Q1 FY27 | Q2 FY27 |
|---|---|---|---|---|---|---|---|---|---|---|
| Revenue ($B) | 26.0 | 30.0 | 35.1 | 39.3 | 44.1 | 46.7 | 57.0 | 68.1 | 81.6 | 96.2 |
| YoY % | +262% | +122% | +94% | +78% | +69% | +56% | +62% | +73% | +85% | +106% |
| Data Center ($B) | -- | -- | -- | -- | ~39.9 | 41.1 | 51.0 | 62.0 | 75.2 | 89.0 |
| DC YoY % | -- | -- | -- | -- | -- | +56% | +66% | +75% | +88% | +117% |
| Non-GAAP GM % | -- | -- | -- | -- | -- | 72.7% | 73.6% | 75.2% | 75.0% | 75.0% |
| Non-GAAP EPS | $0.61 | $0.68 | $0.81 | $0.89 | $0.81 | $1.05 | $1.30 | $1.62 | $1.87 | $2.22 |
| EPS YoY % | +455% | +152% | +103% | +71% | +33% | +54% | +60% | +82% | +131% | +111% |
| Period | Rev Est | Rev Act | Rev Surprise | EPS Est | EPS Act | EPS Surprise |
|---|---|---|---|---|---|---|
| Q3 FY2025 | $33.2B | $35.1B | +5.8% | $0.75 | $0.81 | +8.0% |
| Q4 FY2025 | $38.1B | $39.3B | +3.2% | $0.85 | $0.89 | +5.0% |
| Q1 FY2026 | $43.3B | $44.1B | +1.7% | $0.74 | $0.81 | +9.9% |
| Q2 FY2026 | $46.1B | $46.7B | +1.5% | $1.01 | $1.05 | +4.0% |
| Q3 FY2026 | $55.0B | $57.0B | +3.7% | $1.26 | $1.30 | +3.2% |
| Q4 FY2026 | $66.1B | $68.1B | +3.0% | $1.54 | $1.62 | +5.2% |
| Q1 FY2027 | $78.4B | $81.6B | +4.1% | $1.76 | $1.87 | +6.3% |
| Q2 FY2027 | $92.3B | $96.2B | +4.3% | $2.09 | $2.22 | +6.2% |
Pattern: Consistent beater. L12Q revenue beat rate: 12/12 (100%). L4Q average revenue surprise: +3.8%. Revenue surprise magnitude has narrowed from the +9-11% range in FY2024 (when the AI ramp first shocked consensus) to a stable +1.5% to +4.3% corridor over the last 6 quarters. This reflects street catch-up to the trajectory rather than execution deterioration. NVIDIA continues to guide conservatively (zero China, midpoints well below delivery) and beat by a predictable margin.
Q3 FY2027 Guidance
| Metric | Q3 FY27 Guide | Consensus | vs Consensus |
|---|---|---|---|
| Revenue | $108.0B +/-2% | $108.7B | -0.6% |
| Non-GAAP Gross Margin | 74.0% +/-50bps | -- | -100bps vs Q2 |
| Non-GAAP OpEx | ~$9.0B | -- | +8% QoQ |
| Vera Rubin Mix | ~20% of DC revenue | -- | First production Q |
Gross Margin Reset Path
Actual
Guide
Trough
Settling
FY2028 Forward Guide (First-Ever)
This is unprecedented. NVIDIA has never guided a full fiscal year in advance. The ~70% FY2028 revenue growth implies ~$672B in revenue, materially above FMP consensus of ~$570B at the time of the call. Management was explicit that this is supply-constrained: unconstrained demand growth is "significantly" above 100%. The 70% floor reflects secured supply commitments, not a demand ceiling.
Implied revenue build:
| Quarter | Revenue | YoY Growth | Type |
|---|---|---|---|
| Q1 FY2027 | $81.6B | +85% | Actual |
| Q2 FY2027 | $96.2B | +106% | Actual |
| Q3 FY2027 | $108.0B | +89% | Guide mid |
| Q4 FY2027 | ~$115-120B | +69-76% | Implied |
| FY2027 Total | ~$401-406B | +86-88% | Estimated |
| FY2028 (at 70%) | ~$681-690B | ~70% | Company guide |
FY25
FY25
FY26
FY26
TROUGH
FY26
FY26
FY27
FY27
PEAK
Revenue-EPS divergence in Q2 FY2027. Revenue acceleration was +2,060 bps QoQ while EPS decelerated -1,950 bps. The divergence reflects rising operating expenses (+8% QoQ), a higher tax rate (16% vs. lower prior-year rate), and the approaching gross margin compression from memory pricing. This pattern will intensify through Q4 FY2027 (71-72% trough) before stabilizing in FY2028 as price increases take effect.
| Catalyst | Timing | Signal | Detail |
|---|---|---|---|
| Vera Rubin Ramp | Q3 FY27+ | Positive | ~20% of Q3 DC revenue (~$20B). Revenue per GW rises from $25B (Blackwell) to $40B. "Fastest ramp in NVIDIA's history." Purchase orders from every major hyperscaler. |
| Gross Margin Reset | Q3-Q4 FY27 | Negative | 75% to 74% (Q3) to 71-72% trough (Q4). HBM memory pricing surged 435% per rack vs. prior gen. Price increases take effect Q1 FY28, settling margins at 72-73%. |
| AWS 2M GPU Deal | Q3 FY27 - Q2 FY29 | Positive | Multi-year, multi-billion dollar commitment. AWS also adopting full physical AI stack (Omniverse, Cosmos, Isaac, Jetson) for warehouse robotics. |
| OpenAI ~12 GW Commitment | Through 2030 | Positive | Portsmouth site initial 4.25 GW, ~1.5M GPUs per upgrade cycle, 20-year lease. NVIDIA providing credit support up to $105B. |
| NeoCloud Rev-Share Model | Launched Jul 2026 | Positive | Hardware sale + share of rental revenue. NeoClouds expected to exit CY26 at 8 GW installed (up from ~3 GW end CY25). "Potential to drive billions." |
| Groq 3 LPX Production | Q3 FY27 | Positive | 3,400 tokens/sec on Gemma 4 31B (4x nearest alternative). Nebius first adopter. New TAM for ultra-low-latency agentic inference. |
| Goldman Sachs Communacopia | Sep 10, 2026 | Event | Jensen keynote fireside chat. Potential for forward-looking commentary on FY28 demand visibility and Vera Rubin ramp updates. |
| GTC Berlin | Oct 21, 2026 | Event | Product/partnership announcements for European sovereign AI. Could showcase Vera Rubin Ultra roadmap details. |
| Agentic AI Inflection | Ongoing | Positive | 15-100x compute per task vs. prompted AI. Jensen: "Most AI are now agentic." Envisions 400K-4M agents per enterprise. |
| Custom Silicon Competition | Ongoing | Risk | Trainium 4 (AWS), TPU Ironwood (Google), Maia 100 (Microsoft). None offers full-stack fungibility. Watch for any hyperscaler shifting training off NVIDIA. |
| Analyst | Question | Response Summary | Verdict |
|---|---|---|---|
| Moore (Morgan Stanley) | What gives confidence to guide 70% growth a full year out? | Agentic compute intensity, non-hyperscaler TAM growing ~100% YoY, deeper supply chain engagement, 70% reflects secured supply not demand ceiling. | Answered |
| Muse (Cantor) | Inference share trajectory, per-GW TAM, Groq 3 LPX positioning? | 4-phase AI lifecycle fungibility; NVLink 72 architecture spans all workloads; Groq positioned for high-interactivity bolt-on. | Answered |
| Rasgon (Bernstein) | Break down ~$200B FY28 incremental by product. Pricing vs. volume split? | Pivoted to hyperscale vs. ACIE narrative and per-GW framework. No product-level breakdown or pricing quantification. | Deflected |
| Arya (BofA) | Total ecosystem investment size? FY28 cash outlay? Custom chip competition? | Custom chip question well handled. But $500B total unconfirmed and FY28 cash impact sidestepped entirely. | Deflected |
| Arcuri (UBS) | Are open-source models net positive or negative for NVIDIA? | Both positive: nearly all open models run on NVIDIA via CUDA ubiquity; closed models simultaneously thriving. | Answered |
| Reitzes (Melius) | AGI timeline impact on demand? Does recursive self-improvement create a step-function? | Described agentic demand multiplier, then dismissed AGI milestones as "senseless." No modeling framework offered. | Deflected |
| Schneider (Goldman) | Rank-order supply constraints: power, DRAM, foundry, or other? | Fully deflected. "Our entire supply chain is challenged... everybody is running flat out." No ranking provided. | Deflected |
| Rakers (Wells Fargo) | Does $40B/GW scale to $60B+ with Vera Rubin Ultra? | Confirmed directional scaling: $3-5B general purpose, $18B Hopper, $25B Blackwell, $40B Vera Rubin, "higher after that." | Answered |
Deflection pattern: Management was forthcoming on strategic narrative (platform fungibility, demand drivers, competitive positioning) but guarded on anything pinning them to product-level financials, balance sheet commitments, or specific bottleneck identification. The supply constraint deflection is particularly notable given that Colette proactively reset GM lower on memory pricing, yet Jensen would not identify memory as the binding constraint when asked directly.
| Company | Signal | Read-Through |
|---|---|---|
| MU / SK Hynix / Samsung | Positive | HBM memory pricing "exceeded prior expectations and headed even higher." Sustained pricing power through at least FY2028. NVIDIA has deep relationships with all 3 suppliers. |
| AWS (AMZN) | Positive | Deploying 2M additional GPUs Q3 FY27 through Q2 FY29 plus Vera CPUs. Confirms aggressive multi-year commitment and validates AWS CapEx trajectory. |
| CoreWeave / Nebius / Nscale | Positive | Named as NeoCloud success stories NVIDIA helped build. Nebius first to receive Groq 3 LPX. Validates NeoCloud business model and continued NVIDIA ecosystem support. |
| Apollo / BlackRock / Blackstone / Brookfield / GS / KKR | Positive | $500B+ capital for AI infrastructure financing. Confirms AI infra as a major new asset class for institutional capital deployment. |
| ORCL | Positive | Named as lead partner for Vera CPU shipments already underway. Reinforces Oracle's AI infrastructure push. |
| TSMC (implied) | Positive | Entire supply chain "challenged" and "running flat out." Demand far exceeds capacity. Pricing power intact; capacity additions highly valuable. |
| NVDA (own margins) | Negative | The primary negative read-through is for NVIDIA itself: gross margin compression from 75% to 71-72% trough driven by the same memory cost inflation that benefits suppliers. |
Macro backdrop: Cloud industry backlog exceeds $2T. Top-5 hyperscaler CapEx expected ~$800B in CY2026, ~$1.3T in CY2027. Global VC funding in AI exceeded $400B in H1 2026 (surpassing $265B for all of 2025), with ~70% flowing to compute. Nearly 20 AI-native companies now exceed $1B ARR, up from 13 in Q4 FY2026. No negative macro signals were offered; the discussion was entirely about allocation of scarce supply rather than demand risk.