Company Overview
NVIDIA in one sentence: NVIDIA designs the graphics processing units (GPUs) that power virtually all of the world's artificial intelligence — from data centers to autonomous vehicles, robotics, and gaming.
1993
Founded (Santa Clara, CA)
Business Segments
| Segment | Revenue Share | YoY Growth | Trend |
|---|
| Data Center (AI GPU) | ~88% | +95% | Exponential acceleration |
| Gaming (GeForce) | ~8% | +9% | Stable |
| Automotive (DRIVE) | ~2% | +30% | Growing |
| Professional Visualization | ~2% | +5% | Stable |
The CUDA moat — the deepest strategic moat in tech: The CUDA software ecosystem has 5 million developers and 500+ optimized libraries. Every hyperscaler, every AI startup, every research lab has millions of lines of code written for CUDA. Switching platforms means rewriting years of work. This is why Google, with its TPUs since 2015, has failed to dent NVIDIA's position. Hardware can be replaced; a software ecosystem cannot.
NVIDIA CUDA
News & Catalysts
EARNINGS ALERT: NVIDIA reports Q4 FY2026 results on Wednesday, February 25 after the close. Guidance $65B, whisper $67B. Historic move expected: plus or minus $10-12.
After the Supreme Court struck down his tariffs, the president raises global duties to 15%. Direct impact on the TSMC to NVIDIA supply chain. Dow futures could open lower on Monday.
IBD
B200/GB200 GPUs are sold out through mid-2026. AI server orders jump from 28,000 to 60,000 units in 2026 according to Morgan Stanley. Backlog of 3.6 million units.
FinancialContent
Meta expands its partnership to deploy millions of AI chips in its data centers, including standalone CPUs. Meta plans $135B in AI CapEx for 2026.
CNBC
The US authorizes H200 with a 25% tariff, but Beijing blocks imports and pushes Huawei Ascend 910C. Estimated loss: $4.5B. The AI OVERWATCH Act aims to ban Blackwell exports to China.
Axios
3nm architecture, HBM4, 288GB, 13 TB/s bandwidth. Performance 2.5x greater than Blackwell. Rubin Ultra in 2027 with 100 PFLOPS and 1TB HBM4E.
Tom's Hardware
Barclays: 70% of AI compute will be inference by 2026. A $50B+ market. OpenAI spent $7B on inference in 2025 (3.5x vs 2024). Jevons paradox: efficiency creates more demand.
PIIE
Billionaire Stanley Druckenmiller is adding to his tech positions despite sector rotation. A positive contrarian signal for the sector.
Fundamentals
$187.1B
Revenue TTM
+62.5% YoY
$112.7B
EBITDA
Margin 60.2%
$99.2B
Net Income TTM
Margin 53.0%
$60.6B
Cash
Net Cash $49.8B
Valuation
| Metric | Value | Interpretation |
|---|
| Forward P/E | 23.6x | Reasonable for this growth (PEG 0.36) |
| Trailing P/E | 45.3x | High but compressing rapidly |
| EV/Revenue | 24.4x | Significant premium — justified by margins |
| EV/EBITDA | 40.5x | Standard for +60% growth |
| Price/Book | 38.8x | Very high (asset-light business) |
| Enterprise Value | $4.56T | Larger than Germany's GDP |
Profitability — Sector Comparison
| Metric | NVDA | AMD | AVGO | Semis Avg |
|---|
| Gross Margin | 70.1% | 49% | 62% | 55% |
| Operating Margin | 63.2% | 7% | 32% | 30% |
| Net Margin | 53.0% | 7% | 25% | 20% |
| ROE | 107.4% | 3.8% | 10.5% | 15% |
| ROA | 53.5% | 2.1% | 4.2% | 8% |
Why the 23.6x Forward P/E is more relevant than the 45.3x Trailing: NVIDIA is growing so fast that its past earnings massively underestimate its future earning power. A 23.6x Forward P/E for +62.5% growth gives an implied PEG of 0.36x — you're paying for growth at just 36 cents on the dollar. For comparison: Apple (PEG ~2.5), Microsoft (PEG ~1.9), S&P 500 (PEG ~1.8).
Earnings History & Q4 Preview
Q4 FY2026 — Wednesday, February 25 after the close
Revenue consensus: $65.6B | Wall Street whisper: $67B | Official guidance: ~$65B
Citigroup (Atif Malik) estimates NVDA could reach $67B — a 3% beat.
Last 4 Quarters — Systematic beats, but surprise is compressing
| Quarter | Actual EPS | Est. EPS | Surprise | Beat |
|---|
| Q4 FY2024 | $0.89 | $0.85 | +5.25% | |
| Q1 FY2025 | $0.81 | $0.75 | +8.03% | |
| Q2 FY2025 | $1.05 | $1.01 | +4.10% | |
| Q3 FY2025 | $1.30 | $1.26 | +3.46% | |
4/4 beats — Caution: the surprise rate is compressing (from +8% to +3.5%). Expectations are catching up to reality.
Annual Trajectory
| Fiscal Year | Revenue | Net Income | Revenue Growth |
|---|
| FY2022 | $26.91B | $11.26B | — |
| FY2023 | $26.97B | $8.37B | +0.2% |
| FY2024 | $60.92B | $32.31B | +126% |
| FY2025 | $130.50B | $74.27B | +114% |
| FY2026E (TTM) | $187.1B | $99.2B | +43% |
5 Things to Watch on February 25:
- Q1 FY2027 guidance — More important than Q4 results. The market expects $70B+. If <$65B, the stock plunges.
- Gross margins — Blackwell is more complex (CoWoS-L, HBM3e). If margin drops below 68%, negative signal.
- Rubin commentary — Timing, pricing, first customers. The Blackwell to Rubin transition is critical.
- China impact — Quantification of lost revenue ($4.5B estimated) and compensation strategy.
- Client CapEx — Meta ($135B), Microsoft, Google, Amazon: their 2026 AI budgets determine NVDA revenue.
Insiders & Institutions
Top Insiders
| Name | Role | Shares | Value |
|---|
| Jensen Huang | CEO & Chairman | 349.3M | $66.3B |
| Mark A. Stevens | Director | 34.2M | $6.5B |
| Tench Coxe | Director | 30.6M | $5.8B |
| Colette Kress | EVP & CFO | 4.8M | $911M |
| Ajay K. Puri | EVP, Worldwide Ops | 4.2M | $798M |
Top Institutions (13D/G)
| Filer | Date | Filing | Change |
|---|
| The Vanguard Group | 12/31/2025 | 13G | +10.83% |
| BlackRock Inc. | 12/31/2023 | 13G | +0.43% |
| FMR LLC (Fidelity) | 11/08/2024 | 13G | -21.93% |
Insider sales = no bearish signal here: Recent sales by Colette Kress (~47K at $175-185) and Ajay Puri (~200K at $180) are automatic selling programs (Rule 10b5-1), scheduled months in advance. Jensen Huang holds 349M shares ($66B) — a massive alignment of interests. Vanguard increased its position by +10.83% in Q4 2025.
Options & Derivatives — Pre-Earnings Positioning
0.58
P/C Vol Ratio (bullish)
ATM Chain — February 23 Expiration
| Strike | Call $ | Call IV | Call OI | Put $ | Put IV | Put OI |
|---|
| $185.00 | $5.14 | 38.4% | 3,889 | $0.43 | 35.6% | 6,547 |
| $187.50 | $3.20 | 33.5% | 9,914 | $0.85 | 32.0% | 4,041 |
| $190.00 | $1.54 | 29.1% | 14,783 | $1.70 | 28.9% | 3,651 |
| $192.50 | $0.52 | 26.7% | 16,302 | $3.21 | 26.5% | 1,326 |
| $195.00 | $0.16 | 27.1% | 13,199 | $5.28 | 33.0% | 874 |
Reading the options positioning: Max pain ($187.50) is just below the price ($189.82) — it's a short-term "magnet." The 30-35% IV prices in a post-earnings move of plus or minus $10-12 (~6%). The P/C volume ratio of 0.58 is bullish: traders are buying more calls than puts. Beware IV Crush: after earnings, IV will collapse from ~35% to ~25%, destroying time value on options. Straddles are expensive — prefer spreads if you're playing options.
Sector & Comparables
| Ticker | Price | MCap | Fwd P/E | RSI | 5D | 90D |
|---|
| NVDA | $189.82 | $4.62T | 23.6x | 54.9 | +3.8% | +10.6% |
| AMD | $200.15 | $326B | 18.8x | 33.7 | -3.5% | -7.5% |
| AVGO | $332.65 | $1.58T | 22.9x | 50.7 | +2.3% | -6.7% |
| TSM | $370.54 | $1.92T | 20.6x | 45.3 | +1.7% | +29.0% |
| MU | $428.17 | $482B | 9.6x | 53.2 | +4.0% | +122% |
| ARM | $125.58 | $133B | 58.2x | 78.2 | +0.2% | -27.0% |
| MRVL | $79.48 | $67B | 22.7x | 51.4 | +1.1% | -11.1% |
| QCOM | $142.88 | $153B | 12.6x | 24.7 | -5.9% | -17.0% |
| INTC | $44.11 | $220B | 44.5x | 42.4 | -5.7% | +18.5% |
Semiconductor ETFs — Near ATH
| ETF | Price | vs 52W High | 5D | 90D |
|---|
| SMH (VanEck) | $415.03 | -1.3% | +1.8% | +22.2% |
| SOXX (iShares) | $359.43 | -1.6% | +1.3% | +26.2% |
Mega-Cap Tech — NVIDIA Customers
| Ticker | Price | MCap | RSI | 90D | Note |
|---|
| MSFT | $397.23 | $2.95T | 30.4 | -22.7% | Oversold — Azure + Copilot |
| GOOGL | $314.98 | $3.81T | 32.1 | +29.0% | TPU v6, Gemini |
| AMZN | $210.11 | $2.26T | 25.3 | -4.5% | Trainium, AWS |
| META | $655.66 | $1.66T | 26.1 | -8.4% | NVDA deal millions of chips |
Remarkable sector divergence: Semi ETFs (SMH, SOXX) are within 1-2% of their ATH, driven by TSM (+29%) and MU (+122%). But Big Tech customers are struggling: MSFT -22.7%, AMZN -4.5%, META -8.4% over 90D — all in oversold territory (RSI 25-32). This means: structural AI demand is intact (semis are rising) but mega-cap valuations are compressing (rotation toward value/defensive). NVIDIA sits between the two — beneficiary of AI demand, but exposed to tech rotation risk.
Trade Idea — Pre/Post Earnings Swing
TP2
$254
Consensus (+38%)
R/R
1:2.0
to TP1 (mid-entry $180)
Trade thesis: NVDA is consolidating in a $185-194 range ahead of earnings on 02/25. The V-shaped bounce from $171 (Feb 5, +11% in 11 sessions) confirms structural demand on pullbacks. The optimal entry is on a pullback to $175-185 (EMA 20/50 zone + support). The stop below $165 protects against a catastrophic miss while leaving room above the $170.74 support.
Reinforcement Signals
- Earnings beat >$67B + Q1 guidance >$70B
- Gross margins maintained >70%
- Positive commentary on Rubin and Blackwell 2H demand
- Breakout above $194.49 (swing high) with volume >200M
Invalidation Signals
- Q4 revenue <$65B or Q1 guidance <$65B
- Gross margin <68% (manufacturing pressure)
- China escalation: total export ban
- Break of $170.74 support with volume >250M
Timing & Sizing:
- Horizon: Swing 4-8 weeks (post-earnings through mid-April)
- Catalysts: Earnings 02/25, NFP 03/06, CPI 03/13, Q1 FY2027 guidance
- Sizing: 4-6% portfolio max. Beta 2.31 means reduced size vs standard position.
- Scaled entry: 50% at $180-185 (pre-earnings) / 50% post-earnings on confirmation ($195+) or dip ($170-175)
- Conservative alternative: SMH (semi ETF, beta ~1.5) for the theme without single-stock risk
Part 2 — The Tribunal
Should you buy NVDA? The prosecution attacks, the defense responds. The judge decides.
Inspired by the Stock Picking series — "Always challenge the AI, always verify the numbers"
The Judge
Court is now in session. We are examining the case of NVIDIA Corporation, ticker NVDA. Market cap: $4.62 trillion. The prosecution will attempt to demonstrate that the stock is overvalued and that buying now is imprudent. The defense will argue otherwise. Each argument must be supported by verifiable data.
Reminder: as the Stock Picking series teaches, AI-generated financial figures must always be verified against at least 2 independent sources.
Valuation: Bubble or Merit?
P
Prosecution (Bear)
$4.62 trillion. That's more than Germany's GDP. P/B of 38.8x, EV/Revenue of 24.4x. NVIDIA is priced for a perfect future where AI demand never falters. Let's remember that
Cisco in 2000 had a P/S of 31x before losing 80%. Semiconductors are historically cyclical — just because demand is "off the charts" today doesn't mean it will be in 3 years. The last time we paid these multiples for a chipmaker, it ended in tears.
D
Defense (Bull)
The Cisco comparison is intellectually dishonest. Cisco had a P/E of
130x with 30% growth. NVIDIA has a Forward P/E of
23.6x with
62.5% growth. The implied PEG is 0.36x — growth isn't in the price, it's
underpriced. Net margin 53%, ROE 107%. Cisco never came close to these figures. And the $60.6B in cash (near-zero net debt) is a cushion Cisco never had. Verify on
StockAnalysis.
P
Prosecution (Bear)
The 0.36x PEG rests on the assumption that +62.5% is sustainable. But growth is
already decelerating: +126% (FY2024) to +114% (FY2025) to an estimated +43% (FY2026). When this growth hits +20% — which is mathematically inevitable at $200B+ in revenue — the Forward P/E will climb back to 35-40x, and the correction will be violent. We're up against the law of large numbers.
D
Defense (Bull)
The deceleration is
mathematical, not structural. Going from $60B to $130B (+114%) is mechanically easier than from $130B to $260B. But the TAM (total addressable market) for AI CapEx grows from $360B (2025) to
$480B (2026), then to an estimated
$7 trillion over the supercycle. Even at "only" +30% growth, NVIDIA generates $240B+ in revenue by FY2028 — at a 20x P/E, that justifies the current market cap without any multiple expansion.
The CUDA Moat: Unassailable or Cracking?
P
Prosecution (Bear)
The "CUDA fortress" is cracking.
Google TPU v6 + the PyTorch plugin = developers can run their code without modification. Meta is negotiating TPU access.
AMD MI350: 288GB HBM3E, ROCm gap reduced from 40-50% to
10-30%.
OpenAI is developing its own ASICs with Broadcom for 2026. Amazon Trainium. Meta MTIA. Every NVIDIA customer is actively working to reduce their dependency. This is the classic sign of a supplier with too much pricing power — its customers route around it.
D
Defense (Bull)
These threats have existed since
2015 — 11 years since Google launched TPUs. Result: NVIDIA has
increased its GPU data center market share from ~70% to >80%. Why? Because CUDA isn't software, it's an
ecosystem of 5 million developers and 500+ libraries. The switching cost is astronomical. ASICs are optimal for inference of a specific model, but NVIDIA wins on
versatility: training + inference + fine-tuning + multi-model. And Rubin arrives Q3 2026 with 50 PFLOPS (2.5x Blackwell). The innovation cycle locks customers in with each generation.
DeepSeek and the End of "More Compute"
P
Prosecution (Bear)
DeepSeek demonstrated that you can train GPT-4-class models with
a fraction of the GPU power. It wiped $500 billion from NVIDIA's market cap in a single day. The message is clear: algorithmic efficiency is advancing faster than expected. If models need 10x fewer GPUs for the same result, compute demand collapses in the medium term.
D
Defense (Bull)
DeepSeek
validated the NVIDIA thesis via the Jevons paradox. Cheaper models to train = more models trained by more players =
more total demand. Barclays: 70% of AI compute will be inference by 2026 — a
$50B+ market. OpenAI spent
$7B on inference in 2025,
3.5x more than 2024. When unit cost drops, total consumption explodes — this is exactly what happened with cloud, storage, and bandwidth. Jensen Huang speaks of a "new scaling law" that accelerates demand.
The China Bomb
P
Prosecution (Bear)
NVIDIA took a
$4.5B hit from export restrictions. The H200, even authorized with a 25% tariff, is
blocked by Chinese customs. The AI OVERWATCH Act wants to ban Blackwell. Beijing is pushing Huawei Ascend 910C as a domestic alternative. This isn't a temporary risk — it's a
permanent market loss. China is building its domestic AI ecosystem, and NVIDIA will never return to 20% of revenue from China.
D
Defense (Bull)
China is a risk, not a definitive bear argument. China revenue has fallen to
~10-12% of the total, and NVIDIA has replaced it with insatiable demand from Western hyperscalers and sovereign states. The backlog stands at
3.6 million units. Even without China, NVIDIA is
sold out. Huawei's Ascend 910C is ~40-50% less performant than the B200. And the February 17 Meta deal (millions of chips) shows that Western customers are more than compensating.
Timing: Buying 3 Days Before Earnings?
P
Prosecution (Bear)
Buying 3 days before the most-watched earnings in the world is
gambling. Historical post-earnings moves: plus or minus 8-12%. Beta 2.31 = risk of -15% to -20% in a few hours. The whisper at $67B is 3% above guidance. An in-line "beat" will be interpreted as a disappointment. Trump just imposed 15% tariffs on Saturday — the market hasn't priced this in yet. The timing is
the worst possible.
D
Defense (Bull)
This is precisely why the trade idea recommends a
scaled entry: 50% before, 50% after. If NVDA beats big + guides $70B+, the gap up to $200-210 is the cost of not being positioned. If it disappoints, the pullback to $170-175 is a proven accumulation zone (support $170.74, 3 touches, held on Feb 5).
Fundamentals have never been stronger. Druckenmiller just put $152M into mega-tech — he doesn't trade on a single quarter. And NVDA has a 100% beat rate over the last 4 quarters.
4 Clients = 50% of Revenue
P
Prosecution (Bear)
Microsoft, Google, Amazon, Meta —
~50% of revenue. And each is developing its own AI chips. When your biggest customer is also your future competitor, that's a fragile business model. Meta signs an NVIDIA deal
and negotiates Google TPU access
and develops MTIA. That's supply chain diversification — not loyalty.
D
Defense (Bull)
If the hyperscalers could replace NVIDIA, they would have done so. Google has had TPUs for
11 years. Result: NVIDIA has taken the market. The Meta-NVIDIA deal of
February 17, 2026 — 5 days ago — for
millions of chips is not the strategy of a customer planning to leave. ASICs are complementary (single-model inference) not substitutive (training, research, fine-tuning). As long as AI innovation accelerates, customers need versatile GPUs.
Money Is Leaving Tech
P
Prosecution (Bear)
The data doesn't lie. Over 5 days:
Consumer Staples leading, Tech lagging. MSFT -22.7% over 90D, META -8.4%, AMZN -4.5%. QCOM RSI 24.7. AMD RSI 33.7. Gold at $5,081 (ATH). Yahoo Finance headline: "Money is leaving tech." This is a classic
late-cycle signal. When Consumer Staples outperform, historically it ends badly for growth.
D
Defense (Bull)
The rotation is hitting
tech clients, not semis. SMH is within -1.3% of its ATH. SOXX at -1.6%. TSM +29% over 90D. MU +122%. AI demand is
structural — it doesn't depend on the economic cycle but on technology adoption. NVIDIA correlates 0.69 with AMD, but only
0.001 with SPY. It's a universe apart. And VIX at 19 + Risk-On regime (composite score 0.45) are not end-of-cycle signals.
Cross-Verification — Data Reliability
The Judge
Procedural pause. As the Stock Picking series teaches:
"AI chatbots can invent financial figures with total confidence." This analysis uses real-time MCP Gateway data (Fintel, ChartExchange, Yahoo Finance, SEC EDGAR, AmericanBulls). The figures are verifiable:
Golden rule: Verify against at least 2 independent sources before any decision.
Setup Validation — The 4 Pillars
D
Defense (Bull)
1) Fundamentals: Revenue +62.5%, net margin 53%, ROE 107%, $60B cash, PEG 0.36x —
✓ VALIDATED
2) Technicals: Price > EMA 20/50/200, bullish MACD, RSI 54.9 neutral, support $170.74 tested and held —
✓ VALIDATED
3) Macro: Risk-On regime (0.45), VIX 19, AI CapEx $480B —
⚠ MIXED (15% tariffs, sector rotation)
4) Catalyst: Q4 earnings on 02/25, Blackwell sold-out, Rubin Q3 2026, Meta deal —
✓ VALIDATED
P
Prosecution (Bear)
The Macro pillar is
red, not "mixed." Gold at $5,081 is not Risk-On. The rotation out of tech is active. Consumer Staples leading is a historical end-of-cycle signal. Trump is imposing 15% global tariffs — the market hasn't priced this in. And the fundamental rule remains:
never buy when information asymmetry is at its maximum — which is 3 days before binary earnings.
Tribunal Verdict
The Judge — Final Verdict
After adversarial examination across 7 rounds of debate, the tribunal renders its verdict:
The tribunal rules that:
- Fundamentals are historically exceptional — No company has ever combined $187B in revenue, 53% net margin and 62% growth. The defense wins on this point (9/10).
- Valuation is stretched but justifiable — 23.6x Forward P/E for this growth is defensible. The prosecution is right on P/B (38.8x), wrong on the Cisco comparison (7/10).
- The CUDA moat holds — 11 years of TPUs haven't dented the position. The ASIC threat is real but not imminent (8/10 defense).
- DeepSeek validates the thesis — The Jevons paradox works. More efficiency = more total demand (7/10 defense).
- China is a permanent loss — The prosecution scores a point. But the backlog compensates (6/10 defense).
- Pre-earnings timing is risky — The prosecution is right. You don't go all-in before a binary event (4/10 defense).
- Sector rotation is a warning signal — But it's hitting clients, not semis. Nuanced (5/10 defense).
Sentence:
- Long-term investors (>1 year): BUY on pullback $175-185. The structural AI thesis is intact. Entry 50% now, 50% post-earnings.
- Swing traders (4-8 weeks): WAIT post-earnings. Buy on dip ($170-175) if moderate miss, or breakout $195+ if strong beat.
- Conservative profiles: SMH (semi ETF) for the AI theme without single-stock risk and the 2.31 beta.
Tribunal confidence score: 7/10 — High fundamental conviction, elevated short-term uncertainty (earnings + tariffs). To be revised after February 25.
Overall Rating
Positive Key Takeaways
- AI monopoly unchallenged for 11 years, CUDA moat
- Growth +62.5% + margin 53% + PEG 0.36x = undervalued
- Locked roadmap Blackwell then Rubin then Feynman
Major Risks
- $4.62T market cap = zero margin for error
- China + Trump tariffs = permanent headwind
- Earnings on 02/25 = binary event of plus or minus 10%
Mindset Tip: NVIDIA is the most covered stock in the world. Risk #1 is neither valuation nor China — it's FOMO. Don't let the fear of missing the train make you ignore your sizing plan and your stops. A beta of 2.31 means every pullback feels like the end of the world. It isn't — but only a disciplined plan protects you from your own emotions.
Sources & Methodology
Financial data: Yahoo Finance, SEC EDGAR (10-Q, 8-K), MCP Gateway (Fintel, ChartExchange) — February 22, 2026
Technicals: TradingView, AmericanBulls, internal calculations (EMA, RSI, MACD, ATR, OBV, VWAP)
Options: CBOE chain via Yahoo Finance — OI and volumes from February 21, 2026
Sentiment: StockTwits (638,915 watchers, score 0.242), Reddit, YouTube
News: Reuters, Bloomberg, CNBC, Axios, Motley Fool, IBD, Yahoo Finance, Tom's Hardware, Seeking Alpha
Research: Morgan Stanley (AI server demand), Barclays (inference shift 70%), Citigroup (Atif Malik, $67B est.), PIIE (DeepSeek), Semiconductor Digest ($7T supercycle)
Competition: AMD MI350 (Seeking Alpha), Google TPU v6 (Tom's Hardware), Broadcom ASICs (CNBC), Huawei Ascend (CSIS)
Disclaimer: This document is provided for informational and educational purposes only. It does not constitute personalized investment advice. Past performance does not guarantee future results. Stock investing involves risk of capital loss. NVIDIA is a highly volatile stock (beta 2.31) — price movements can be extreme, particularly around earnings releases. Consult a licensed financial advisor before making any investment decision. Market Watch is not a registered investment advisor.
MARKET WATCH