Companies / Technology / NVIDIA Corp.
NVIDIA Corp.
NASDAQ: NVDA · Semiconductors · United States · Fiscal year ends late January
NVIDIA designs accelerated-computing platforms — GPUs, networking silicon and the CUDA software layer that binds them. Over six years the company has moved from selling graphics cards to gamers to supplying the substrate on which most large AI models are trained.
Analysis last updated 12 August 2026 · Based on the FY2025 Form 10-K, filed 26 February 2026
Revenue FY25
$130.5B
+114% YoY
Rev. CAGR 5y
+64.1%
FY20 to FY25
Operating margin
62.4%
+8.3 pts
Free cash flow
$60.9B
47% of revenue
ROIC
118%
Excl. goodwill
Net cash
$38.0B
No net debt
The business in one page
Six years ago NVIDIA was a gaming company with a promising side business. In FY2020, Gaming produced $5.5 billion of $10.9 billion in revenue and Data Center $3.0 billion. In FY2025 Data Center produced $115.2 billion of $130.5 billion — 88% of the company — and Gaming, though larger in absolute terms than it has ever been, is now a rounding error in the investment case.
What makes the position unusual is not the hardware. It is that fifteen years of CUDA has made the NVIDIA instruction set the default target for machine-learning frameworks. A competitor selling a faster chip must also persuade customers to rewrite and re-validate their software. That is the moat, and it is a software moat sitting inside a semiconductor company.
The obvious risk is on the demand side, and it is concentrated: four customers accounted for roughly 46% of FY2025 revenue, and those customers are simultaneously funding internal silicon programmes.
Revenue
Fiscal years, $ billions
Source: NVIDIA Form 10-K, FY2020-FY2025.
Free cash flow
Operating cash flow less capex, $ billions
Source: consolidated statements of cash flows.
What we think matters
The switching cost is written in software, not silicon
CUDA, cuDNN and the framework integrations built on them mean a competitor has to win on total migration cost, not on peak FLOPS.
Negative working capital funds the growth
Customer prepayments and supplier terms mean revenue growth releases cash rather than consuming it – visible in the cash flow statement, not the P&L.
Four customers, 46% of revenue
Concentration is disclosed in the 10-K but not named. Each of those customers has an internal accelerator programme in flight.
Inventory and supply commitments
Purchase obligations rose faster than revenue in FY2025. In a demand pause, that becomes a margin problem before it becomes a revenue problem.
How NVIDIA makes money
NVIDIA sells silicon and the systems built around it. Revenue is recognised on shipment, which means the income statement is a shipment record, not a demand record — an important distinction in a period when customers are ordering ahead of deployment capacity.
Revenue by segment
FY2025, share of total revenue
Data Center
GPUs, networking and full rack-scale systems sold to hyperscalers, neoclouds, enterprises and sovereign programmes. Increasingly sold as a system rather than a chip, which raises revenue per unit and lowers the customer freedom to mix suppliers.
Gaming
GeForce discrete GPUs sold through add-in-board partners and laptop OEMs. Structurally profitable, seasonally cyclical, and now competing for the same wafer supply as Data Center.
Professional Visualization and Automotive
Workstation graphics and automotive compute platforms. Together under 3% of revenue; relevant as option value, not as a driver of the current investment case.
Revenue drivers
|
Driver
|
What it depends on
|
Direction FY25
|
Disclosed in
|
|---|---|---|---|
|
AI capital expenditure
|
Hyperscaler and neocloud budgets
|
Strongly positive
|
Customer filings
|
|
Foundry and packaging supply
|
TSMC CoWoS and HBM capacity
|
Easing
|
10-K, supply commitments
|
|
Average system price
|
Mix shift from chips to racks
|
Positive
|
CFO commentary
|
|
Customer self-design
|
Internal accelerator programmes
|
Negative
|
Not disclosed
|
|
Export controls
|
Rules on advanced compute sales
|
Negative
|
10-K, Item 1A
|
|
Gaming replacement cycle
|
Installed base age, console competition
|
Neutral
|
Segment results
|
Financial analysis
Six fiscal years, presented on a consistent basis. Where we adjust a reported figure, the adjustment is stated beneath the table.
Operating income
GAAP, $ billions
Operating margin and ROIC
Percent - both series share one axis
Income statement
|
$ billions
|
FY20
|
FY21
|
FY22
|
FY23
|
FY24
|
FY25
|
|---|---|---|---|---|---|---|
|
Revenue
|
10.9
|
16.7
|
26.9
|
27.0
|
60.9
|
130.5
|
|
Gross profit
|
6.8
|
10.4
|
17.5
|
15.4
|
44.3
|
97.9
|
|
Gross margin
|
62.0%
|
62.3%
|
64.9%
|
56.9%
|
72.7%
|
75.0%
|
|
R&D expense
|
2.8
|
3.9
|
5.3
|
7.3
|
8.7
|
12.9
|
|
Operating income
|
2.8
|
4.5
|
10.0
|
4.2
|
33.0
|
81.5
|
|
Operating margin
|
26.1%
|
27.2%
|
37.3%
|
15.7%
|
54.1%
|
62.4%
|
|
Net income
|
2.8
|
4.3
|
9.8
|
4.4
|
29.8
|
72.9
|
FY2023 reflects a $1.4bn inventory charge and the terminated Arm acquisition cost. We have not adjusted it out: the charge is a real cost of the demand mis-forecast.
Cash flow and capital allocation
|
$ billions
|
FY23
|
FY24
|
FY25
|
|---|---|---|---|
|
Operating cash flow
|
5.6
|
28.1
|
64.1
|
|
Capital expenditure
|
(1.8)
|
(1.1)
|
(3.2)
|
|
Free cash flow
|
3.8
|
27.0
|
60.9
|
|
Share repurchases
|
(10.0)
|
(9.7)
|
(33.7)
|
|
Dividends
|
(0.4)
|
(0.4)
|
(0.8)
|
|
Acquisitions
|
(0.0)
|
(0.1)
|
(1.0)
|
Buybacks in FY2025 absorbed 55% of free cash flow and roughly offset stock-based compensation dilution of 1.1% of shares.
Balance sheet
Net cash position
Cash and investments less total debt, $ billions
Cash and investments
$46.6B
Total debt
$8.6B
Inventory days
98
Valuation framework
We do not publish price targets. What follows is the arithmetic that connects a set of assumptions to a range of values, so you can change the assumptions and see what happens.
How to read this section. Every number below is a consequence of an assumption stated next to it. None of them is a forecast, and none should be treated as a recommendation to buy or sell anything.
1. Where the multiple sits historically
|
Multiple
|
Current
|
5-year median
|
5-year low
|
5-year high
|
|---|---|---|---|---|
|
EV / Sales
|
18.4x
|
21.7x
|
9.1x
|
42.6x
|
|
EV / EBIT
|
29.5x
|
54.2x
|
26.8x
|
168x
|
|
Price / Free cash flow
|
39.4x
|
61.0x
|
33.2x
|
205x
|
|
Price / Earnings
|
32.9x
|
58.4x
|
29.1x
|
212x
|
A multiple below its own median is not a signal. In a business whose earnings may not be at a mid-cycle level, the denominator is the question, not the ratio.
2. What the current price implies
Reverse-engineering the discounted cash flow: holding the discount rate at 9.5% and terminal growth at 3%, these are the free-cash-flow paths that would justify the current enterprise value.
Conservative
+11%
FCF CAGR required over ten years if margins compress to 45% by FY2030.
Base
+16%
FCF CAGR implied if operating margin holds near 55% and capital intensity stays low.
Optimistic
+23%
Required if a second demand wave from inference and sovereign build-outs extends the cycle.
Sensitivity: a 100bp change in the discount rate moves implied value by roughly 14%. A 5-point change in the terminal operating margin moves it by roughly 11%.
3. The assumptions you should argue with
Terminal operating margin
We use 45-55%. History offers no analogue for a semiconductor company sustaining 60%+ through a full cycle.
Duration of the build-out
We assume elevated AI capex through FY2029, then normalisation. This single assumption drives most of the range.
Not investment advice. Fundamenta does not publish price targets, buy/sell ratings or portfolio recommendations. Valuation ranges are educational illustrations of stated assumptions.
Key risks
Ordered by how much each would change the investment case, not by how likely we think it is. Each states what evidence would tell you it is happening.
Customer concentration and in-house silicon
Four customers represent roughly 46% of revenue. Each is developing internal accelerators, and each has a structural incentive to reduce dependence on a single supplier earning 75% gross margins.
Watch for: the concentration disclosure in the next 10-K; hyperscaler capex commentary that separates internal from purchased silicon.
Demand is capital expenditure, and capital expenditure is cyclical
Revenue depends on a small number of budgets that are set annually and can be paused. Semiconductor history is a record of demand that looked structural until it was not.
Watch for: order backlog commentary, inventory days rising above 110, purchase obligations growing faster than revenue.
Export controls and geopolitical exposure
Rules on advanced compute exports have already removed a material market once. Manufacturing is concentrated in Taiwan, which is a separate and larger tail risk.
Watch for: revenue by geography in the 10-K; changes to Item 1A language on licensing.
Erosion of the software moat
Compiler-level abstraction layers reduce the cost of targeting non-NVIDIA hardware. This is slow, and it is the risk least visible in any quarterly number.
Watch for: framework default backends; adoption of vendor-neutral compilation in production training stacks.
Supply chain single points of failure
Advanced packaging and high-bandwidth memory both run through a handful of suppliers. Constraint caps upside; it does not threaten the business.
Watch for: supplier capacity announcements; lead-time commentary in CFO remarks.
Research on NVIDIA
Everything we have published on this company, newest first.
The full six-year review: business model, segment economics, capital allocation and the valuation framework.
Aug 12, 2026 · 18 min · Company analysis
Revenue, gross margin and inventory read against the prior four quarters, plus two disclosure changes and what they obscure.
Jul 24, 2026 · 9 min · Quarterly analysis
Three semiconductor equipment makers on revenue durability, gross margin, R&D intensity and returns over a full cycle.
Jul 11, 2026 · 15 min · Comparison