Market Cap
$4.34T
Price: $178.68 | Shares: ~24.3B
Revenue FY2025
$130.5B
+114% Y/Y
Net Income
$72.9B
+145% Y/Y | Margin 55.8%
EPS
$2.94
+147% Y/Y (Diluted)
Gross Margin
75.0%
FY24: 72.7% | Continued Expansion
FCF
$60.9B
FCF Yield ~2.3% | ROIC ~115%
Net Cash
$51.5B
Total Debt $11B only | D/E 0.07
NVIDIA Corporation is an American semiconductor company and the global leader in AI accelerators (GPU / AI Accelerators). The company holds ~87% market share in AI Training & Inference, with a software ecosystem (CUDA) that has created lock-in with more than 5 million developers. Its customers — Microsoft, Google, Amazon, Meta — invest hundreds of billions of dollars in AI infrastructure.
Model: Fabless — chip design, manufactured at TSMC | HQ: Santa Clara, California | Currency: USD | Fiscal Year: ends in January
AI / Data Center
Gaming GPUs
Autonomous Vehicles
CUDA Ecosystem
NVLink / DGX
Annual Revenue
FY2021–FY2025 | $ Billions
Net Income
FY2021–FY2025 | $ Billions
Profit Margins
Gross | Operating | Net
EPS — Earnings Per Share
$ per diluted share
| Metric | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | 4Y Change |
|---|
| Revenue ($B) | 16.7 | 26.9 | 27.0 | 60.9 | 130.5 | +681% |
| Gross Profit ($B) | 10.4 | 17.5 | 15.4 | 44.3 | 97.9 | +841% |
| Gross Margin | 62.3% | 64.9% | 56.9% | 72.7% | 75.0% | +12.7pp |
| Operating Income ($B) | 4.5 | 10.0 | 4.2 | 33.0 | 81.4 | +1,709% |
| Operating Margin | 27.2% | 37.3% | 15.7% | 54.1% | 62.4% | +35.2pp |
| Net Income ($B) | 4.3 | 9.8 | 4.4 | 29.8 | 72.9 | +1,595% |
| EPS ($) | 0.17 | 0.39 | 0.17 | 1.19 | 2.94 | +1,629% |
Free Cash Flow
FCF | $ Billions
Revenue by Segment
FY2025 | % of Revenue
| Balance Sheet Metric | FY2023 | FY2024 | FY2025 |
|---|
| Total Assets | $41.2B | $65.7B | $112.2B |
| Cash & Equivalents | $13.3B | $26.0B | $43.2B |
| Shareholders' Equity | $22.1B | $42.5B | $65.9B |
| Total Debt | $11.0B | $9.7B | $11.0B |
| Net Cash | $2.3B | $16.3B | $51.5B |
| D/E | 0.41 | 0.17 | 0.07 |
| Current Ratio | 3.52 | 4.17 | 4.44 |
| Operating Cash Flow | $5.6B | $28.9B | $64.1B |
| FCF | $3.8B | $27.0B | $60.9B |
| ROIC | ~34% | ~69% | ~115% |
| ROE | ~17% | ~91% | ~119% |
Market Share — AI Accelerators
2025 Estimate
Historical P/E
TTM | FY2021–FY2025
CEO
Jensen Huang
Founder, ~30 years | 3.5% ownership
Net Cash
$51.5B
Total Debt $11B only
Debt / Equity (D/E)
0.07
Negligible — Fabless model
Authorized Buyback
$50B
Share repurchase program
🏰 Moat Assessment: Very Strong & Widening
Customer Switching Cost (CUDA — 5M+ developers)★★★★★
Intangible Assets (Software Ecosystem)★★★★★
Dominant Market Share (~87%)★★★★★
Systems Integration (NVLink, DGX)★★★★☆
Pricing Power (H100→B200: $30K→$70K)★★★★★
Moat is very strong and widening — CUDA lock-in with 5M+ developers. A new entrant would need years and tens of billions of dollars to replicate the ecosystem.
👥 Key Customers
MicrosoftAzure AI — largest customer
Google / AlphabetGCP AI / DeepMind
Amazon / AWSEC2 GPU Instances
Meta PlatformsAI Research / LLMs
⚠ High concentration: 2 customers = 39% of revenue, 4 customers = ~61%
NVIDIA is not an ordinary chip company. It is the bridge between two distinct industry eras —
the era when the GPU served video games, and the era when that same architecture became the foundational
infrastructure of every professional AI system in the world. This transition — and the way the company
shaped it through CUDA as a software layer — is the basis for understanding every other angle of the business.
The center of gravity of the thesis is not the GPU but the ecosystem. Over 5 million developers
have written code on top of CUDA. The world's leading AI libraries (PyTorch, TensorFlow, JAX) are tuned
to NVIDIA's architecture. This software layer is what makes a transition to competing hardware
a multi-year process — not months. That is why the instinctive reaction "a competitor will arrive and ease pricing"
misses the real dynamic: competition is not between chips, it is between ecosystems.
The key variables to monitor are four, in this order: first,
the CapEx pace of hyperscalers (Microsoft, Google, Amazon, Meta) — this is the direct driver
of Data Center revenue, which represents ~88% of the business [10-K FY25]. Second, the product cycle —
each transition (H100 → B200 → Rubin) drives a Gross Margin Uplift; whether the next transition sustains the pace.
Third, internal ASICs from the customers themselves (Google TPU, AWS Trainium, Microsoft Maia) —
the question is not "will they succeed" but "how much will they affect the percent of workload still requiring NVIDIA".
Fourth, export restrictions to China — an external factor that is hard to model.
Where the analysis can go wrong — first error: pricing it like a cyclical semiconductor company.
Historical chip companies (Intel, AMD, Broadcom) trade at a P/E of 15-25x because of
cyclicality concerns. NVIDIA trades today at a TTM P/E of ~36x [yfinance snapshot]
versus a historical average of 60-73x. The naive reading says "still expensive" or "no longer cheap".
A professional reading asks a different question: is the 115% ROIC [FY25] a temporary state
(which will be erased in a downturn) or a feature of the IP/royalty model the company holds — and the answer
changes the entire analytical framework.
Where the analysis can go wrong — second error: ignoring customer concentration.
Two of NVIDIA's customers generate 39% of revenue; four customers = 61% [10-K FY25].
This is a level of concentration that for most companies would create a meaningful discount. At NVIDIA the market
overlooks it, because those customers themselves depend on NVIDIA symmetrically. The problem — that symmetry
could break if one of the hyperscalers announces a material shift to internal ASICs.
This is not a one-year scenario; it is a scenario that could unfold over 3-5 years.
Cash allocation — an atypical structure for a growth company.
NVIDIA holds $51.5B in Net Cash [FY25] against total debt of just $11B (D/E 0.07).
The decision to hold this much cash rather than invest aggressively in additional R&D or M&A (the ARM
acquisition failed in 2022) reflects exceptional management discipline. The open question: whether the $50B
buyback authorized [FY25] is the right tool to release capital, or whether other options for
value creation should be considered.
The competitive context is not what appears in the headlines.
AMD (Instinct MI300X) shows 5-8% market share with a reasonable product line. Google TPU
is mostly used internally. Amazon Trainium is in its early stages. The real long-term competitor is not a company —
it is a pattern: if a large hyperscaler decides that shifting 30-50% of its workload to internal ASIC
is worthwhile, this can occur even without AMD's chart showing any change. This is the dynamic to monitor.
The material risks — not only those competing on the next chip.
The bigger risk is the overall market: if global AI CapEx slows (say,
from +60% Y/Y to +15%), NVIDIA's revenue will be hit even if it holds 100% market share.
This is an industry risk, not a company risk — and its place in the analysis differs.
Clear invalidation conditions: gross margin falling below 65% for two consecutive quarters
[FY25: 75%], or FCF falling below $30B without a one-time explanation [FY25: $60.9B].
What distinguishes a professional analysis of NVIDIA from a "tip".
A tip presents an outcome: "what to do". A professional analysis presents a process: which conditions need
to hold for the thesis to last, which risks would break it, what the impact would be
if a key counterparty (a hyperscaler, say) changes its behavior. These answers
are more important than a price target, because they allow the client to update their understanding
in real time — something a fixed price target does not allow.
The position of this analysis.
NVIDIA is a company that exists in most high-quality global investment portfolios. The analysis here
does not answer the question "hold or not" — it provides the angles through which
that decision is made better. The site does not participate in the decision.
The decision, in every case, belongs to the client.
The difference between surface-level analysis and professional thinking is often found in variables that are not visible at first glance.
The difference between surface-level analysis and professional thinking
often lies in the variables that are not immediately visible.
⚠ Risks
●Internal ASICs from hyperscalers — Google TPU, Amazon Trainium, Microsoft Maia. The largest customers are also potential competitors
●Extreme customer concentration — 2 customers = 39% of revenue, 4 customers = ~61%
●TSMC dependency — geopolitical exposure to Taiwan. NVIDIA does not manufacture itself
●China export restrictions — the U.S. government has already blocked H100/A100. Further restrictions are possible
●AMD competition — Instinct MI300X, 5–8% market share. The only listed competitor
🚀 Items Under Watch
●Accelerating AI demand — Microsoft, Google, Amazon, Meta continue to invest hundreds of billions in AI infrastructure
●Blackwell → Rubin cycle — each product cycle drives a new wave of demand. B200 sold out through mid-2026
●Sovereign AI — governments are building national AI infrastructure — an expanding new market
●Autonomous vehicles — +55% Y/Y — a rapidly growing market
●Expansion into software — NIM, AI Enterprise — growing recurring revenue
❗ Invalidation Conditions — What Would Change the Thesis
!Gross margin falls below 65% for two consecutive quarters
!A key hyperscaler announces a reduction of >30% in NVIDIA purchases
!FCF falls below $30B with no one-time explanation
!A hyperscaler announces success of an internal ASIC with performance comparable to Blackwell
✓ No active deal-breakers — positive FCF, Net Cash $51.5B, D/E 0.07, no going-concern issues
📊 Valuation
P/E TTM35.8x
Historical Average P/E60–73x
Forward P/E (FY2026E)~25x
EV/EBITDA~30x
P/FCF TTM~71x
TTM P/E of 35.8x stands below the company's own historical average (60–73x). This gap can be read in several different ways and is one of the central variables of the analysis.
The framework below describes which conditions need to hold in each scenario — not
a price forecast. The conditions are what a professional analyst monitors over time, and they are what allow
assumptions to be updated as reality changes.
Scenarios are descriptive, not predictive. They outline possible conditions, not expected outcomes.
These scenarios contain no probability assessment, no preferred direction, and no expectation as to which will materialize.
🟢 Bull Scenario
AI CapEx Acceleration + Margin Preservation
Conditions that need to hold:
- Hyperscaler AI CapEx grows above current consensus
- The Blackwell → Rubin cycle drives an additional demand wave without manufacturing constraints
- Gross margin sustains at 73%+ (FY25: 75%)
- Sovereign AI expands as a measurable new market
- No exceptional regulatory event (China export, antitrust)
What this means: under such conditions, the platform thesis holds
and the model continues to generate exceptional FCF.
🔵 Base Scenario
Continued Growth at Slower Pace
Conditions that need to hold:
- AI CapEx continues to grow, but at a more moderate pace
- Data Center remains >85% of revenue
- Gross margin in the 72-75% range
- AMD expands slightly but not materially
- Hyperscaler internal ASICs remain a secondary workload
What this means: the model continues to work, ARR grows, with no
re-rating but also no material erosion.
🔴 Bear Scenario
Structural Pressure on the Platform
Conditions that need to hold:
- Material slowdown in hyperscaler AI investments
- A key hyperscaler announces a material shift to internal ASIC
- Gross margin falls below 65% for two consecutive quarters
- China export restrictions expand
- FCF falls below $30B with no one-time explanation
What this means: under such conditions, even if NVIDIA holds the core,
pressure on the multiple may grow; the resulting state is reverse compounding.
In a professional company analysis, the question is not "is this good", but
"through which angles must the company be examined so that the essence is not missed".
Every analysis at Bakshi Finance passes through six angles. The text below is not an assessment.
It is the mapping of the questions this analysis is meant to answer. The specific answers for NVIDIA
appear in the "How to Think About This Company" section above.
The analysis is based on an internal multi-factor analytical framework
used in professional portfolio management. The framework maps the questions;
the answers appear woven through the analysis above.
What this lens is not: there is no rating, no comparison
between this company and another, and no preference. The same six questions are asked of every company on the site —
the variation is in the answers, not in the tool.
This framework is intended to structure analysis, not to produce an investment conclusion.
📈
Growth
How does the company grow? Does growth come from volume, price, or mix?
Is it stable across cycles? Which catalysts drive it?
💰
Profitability
How do profit margins behave over time? Which portion of gross profit
is retained as operating profit? How much of accounting profit actually converts to free cash flow?
⚖
Leverage
What is the capital structure? What kind of debt (short / long, indexed / not)?
With what flexibility will the company handle a downturn?
🏰
Competitive Position
What protects its revenue from erosion? How long can that protection
realistically hold? What could damage it?
👴
Management Quality
How does management allocate capital? What is its track record on
strategic decisions? How transparent and consistent is the reporting?
🧩
Business Complexity / Risk
Where would a simplistic analysis go wrong? Which management data
requires deeper understanding? What is exposed to regulation, cyclicality, or technological change?
1. Business identity. NVIDIA is a supplier of AI chips and accelerated computing infrastructure, with ~87% dominance in Data Center GPUs [FY25]. FY2025 revenue of $130.5B (+114% Y/Y), of which Data Center represents ~88%.
2. The pillar of the thesis. CUDA ecosystem with 5M+ developers, ROIC ~115%, 75% gross margin [FY25], FCF $60.9B. Net Cash $51.5B with D/E 0.07 — a balance-sheet structure that affords unusual flexibility within the industry.
3. The current critical juncture. The company trades at a TTM P/E of ~35.8x versus a historical average of 60-73x. The reason is not simple: it depends on hyperscaler AI CapEx expectations, on the Blackwell → Rubin cycle, and on the progress of customers' own internal ASICs.
4. What matters for monitoring. The CapEx pace of 4-5 key hyperscalers, gross margin each quarter (invalidation condition: below 65% for two consecutive quarters), and the progress of internal ASICs. These three variables will determine the path over the next 12-24 months.
This summary is not a recommendation. It is a factual list of what the analysis identified. The decision — belongs to the client.
Operating Framework & Regulatory Disclosure
Bakshi Finance operates as a Family Office serving qualified clients only. Mr. Yaron Bakshi was a licensed investment advisor in Israel during 2008–2023. As of the publication date of this document, the firm does not hold any license for investment advice, investment marketing, or portfolio management.
This document is intended for professional research and educational purposes only. Nothing herein constitutes a recommendation to buy, sell, hold, or take any action in any security. Nothing herein is a substitute for advice that takes into account the data and needs of any specific person. Every decision — is the sole responsibility of the investor.
Data is drawn from official sources (10-K FY2025, Q4 FY2025 Earnings, SEC filings). Updated filings may change the picture. Past performance is not indicative of future results.
