NVDA · NVIDIA Corporation — AI Compute Infrastructure Dominance
Research Date: May 11, 2026 Market Cap: ~$5,229B Research Type: Phase 2 Formal — Fact-based draft with cross-verified public sources
Data Credibility & Verification Layer
This report is based on cross-verified public data sources:
| Data Type | Source | Confidence |
|---|---|---|
| 8Q quarterly financials | NVIDIA IR press release + SEC EDGAR | L1–L2 |
| Segment revenue breakdown | NVIDIA Q4 FY2026 earnings PR | L2 |
| FY27 Q1 guidance | NVIDIA IR (published 2026-02-25) | L2 |
| GTC 2026 $1T demand forecast | Jensen Huang keynote (CNBC/Fortune coverage) | L2–L3 |
| Competitive / custom silicon | Multiple sources (CNBC/Motley Fool/tech-insider) | L3 |
Limitations:
- No FactSet/Bloomberg consensus subscription
- SEC 10-K MD&A not directly accessed
- Earnings call transcripts accessed via secondary summaries only
- FY25 Q4 gross margin is a derived estimate (back-calculated from annual figures)
Key Takeaways
Thesis: NVIDIA is the dominant force in global AI compute infrastructure — Data Center revenue accounts for 91% of total, FY26 full-year revenue reached $215.9B (+65% YoY), TTM net income of $120.1B, and exceptionally strong free cash flow. Backlog visibility extending to 2027 across Blackwell and Vera Rubin platforms exceeds $1T. The CUDA software ecosystem, built over 20 years, constitutes a deep competitive moat that rivals (AMD ROCm / Google TPU / AWS Trainium) cannot replicate in the near term.
Coverage Status: Active · Last Updated May 11, 2026 Data Source: SEC EDGAR 10-K/10-Q + NVIDIA Investor Relations + GTC 2026 Materials
Scenario Analysis (Educational Illustration Only):
- Bear Case Revenue Assumption: Forward PE 25x — assumes AI CapEx cycle peak + full China export restriction
- Base Case Revenue Assumption: Forward PE 31x — FY27 consensus realization at $370B revenue / EPS $8.34
- Bull Case Revenue Assumption: Forward PE 42x — inference inflection exceeds expectations + Vera Rubin early ramp
Note: These are arithmetic scenarios derived from publicly disclosed guidance ranges and consensus estimates, not price forecasts or investment recommendations.
Key Risks:
- Customer custom silicon (Google TPU v8 / AWS Trainium 3–4 / Microsoft Maia 200 / Meta MTIA) — inference market share may decline from 90%+ to 20–30% by 2028 (analyst projections)
- China export controls: Market share dropped from 95% to ~0%; FY26 Q1 included a $4.5B inventory write-down
- Valuation concentration risk: $5.2T market cap = world's largest; PE 44x implies perpetual growth assumptions
- Gross margin volatility: FY26 full-year 71.1% (vs FY25 75%), reflecting Blackwell/Rubin generational transition costs
This section is for educational purposes only. See full Disclaimer.
1. Company Fundamentals
| Dimension | Data | Source |
|---|---|---|
| Company | NVIDIA Corporation | Polygon ticker_details |
| SIC Code | 3674 — SEMICONDUCTORS | Polygon ticker_details |
| Employees | 42,000 | Polygon ticker_details |
| Primary Exchange | NASDAQ (XNAS) | Polygon ticker_details |
| Fiscal Year | Ends late January (FY26 = 2026-01-25) | NVIDIA IR |
| Beta vs SPY | 1.82 (18-month window) | Calculated |
| Volatility | 46% | Calculated |
| Current 10Y Treasury | 4.41% (2026-05-07) | Polygon treasury_yields |
Business Segments (Q4 FY26, ended 2026-01-25)
| Segment | Q4 FY26 Revenue | YoY | Share | Description |
|---|---|---|---|---|
| Data Center | $62.3B | +75% | 91% | Core AI GPU business |
| ↳ Compute | $51.3B | +58% | 75% | GPU + DGX systems |
| ↳ Networking | $11.0B | — | 16% | InfiniBand / Spectrum-X |
| Gaming | $3.7B | +47% | 5% | GeForce RTX series |
| Professional Visualization | $1.3B | +159% | 2% | Quadro / Omniverse |
| Automotive | $0.6B | +6% | 1% | DRIVE / autonomous driving |
[Sources: NVIDIA Q4 FY2026 Earnings (NVIDIA Newsroom) · ServeTheHome Q4 FY2026 Analysis]
FY26 Full-Year Results
| Metric | FY26 Full Year | YoY | Notes |
|---|---|---|---|
| Total Revenue | $215.9B | +65% | All-time high |
| Data Center | $193.7B | +73% | 89.7% of total revenue |
| GAAP Gross Margin | 71.1% | vs FY25 75% | Blackwell ramp transition costs |
| GAAP Net Income | $120.1B | — | Net margin 55.6% |
| Operating Cash Flow | $102.7B | — | — |
| Shareholder Returns | $41.1B | — | Buybacks + dividends |
[Source: NVIDIA FY2026 Annual Results]
FY27 Q1 Official Guidance (issued with Q4 FY26 earnings)
| Metric | FY27 Q1 Guidance | Notes |
|---|---|---|
| Revenue | $78.0B ± 2% | ~+15% QoQ, +77% YoY (vs Q1 FY26 $44.1B) |
| GAAP Gross Margin | ~71% | Vera Rubin ramp impact expected to be manageable |
| China Revenue | $0 (excluded from guidance) | Conservative assumption under export controls |
[Sources: NVIDIA Q4 FY26 Earnings · Fortune Coverage]
2. Supply Chain Position
Upstream
| Supplier | Relationship | Risk |
|---|---|---|
| TSMC | Sole foundry (N4/N5 Blackwell → N3 Rubin) | Geopolitical risk across the Taiwan Strait; capacity competition |
| SK Hynix / Samsung | HBM3e (Blackwell) / HBM4 (Rubin) | HBM4 mass production not until 2026H2; potential bottleneck |
| Broadcom | Networking silicon (partial NVLink Switch) | Dual-sourced |
| CoWoS Packaging (TSMC) | 2.5D advanced packaging | Capacity-constrained; 18-month expansion cycle |
Product Roadmap:
| Generation | GPU | Process | HBM | Production | Key Features |
|---|---|---|---|---|---|
| Blackwell | B200/B300 | N4 | HBM3e (192GB) | In production | NVL72 liquid-cooled; sold out through 2026H1 |
| Vera Rubin | R100 | N3 | HBM4 (288GB) | 2026H2 | 336B transistors, 5x Blackwell, 50 PFLOPS FP4 |
| Vera Rubin Ultra | — | N3 | HBM4e | 2027H1 | Next-generation upgrade |
[Sources: CNBC GTC 2026 Coverage · NVIDIA Vera Rubin Platform · Tech-Insider Rubin Analysis]
Downstream (4 Customer Types)
| Customer Type | Key Names | Revenue Share Trend | Pricing Power |
|---|---|---|---|
| Hyperscaler | Microsoft, Google, Meta, Amazon | ~60–70% of Data Center revenue | Low (NVIDIA holds pricing power) |
| Neocloud | CoreWeave, Lambda, Nebius, Nscale | Fastest-growing segment | Medium |
| Sovereign AI | G7 governments, Middle East sovereign funds | Highest margin | High |
| Enterprise | Fortune 500 | Growing | Medium |
Key Customer AI CapEx Plans (2026):
| Customer | 2026 CapEx Plan | YoY |
|---|---|---|
| Amazon | $200B | ~2x |
| Alphabet (Google) | $175–185B | ~2x |
| Microsoft | $120B+ | ~2x |
| Meta | $115–135B | ~2x |
| Big Four Total | ~$700B | +77% |
Approximately 75% of hyperscaler CapEx (~$450B) is allocated to AI-related infrastructure.
[Sources: Fortune AI CapEx Report · CNBC Report · Tom's Hardware]
3. Sector Cycle Assessment
AI Compute Cycle: Mid-Acceleration Phase
| Signal | FY26/2026 Data | Assessment |
|---|---|---|
| Hyperscaler AI CapEx | $700B (2026 plan, +77% YoY) | Accelerating, no deceleration signals |
| NVIDIA Data Center Revenue | $193.7B FY26 (+73%) | Primary growth engine intact |
| Blackwell Capacity | Sold out through 2026H1 | Demand exceeds supply |
| Vera Rubin Orders | GTC 2026: $1T backlog visibility announced | Demand far exceeds capacity |
| FY27 Q1 Guidance | $78B (+77% YoY) | Growth rate sustained |
| Inference Inflection | Jensen: "the inference inflection has arrived" | New addressable market opening |
Core Assessment: AI compute demand is transitioning from training-driven to inference-driven, expanding rather than contracting the total addressable market. Agentic AI requires continuous 24/7 inference capability, opening a TAM potentially 3–5x larger than training alone.
[Sources: Fortune — Jensen Huang $1T · Quartz — Inference Inflection]
Counter-Signals (Cycle Risk Factors)
- ⚠️ Hyperscaler FCF pressure: Amazon may generate negative FCF of $17–28B in 2026 (Morgan Stanley/BofA estimates)
- ⚠️ Gross margin declined from FY25 75% to FY26 71.1% (generational transition costs)
- ⚠️ China market share fell from 95% to ~0% (export controls)
- No hyperscaler has signaled CapEx reductions as of this writing
- No signs of an "AI winter" in current data
4. Eight-Quarter Financial Trends
| FQ | Period End | Rev (B) | GM% | OI (B) | OM% | NI (B) | EPS | Notes |
|---|---|---|---|---|---|---|---|---|
| Q1 FY25 | 2024-04-28 | $26.0 | 78.4% | $16.9 | 65.0% | $14.9 | $0.61 | Hopper at peak |
| Q2 FY25 | 2024-07-28 | $30.0 | 75.1% | $18.6 | 62.0% | $16.6 | $0.68 | DC $26.3B |
| Q3 FY25 | 2024-10-27 | $35.1 | 75.0% | $21.9 | 62.4% | $19.3 | $0.78 | Then-record high |
| Q4 FY25 | 2025-01-26 | $39.3 | ~72%¹ | $24.3² | ~62% | $22.1 | $0.89 | FY25 close |
| Q1 FY26 | 2025-04-27 | $44.1 | 69.8% | $27.4 | 62.1% | $24.1 | $0.98 | Blackwell ramp begins |
| Q2 FY26 | 2025-07-27 | $49.0 | 69.0% | $29.8 | 60.8% | $26.5 | $1.08 | GM% trough |
| Q3 FY26 | 2025-10-26 | $56.3 | 70.7% | $35.3 | 62.7% | $30.0 | $1.22 | Margin recovery |
| Q4 FY26 | 2026-01-25 | $66.6 | 72.2% | $37.9² | 56.9% | $29.2 | $1.19 | Quarterly all-time high |
¹ FY25 Q4 GM% derived from FY25 full-year 75% minus Q1–Q3 weighted average ² DERIVED (full year minus first three quarters)
Key Observations
- Q4 FY26 revenue of $66.6B = all-time quarterly record, +18% QoQ, +69% YoY
- Gross margin V-shaped recovery: Q2 FY26 69.0% (trough) → Q4 FY26 72.2% (+3.2pp). Initial Blackwell line adjustment costs have been absorbed
- Full-year net margin of 55.6%: Exceptional for the semiconductor industry (compare TSMC ~39%, AMD ~7%)
- Revenue growth trajectory: Q1 FY25 $26B → Q4 FY26 $66.6B = 156% growth over 7 quarters
- Q4 FY26 operating margin dipped to 56.9% (vs Q3 62.7%), potentially related to accelerated Vera Rubin R&D and one-time charges
FY26 Full-Year Summary
| Metric | FY26 | FY25 | YoY |
|---|---|---|---|
| Revenue | $215.9B | $130.5B | +65% |
| GAAP Gross Margin | 71.1% | 75.0% | -3.9pp |
| Net Income | $120.1B | $72.9B | +65% |
| Operating Cash Flow | $102.7B | — | — |
| Diluted EPS | $4.90 | $2.94 | +67% |
Shareholder Returns (FY26)
| Item | FY26 | Source |
|---|---|---|
| Total Returns (buybacks + dividends) | $41.1B | NVIDIA IR |
| Remaining Buyback Authorization | $58.5B | Q4 FY26 earnings |
[Sources: NVIDIA FY2026 Annual Results · CNBC Shareholder Returns Report]
5. Balance Sheet
| Metric | Q4 FY26 (2026-01-25) | Notes |
|---|---|---|
| Total Assets | $206.8B | — |
| Total Liabilities | $49.5B | — |
| Total Debt | $8.5B | Very low leverage |
| Shareholders' Equity | $157.3B | Strong |
| Debt-to-Equity | 5.4% | Near-zero leverage |
| Cash & Short-Term Investments | Exceeds total debt | Net cash position |
Key Takeaways
- Net cash position: NVIDIA's cash holdings far exceed total debt, in stark contrast to companies like DELL (which has negative shareholders' equity). This is a nearly unleveraged balance sheet.
- $157.3B shareholders' equity: Provides substantial capital cushion even in extreme downside scenarios.
- Total assets of $206.8B: Grew rapidly during FY26, driven by retained earnings and inventory build-up (Vera Rubin pre-production).
- Context: In late 2022 (before the AI wave), NVIDIA's total assets were ~$41B — a 5x increase in three years.
[Sources: StockAnalysis NVDA Balance Sheet · MacroTrends NVDA Balance Sheet]
6. Data Quality Summary
| Dimension | Status | Notes |
|---|---|---|
| 8Q Continuity | Complete | Q1 FY25 → Q4 FY26, no gaps |
| Q4 Derived Values (income) | 2 items (Q4 FY25 OI, Q4 FY26 OI) | FY minus Q1–Q3; margin of error is manageable |
| Segment Data Source | NVIDIA IR press release | L2 (not in statutory filings) |
| FY27 Q1 Guidance | L2 (NVIDIA IR original text) | $78B ± 2% |
| GTC $1T Forecast | L2–L3 (CNBC/Fortune coverage) | Jensen Huang keynote |
| Analyst Consensus | L3 (MarketBeat/StockAnalysis) | FY27 revenue $370B / EPS $8.34 |
| Blocking Issues | 0 | Core data cross-verification passed |
Overall Data Quality Assessment: NVIDIA's information disclosure quality is exceptionally high. Quarterly earnings press releases include complete segment revenue, gross margin, and OCF data. The annual GTC conference provides forward-looking guidance. The primary limitation is the absence of precise FactSet/Bloomberg consensus figures.
7. Peer Comparison
| Dimension | NVIDIA | AMD | Intel | Broadcom | TSMC |
|---|---|---|---|---|---|
| Market Cap | $5,229B | ~$240B | ~$130B | ~$1,100B | ~$1,200B |
| TTM Revenue | $215.9B | ~$30B | ~$50B | ~$55B | ~$95B |
| Gross Margin | 71.1% | ~50% | ~42% | ~67% | ~57% |
| Net Margin | 55.6% | ~7% | ~1% | ~30% | ~39% |
| AI GPU Market Share | ~86% | ~5–7% | <1% | N/A | N/A (foundry) |
| PE (TTM) | ~44x | ~60x+ | 100x+ | ~45x | ~28x |
| 1Y Return | +83.4% | ~+15% | ~-20% | ~+40% | ~+30% |
Key Differentials
| Dimension | Interpretation |
|---|---|
| Net margin 55.6% | Industry-leading — comparable only to payment networks like Visa/Mastercard. AMD's 7% net margin is 1/8th of NVIDIA's |
| AI GPU share ~86% | Near-monopoly. AMD 5–7% + Google/Amazon custom ~10% = NVIDIA still dominates the vast majority |
| PE 44x vs 65% growth | PEG ~0.68. For a company growing at this rate, sub-1.0 PEG is within reasonable range |
| Market cap $5.2T | Top 3 globally; enters territory where "too big to grow" questions arise, yet revenue growth still exceeds 65% |
[Sources: Multiple cross-referenced (MarketBeat/StockAnalysis/Yahoo Finance)]
8. Valuation Framework
Current Valuation Snapshot
| Metric | Value |
|---|---|
| Diluted Shares | ~24.3B |
| TTM Revenue | $215.9B (FY26 full year) |
| TTM Net Income | $120.1B |
| TTM OCF | $102.7B |
| TTM EPS | $4.90 |
| Total Debt | $8.5B |
| Net Debt | Negative (net cash) |
| PE (TTM) | 43.9x |
| PS (TTM) | 24.2x |
| EV/OCF | ~50.9x |
| FCF Yield | ~1.96% |
Forward Valuation (Based on FY27 Consensus)
| Metric | Value |
|---|---|
| FY27 Consensus Revenue | $370B (+71.4% YoY) |
| FY27 Consensus EPS | $8.34 |
| Forward PE | 25.8x |
| Forward PS | 14.1x |
| PEG (fwd PE / growth%) | 0.36 |
Valuation Methods Comparison
| Method | Result | Interpretation |
|---|---|---|
| TTM PE (43.9x) | Appears elevated | But with TTM growth of 65%, PEG of 0.68 is within reasonable range |
| Forward PE (25.8x) | Moderate | Against FY27 expected 71% growth, PEG 0.36 suggests room relative to growth |
| FCF Yield (1.96%) | Well below risk-free rate | vs 10Y Treasury 4.41%; FCF yield risk premium is -245 bp |
| PS TTM (24.2x) | High in absolute terms | But with 55.6% net margin, PS 24x is margin-adjusted equivalent of PE 44x |
Valuation Context
NVIDIA's valuation reflects a "growth pricing" regime — the market is pricing in the following core assumptions:
- FY27 revenue of $370B (+71%) will be achieved
- Gross margin recovers from 71% toward 73–75%
- Inference market explosion expands TAM from ~$300B (training) to $1T+
FCF Yield of 1.96% vs 10Y at 4.41% = negative risk premium. This means investors are paying for future growth, not current cash flow. If FY27 growth falls below 50% (vs the 71% consensus), valuations could compress sharply.
9. Growth Catalysts
Catalyst 1: $1T Backlog Visibility (Blackwell + Vera Rubin through 2027)
- Source: CNBC GTC 2026 · Fortune
- Jensen Huang announced at GTC 2026 that backlog visibility doubled from $500B the prior year. Revenue visibility for FY27–FY28 is exceptionally high.
Catalyst 2: Inference Inflection = TAM Expansion
- Source: Quartz — Inference Inflection
- Agentic AI requires continuous 24/7 inference, unlike one-time training workloads. Inference TAM could be 3–5x training TAM. NVIDIA's TensorRT and Triton inference optimization tools lead the market.
Catalyst 3: Vera Rubin 2026H2 Mass Production
- Source: NVIDIA Vera Rubin Platform · Tech-Insider
- R100 GPU: 336B transistors + 288GB HBM4 + 50 PFLOPS FP4 = 5x Blackwell performance. Per-system NVL72 ASP estimated at $5–8M.
Catalyst 4: Hyperscaler CapEx Reaching $700B (2026, +77% YoY)
- Source: Fortune · Tom's Hardware
- Amazon $200B, Google $175–185B, Microsoft $120B+, Meta $115–135B. Even if NVIDIA captures only 30% of AI CapEx (~$210B), this exceeds FY26 Data Center revenue of $194B.
Catalyst 5: 20-Year CUDA Ecosystem Moat
- Source: Stratum Review CUDA Analysis · Alphastreet CUDA Lock-in
- Millions of developers + thousands of optimized libraries (cuDNN/TensorRT/NCCL/RAPIDS). Migrating to ROCm requires 3–6 months of engineering effort.
Catalyst 6: Gross Margin V-Shaped Recovery
- Source: NVIDIA Q4 FY26 Earnings
- Q2 FY26 69.0% (trough) → Q4 FY26 72.2% (+3.2pp); Q4 non-GAAP reached 75.2%. Management guides FY27 back to mid-70s. Each 1pp of gross margin ≈ ~$3.7B additional profit (based on FY27 $370B revenue).
Catalyst 7: May 20, 2026 Earnings (FY27 Q1) — Near-Term Inflection Point
- Source: Motley Fool Forecast · MarketBeat
- Consensus expects $78B revenue. NVIDIA has beaten consensus in each of the past 8 quarters.
10. Risk Factors & Counter-Evidence
Risk 1: Custom Silicon Eroding Inference Market (Primary Medium-Term Risk)
- Data: Google TPU v8 (+80% inference efficiency/dollar vs v7), AWS Trainium 3–4 (triple-digit growth, sold out), Microsoft Maia 200, Meta MTIA
- Sources: Nerd Level Tech Custom Chip Analysis · Motley Fool
- Analysts project NVIDIA's inference market share may decline from 90%+ to 20–30% by 2028
- Monitor: Quarterly hyperscaler custom chip deployment volumes + NVIDIA Data Center growth rate deceleration
Risk 2: China Export Controls (Realized Loss)
- Data: China market share dropped from 95% to ~0%. FY26 Q1 included $4.5B inventory write-down.
- Sources: Computer Weekly $4.5B Write-down · Tom's Hardware Jensen Comments
- If controls extend to Middle East/Southeast Asian third countries, a new downside risk emerges
Risk 3: Structural Gross Margin Decline
- Data: FY25 75.0% → FY26 71.1% (-3.9pp). While Q4 recovered to 72.2%, Vera Rubin ramp may pressure margins again
- Trigger: FY27 full-year GAAP GM below 68%
- Monitor: Quarterly gross margin trends; HBM4 costs (288GB/GPU, significantly higher than HBM3e)
Risk 4: Valuation Concentration at $5.2T
- Data: NVIDIA market cap is top 3 globally. FCF Yield of 1.96% is well below the 4.41% risk-free rate
- Any growth deceleration below expectations could trigger sharp multiple compression
Risk 5: HBM4 Supply Bottleneck
- Data: Each Vera Rubin GPU requires 288GB HBM4. SK Hynix + Samsung mass production begins 2026H2. Estimated 2026 capacity ceiling: 200,000–300,000 Rubin GPUs
- Source: Introl Blog Rubin Capacity Analysis
- Trigger: Low HBM4 yields → Rubin shipment delays → FY27 H2 revenue miss
Risk 6: TSMC Capacity & Geopolitical Risk
- NVIDIA is 100% dependent on TSMC for fabrication. N3 capacity is shared with Apple and AMD
- Extreme scenario: Taiwan Strait escalation → TSMC production halt (low probability but non-zero)
11. Upcoming Events & Monitoring Calendar
| Timeframe | Event | Key Focus |
|---|---|---|
| 2026-05-20 | NVDA FY27 Q1 Earnings | Revenue vs $78B consensus / gross margin trend / Q2 guidance (consensus $86.6B) / China revenue |
| 2026-08 (est.) | NVDA FY27 Q2 Earnings | Initial Vera Rubin shipment feedback / gross margin / Blackwell backlog |
| 2026-11 (est.) | NVDA FY27 Q3 Earnings | Vera Rubin production ramp / FY27 full-year guidance / inference revenue share |
| 2027-02 (est.) | NVDA FY27 Q4 Earnings | FY27 actuals vs $370B consensus / FY28 guidance (critical inflection) |
Interim Events
| Timeframe | Event | Key Focus |
|---|---|---|
| 2026-06 | COMPUTEX 2026 | Vera Rubin product details / new customer announcements |
| 2026 H2 | Vera Rubin First Shipments | Actual ASP / yield rates / customer feedback |
| Quarterly | Hyperscaler Earnings | AI CapEx maintenance vs cuts |
12. CUDA Software Moat — Deep Dive
Why CUDA Is a Deeper Moat Than Hardware
| Dimension | CUDA Ecosystem | AMD ROCm | Google TPU (JAX) |
|---|---|---|---|
| Ecosystem Age | 20 years (since 2006) | ~8 years | ~10 years |
| Developer Base | Millions | Tens of thousands | Primarily Google-internal |
| Optimized Libraries | 5,000+ | ~300 | N/A |
| PyTorch Native Support | First-class | Second-class (requires ROCm compilation) | JAX-primary (not PyTorch) |
| Migration Cost | — | 3–6 months of engineering | Requires code rewrite |
| Inference Optimization | TensorRT + Triton Server | MIGraphX (immature) | TPU Inference (GCP-only) |
Competitive Progress
- AMD ROCm: As of 2026, PyTorch/JAX have native support; ZLUDA project enables CUDA code compatibility. But production stability remains far behind CUDA.
- Google TPU v8: +80% inference efficiency/dollar (vs v7), but available only on GCP — hardware cannot be purchased.
- AWS Trainium 3–4: Sold out, triple-digit growth. Available only on AWS.
- Meta MTIA: Internal use only, not sold externally.
Assessment: In the near term (2026–2027), the CUDA moat remains unbreakable. In the medium term (2028–2030), the inference market may see 10–30% erosion from custom silicon, but NVIDIA will likely continue to dominate training.
[Sources: GPUnex Blog · Thundercompute ROCm vs CUDA · Rayhan Press CUDA Moat]
13. China Export Controls — Impact Analysis
Timeline
| Date | Event | Impact |
|---|---|---|
| 2022-10 | First U.S. AI chip export controls | A100/H100 banned for sale to China |
| 2023-10 | Second round tightening | Downgraded H800/A800 also banned |
| 2024 | H20 (further downgraded) still exportable | NVIDIA China revenue ~15% |
| 2025-04 | H20 exports banned + $4.5B inventory write-down | NVIDIA China revenue approaches zero |
| 2026-01 | DOC partially eases restrictions (H200 exportable) | Jensen announces Chinese H200 orders received |
| 2026 (ongoing) | 15% tariff on AI chip exports to China | Reduced economic viability |
Financial Impact Quantification
| Scenario | Est. China Revenue | % of Total | Notes |
|---|---|---|---|
| FY24 Peak | ~$17B | ~26% | Pre-controls |
| FY25 | ~$8–10B | ~6–8% | H800/H20 era |
| FY26 | ~$0 | ~0% | $4.5B write-down + full ban |
| FY27 Guidance | $0 (excluded) | 0% | Management's conservative assumption |
Impact Assessment: The ~$17B/year China revenue loss has been fully offset by the global AI CapEx explosion (FY26 Data Center revenue of $194B set a new record even with zero China contribution). However, if controls expand to Middle Eastern and Southeast Asian third countries, a new downside risk emerges.
[Sources: Tom's Hardware Jensen Comments · Computer Weekly $4.5B Write-down]
Sources & Citations
Official Sources (L1–L2)
- NVIDIA Q4 FY2026 Earnings Press Release (2026-02-25) — NVIDIA Newsroom
- NVIDIA Vera Rubin Platform Announcement — NVIDIA Newsroom
- Fortune: Jensen Huang $1T AI Infrastructure — Fortune
- CNBC GTC 2026 Coverage — CNBC
- Fortune: Big Tech $700B AI CapEx — Fortune
Cross-Referenced Sources (L2–L3)
- ServeTheHome Q4 FY2026 Analysis — ServeTheHome
- Tom's Hardware: Jensen on China — Tom's Hardware
- Computer Weekly: $4.5B Write-down — Computer Weekly
- Tech-Insider: Vera Rubin Analysis — Tech-Insider
- Nerd Level Tech: Custom Chip Race — Nerd Level Tech
- Stratum Review: CUDA Moat — Stratum Review
- Introl Blog: Rubin Capacity — Introl