NVIDIA makes money by selling the computing hardware and software that powers artificial intelligence. Its main product is the GPU, a chip originally built to render video game graphics that turns out to be extraordinarily good at the kind of math AI models need. When a cloud company trains a large AI model, or a hospital runs drug discovery software, or a car company tests self-driving systems, the odds are high that NVIDIA chips are doing the work. Customers pay each time they order systems, and some also pay ongoing software license fees for tools like NVIDIA AI Enterprise. The data center business now dominates, accounting for $193.5 billion of the company's $215.9 billion in total revenue in fiscal year 2026. The diagram below traces where the money goes.
How NVIDIA Makes Money
flowchart TD
A["R&D Investment
$76.7B cumulative"] --> B["GPU, CPU, DPU
Chip Design"]
B --> C["Data Center
Platform $193.7B"]
B --> D["Gaming
Platform $16.0B"]
B --> E["Professional
Visualization $3.2B"]
C --> F["Cloud Providers
Enterprises AI"]
D --> G["Gamers
Streamers Creators"]
E --> H["Design & Creative
Professionals"]
F --> I["Total Revenue
$215.9B 71% margin"]
G --> I
H --> I
I --> J["Operating Cash Flow
$102.7B"]
J --> A
C --> K["CUDA Software
Ecosystem 7.5M devs"]
K --> C
Five years of financial data tell a story of near-vertical growth. Revenue was $26.9 billion in fiscal year 2022. It barely moved the following year, sitting at $27.0 billion in 2023. Then the AI spending wave arrived. Revenue jumped to $60.9 billion in 2024, then $130.5 billion in 2025, then $215.9 billion in 2026. That is roughly an eight-fold increase in three years. The growth came almost entirely from one place: the Compute and Networking segment, which includes data center chips and the high-speed networking gear that connects them.
NVIDIA Annual Revenue (Fiscal Years 2022 to 2026)
Revenue in billions of dollars. The flat 2022 to 2023 period gave way to explosive growth once AI infrastructure spending accelerated.
Profitability followed revenue upward. Gross margin, the share of each dollar of revenue left after paying to make the product, rose from 56.9% in fiscal year 2023 to 75.0% in fiscal year 2025. It dipped slightly to 71.1% in fiscal year 2026, partly because of a $4.5 billion charge tied to H20 chips that the U.S. government blocked from being sold into China. Even with that hit, the business generated $96.7 billion in free cash flow in fiscal year 2026, and the company went from carrying $9 billion in net debt in 2022 to a net cash position by 2025.
$96.7B
Free cash flow in fiscal year 2026, up from $3.8B just three years earlier
The cash generation is real and large. But the five-year record also shows how quickly things can reverse. The flat 2022 to 2023 period was not a crisis. It was simply a pause between product cycles. The lesson is that demand for NVIDIA products is not steady. It surges when customers are building out new AI infrastructure, and it can stall when they pause to absorb what they have already bought.
What 'fabless' manufacturing means
NVIDIA designs its chips but does not own the factories that make them. Instead, it pays outside foundries to do the manufacturing. The main chip factory NVIDIA relies on is TSMC, based in Taiwan. This keeps NVIDIA's own costs lower, but it also means the company depends entirely on a small number of outside partners to actually produce its products.
That manufacturing dependency is one of the clearest risks in the business. NVIDIA relies on TSMC in Taiwan and a small number of other suppliers in geopolitically sensitive parts of Asia. If a factory goes offline, or if trade tensions between the United States and Taiwan escalate into something more serious, NVIDIA cannot simply switch to another supplier overnight. The company has said it is working to expand into the United States and Latin America, but that process takes years.
2025
crisis
China Export Controls Force a $4.5B Charge
In April 2025, the U.S. government told NVIDIA it needed a license to export its H20 chips to China. Demand for H20 collapsed almost immediately, and NVIDIA took a $4.5 billion charge for excess inventory and purchase commitments it could no longer fulfill. By the end of fiscal year 2026, NVIDIA said it was effectively shut out of China's data center market entirely. The company noted this helped competitors build larger customer bases in China that could challenge NVIDIA globally.
Geopolitical risk is not the only documented threat. Two customers alone accounted for 22% and 14% of total revenue in fiscal year 2026. If either of those customers slowed their orders, or decided to build their own chips instead of buying from NVIDIA, the revenue impact would be immediate and large. Several of NVIDIA's biggest customers, including cloud giants like Amazon, Alphabet, and Microsoft, are already designing their own AI chips in-house. NVIDIA's own 10-K filing names them as competitors.
36%
Share of total fiscal 2026 revenue from just two direct customers (22% and 14%)
Why product transitions create inventory risk
NVIDIA launches new chip generations roughly every year. When a new generation ships, demand for the old one drops fast. Because chip manufacturing takes many months, NVIDIA has to order components long before it knows exactly how many customers will want them. If a product gets blocked, delayed, or replaced faster than expected, the company can be left holding inventory it cannot sell, which is exactly what happened with the H20.
Supply chain timing risk compounds the customer concentration problem. NVIDIA places large, non-cancellable orders with its suppliers many months in advance because manufacturing lead times can stretch beyond twelve months for some products. If customer demand shifts suddenly, either because a new chip generation arrives, because export rules change, or because AI spending slows, NVIDIA can find itself holding inventory it has already paid for but cannot sell. The company acknowledged that inventory provisions reduced gross margin by 2.6% in fiscal year 2026.
$3.8B
Fiscal 2023 Free Cash Flow
$96.7B
Fiscal 2026 Free Cash Flow
The scale of the cash generation in three years is extraordinary. It also shows how dependent the business is on a single sustained wave of AI infrastructure spending.
There is also a competitive pressure building from an unexpected direction. The rise of high-quality open-source AI models is making powerful AI available to more developers at lower cost. NVIDIA's own filing notes that if those open-source models are deployed on competitors' platforms, demand for NVIDIA products could fall. The company's software ecosystem, built around a platform called CUDA that developers have used for nearly two decades, is a meaningful barrier to switching, but it is not an absolute one.
NVIDIA has invested $17.5 billion in private companies and infrastructure funds in fiscal year 2026 alone, including early-stage AI startups that buy NVIDIA products through cloud providers. Some of those startups may never become profitable.
The Bet
Spending on AI infrastructure keeps growing fast enough, and for long enough, that NVIDIA's current customers do not slow their orders before the next wave of customers steps in to replace them. The model also assumes that no competitor, whether a cloud company building its own chip, a Chinese rival now free to develop in a market NVIDIA cannot enter, or a new entrant, captures enough of the workload to matter before NVIDIA's next architecture arrives. If AI capital spending plateaus, or if the customers who today account for more than a third of revenue decide to source chips elsewhere, the revenue trajectory that produced $96.7 billion in free cash flow in a single year has no guaranteed floor.
Open question
NVIDIA's numbers over the last three years are unlike almost anything seen in the history of technology hardware. The cash flows are real, the gross margins are high, and the company has moved from net debt to a net cash position. But the business sits at the intersection of three forces it cannot fully control: U.S. export policy, the capital spending decisions of a handful of giant cloud customers, and the pace at which those same customers build chips of their own. Is the AI infrastructure buildout a sustained multi-decade shift in how computing works, or is it a concentrated spending surge that will level off once the first wave of data centers is built, and does NVIDIA's software ecosystem run deep enough to hold its customers even when the hardware competition intensifies?
[1]
NVIDIA 10-K filing for fiscal year ended January 25, 2026, Item 1 Business Description
[2]
NVIDIA 10-K filing for fiscal year ended January 25, 2026, Item 7 Management Discussion and Analysis
[3]
XBRL financial data fiscal years 2022 through 2026 as provided
Compiled · 10-K · FY2026
Supply Chain and Manufacturing
Long lead times for manufacturing combined with difficulty predicting customer demand have caused major mismatches between supply and inventory. The company has experienced inventory write-downs, cancellation penalties, and gross margin declines from these demand forecast errors, and this pattern is expected to continue.
Product Quality and Defects
Complex hardware and software products can contain defects, security vulnerabilities, or performance failures that may only be discovered after customers receive them. The company remains responsible for warranty costs, recalls, and repairs, which can divert engineering resources and damage customer relationships and brand reputation.
Customer Concentration
Two direct customers represented 22% and 14% of total revenue in fiscal 2026. Loss of any major customer, significant purchase reductions, or trade restrictions preventing sales to large customers would materially harm financial results and operations.
Supplier Dependency
The company depends on a small number of foundries and contract manufacturers located in geopolitically sensitive regions including Taiwan and China. Loss of a supplier, supply disruptions, manufacturing delays, or quality failures at these critical partners could prevent the company from meeting customer demand and scaling production.
Geopolitical and Trade Risks
The company has approximately 6,000 employees in Israel and relies on suppliers and manufacturers in Taiwan, China, and other geopolitically tense regions. Military conflicts, export controls, tariffs, sanctions, or other trade restrictions in these areas could materially disrupt operations, product development, and revenue.
10-K Item 1A · Risk Factors
