NVIDIA makes money by selling chips — specifically GPUs, which are processors built to do millions of calculations at the same time. Most of those chips go into giant data centers run by cloud companies, AI developers, and large businesses. When a company wants to train an AI model or run AI-powered services, they almost always need NVIDIA's hardware to do it. NVIDIA also sells gaming GPUs to consumers, chips for self-driving car systems, and software that makes all of it work together. The company designs everything — the chips, the networking gear, and the software stack — but pays outside factories to manufacture the actual hardware. Every time a data center adds more AI capacity, NVIDIA collects revenue. The diagram below traces where the money goes.
Five years of financial data tell a story that is almost without precedent in corporate history. In fiscal year 2022, NVIDIA had $26.9 billion in revenue. That number barely moved the following year — $27.0 billion in fiscal year 2023. Then something broke open.
Revenue more than doubled in fiscal year 2024 to $60.9 billion. Then it doubled again to $130.5 billion in fiscal year 2025. Then it grew another 65% to $215.9 billion in fiscal year 2026. This is not gradual growth. It is a near-vertical line. The trigger was the explosion in demand for AI computing, and NVIDIA's Blackwell chips — connecting up to 36 CPUs and 72 GPUs into a single liquid-cooled system — became the hardware that every major AI company needed. Data Center revenue alone grew 68% in fiscal year 2026. The cash the business generates reflects that same trajectory.
Gross margin tells you how much profit is left after paying to make the product. NVIDIA's gross margin has expanded dramatically over this period. It sat at 64.9% in fiscal year 2022, compressed to 56.9% in fiscal year 2023 during a difficult inventory period, then surged to 75.0% in fiscal year 2025 as Blackwell demand overwhelmed supply. 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 suddenly restricted from being exported to China. Even with that hit, seven out of every ten dollars of revenue flowed through to gross profit. The balance sheet transformed too: NVIDIA went from carrying $9.0 billion in net debt in fiscal year 2022 to holding a net cash position by fiscal year 2025.
But the financial picture also contains real threats that are already visible, not just theoretical. The first is geography. The U.S. government has progressively restricted what NVIDIA can sell to China — blocking its most powerful chips, then restricting the H20 (a chip designed specifically to comply with earlier rules), and leaving NVIDIA effectively shut out of China's data center market by the end of fiscal year 2026. That H20 restriction alone cost $4.5 billion in a single quarter. At the same time, by being locked out of China, NVIDIA's competitors gained space to build larger developer and customer ecosystems there — a disadvantage that compounds over time.
The second risk is customer concentration. In fiscal year 2026, one customer accounted for 22% of total revenue and another accounted for 14%. Those two customers together represent more than a third of the entire business. If either slows its purchases — or builds its own chip instead — the revenue impact would be immediate and large. Several large cloud companies are already developing custom AI chips in-house, which is listed directly in NVIDIA's own filings as a competitive threat.
That dependency on outside factories is a third documented risk. NVIDIA's supply chain is mainly concentrated in Asia. It relies on TSMC and Samsung for chip manufacturing, SK Hynix, Micron, and Samsung for memory, and a small number of assembly partners. Because AI demand grew so fast, NVIDIA has had to place large, non-cancellable orders far in advance — sometimes more than twelve months ahead. If demand softens or shifts, those orders become liabilities. In fiscal year 2026, provisions for excess inventory and purchase obligations totalled $7.2 billion.
NVIDIA's business logic rests heavily on that CUDA ecosystem remaining the default. The company has invested over $76.7 billion in research and development since its inception. More than half of its engineers work on software, not hardware. The bet is that this software moat — built over nearly two decades — is deep enough that no competitor can replicate it quickly. But the rise of high-quality open-source AI models, which can run on a wider range of hardware, introduces an alternative path that could reduce that dependency over time.