stockrowWhat it does
NVIDIA is a data-center-scale AI infrastructure company headquartered in Santa Clara, California. Jen-Hsun Huang is President and Chief Executive Officer. It designs computing platforms that help customers train and run AI models, process data, simulate scientific problems, create graphics, operate robots and develop automated-driving systems.
Its basic building blocks are GPUs, CPUs, data-processing units, networking equipment and software. GPUs perform many calculations at once, which suits AI training and inference, the process of using a trained model. NVIDIA combines those components into systems that can link hundreds of thousands of GPUs as one large computer. Its CUDA development platform, libraries, application frameworks and tools let developers write software for those systems.
The Data Center platform is sold as rack-scale systems, subsystems or modules, plus software and services. Customers use it in cloud, hyperscale, on-premises and edge data centers. NVIDIA also sells paid licenses for NVIDIA AI Enterprise and vGPU software.
Outside data centers, GeForce GPUs and GeForce NOW serve gaming; RTX and Quadro/NVIDIA RTX products serve professional graphics and design; and the DRIVE Hyperion platform combines in-vehicle computing, operating-system software, sensors and development tools for automotive applications. Management said on the call that the company has commenced production shipments of Vera Rubin, its next platform for AI infrastructure.
How it makes money
NVIDIA reports two segments. Compute & Networking includes Data Center accelerated-computing and networking platforms, AI solutions and software, plus Automotive platforms and software. Graphics includes GeForce GPUs for gaming PCs and Quadro/NVIDIA RTX GPUs for enterprise workstation graphics.
The company earns most of its revenue by selling hardware systems and components through direct customers, including OEMs, ODMs, cloud providers, distributors and system integrators. Its data-center offering extends beyond chips: it includes GPUs, CPUs, DPUs, NVLink interconnects, InfiniBand and Ethernet equipment, cables, switch chips and systems, and software. That wider system content is part of the company’s revenue opportunity per installed data-center capacity.
Software is both embedded in platforms and sold separately. NVIDIA AI Enterprise is licensed software for building and deploying AI applications, while NVIDIA vGPU software serves graphics-rich virtual desktops and workstations. Management said on the call that NVIDIA has introduced a model in which it makes take-or-pay commitments to a portion of a neocloud facility’s capacity and shares in rental revenue above a minimum floor; it described this as a recurring, usage-linked stream alongside core platform sales.
FY2026 revenue was $215.9 billion; Q2 FY2027 total revenue was $96 billion. Compute & Networking generated $193.5 billion in FY2026, or 89.6% of revenue, while Graphics generated $22.5 billion. Within the reported product markets, Compute revenue was $162.4 billion and Networking revenue was $31.4 billion in FY2026.
| Product or service | Revenue, FY2026 | Share | Change on the year |
|---|---|---|---|
| Compute | $162.4 billion | 75.2% | 58.9% |
| Networking | $31.4 billion | 14.5% | 141.5% |
| Gaming | $16.0 billion | 7.43% | 41.3% |
| Professional Visualization | $3.2 billion | 1.48% | 69.9% |
| Automotive | $2.3 billion | 1.09% | 38.7% |
| OEM and Other | $619.0 million | 0.29% | 59.1% |
Customers and how it sells
NVIDIA’s customers include public and private cloud providers, AI model makers, enterprises, startups, public-sector entities, PC gamers, professional creators and automotive ecosystem partners. Direct customers include add-in-board makers, distributors, OEMs, ODMs, cloud providers, AI model makers and system integrators; indirect customers buy through those channels.
The company’s sales and technical teams work with partners on system design, testing and qualification. Application engineers and solution architects help cloud providers, model developers and enterprise users optimize infrastructure, model training and AI applications. Management said on the call that AWS is deploying an additional 2,000,000 GPUs from Q2 FY2027 through Q2 FY2029.
Revenue is concentrated among a limited number of direct and indirect customers. In FY2026, one direct customer accounted for 22% of total revenue and another accounted for 14%, primarily in Compute & Networking. Revenue by geography reflects the headquarters location of direct customers, rather than necessarily where products ship or are ultimately used.
The United States represented $149.6 billion, or 69.3%, of FY2026 revenue; Taiwan represented $42.3 billion, or 19.6%; and China including Hong Kong represented $19.7 billion, or 9.11%. The company generated no China data-center compute revenue in its forward outlook because of geopolitical uncertainty.
| Region | Revenue, FY2026 | Share | Change on the year |
|---|---|---|---|
| UNITED STATES | $149.6 billion | 69.3% | 93.1% |
| Taiwan | $42.3 billion | 19.6% | 79.4% |
| China (including Hong Kong) | $19.7 billion | 9.11% | −21.4% |
| Other | $4.3 billion | 1.99% | −1.56% |
Suppliers and costs
NVIDIA uses a fabless, contract-manufacturing model. It designs products and relies on suppliers for wafer fabrication, assembly, testing and packaging, while directly procuring some memory, substrates and other components. Taiwan Semiconductor Manufacturing Company Limited and Samsung produce semiconductor wafers. SK Hynix, Micron Technology and Samsung supply memory; independent subcontractors and contract manufacturers including Hon Hai Precision Industry, Wistron and Fabrinet assemble, test and package final products.
The supply chain is mainly concentrated in Asia, although NVIDIA is expanding into the United States and Latin America. The company may place non-cancellable component orders, pay premiums and provide deposits to secure capacity ahead of historical lead times.
Cost of revenue includes semiconductors, fabrication, assembly, testing, packaging, boards and devices, manufacturing support, memory and component costs, tariffs, shipping, inventory and warranty provisions, licenses and service arrangements. Management said on the call that memory-price increases exceeded prior expectations and were expected to rise further. It expects supply to remain a bottleneck through the end of fiscal 2028.
- Revenue$215.9B
- Cost of revenue−$62.5B
- Research and development−$18.5B
- Selling, general and administrative−$4.6B
- Operating income$130.4B
Competitive position
NVIDIA competes in GPUs, CPUs, DPUs, embedded system-on-chips, AI accelerators and data-center interconnects. Named competitors in accelerated computing include AMD, Huawei and Intel. Cloud companies including Alibaba Group, Alphabet, Amazon, Baidu, Huawei and Microsoft develop internal hardware and software with accelerated or AI-computing functions. In automotive and embedded systems, competitors include Ambarella, AMD, Broadcom, Intel, Qualcomm, Renesas, Samsung and Tesla. Networking competitors include AMD, Arista Networks, Broadcom, Cisco, Hewlett Packard Enterprise, Huawei, Intel, Lumentum and Marvell.
The company identifies performance, product breadth, access to customers and partners, distribution, software support, industry-standard APIs, manufacturing capabilities, processor pricing and total system cost as competitive factors. It expects competitors and new entrants to offer lower-priced products or products with different performance or features. It was reported on 26 October 2026 that Arista Networks was winning some AI data-center projects that might otherwise have used NVIDIA InfiniBand interconnects.
Our reading: CUDA, the installed developer base, software libraries and application frameworks are a strong ecosystem moat because they link the hardware to applications and developers across several end markets.
Our reading: NVIDIA’s ability to combine chips, networking, systems and software into data-center-scale platforms is a strong integration moat, but cloud companies’ internal hardware efforts and competing accelerators directly test it.
Our reading: Manufacturing is not a moat; reliance on external foundries, memory suppliers and contractors makes supply availability and cost important competitive constraints.
| Competitor | Where they compete | Threat (our reading) |
|---|---|---|
| Advanced Micro Devices, Inc. | AI accelerators, GPUs, SoCs and networking | High |
| Intel Corporation | GPUs, CPUs, AI processors and networking | High |
| Huawei Technologies Co. Ltd. | AI processors, cloud hardware and networking | High |
| Broadcom | Embedded SoCs and networking products | Medium |
| Arista Networks | AI data-center Ethernet interconnects | Medium |
| Amazon | Internal cloud AI and CPU platforms | Medium |
| Microsoft Corporation | Internal cloud AI and CPU platforms | Medium |
| Alphabet Inc. | Internal cloud AI platforms | Medium |
Growth opportunities
NVIDIA’s growth plans center on expanding accelerated computing and AI infrastructure across cloud providers, AI labs, neoclouds, enterprises, sovereign customers, industrial users and edge deployments. It also targets gaming, professional visualization, automotive, robotics, healthcare, financial services, manufacturing and physical AI.
The company is moving to a one-year Data Center product cadence. Its Rubin platform was expected to commence production shipments in the second half of fiscal 2027, and management said on the call that it had begun production shipments of Vera Rubin. Management expects Vera Rubin to be deployed by major hyperscalers, neoclouds, AI labs and system OEMs. The company also sees growth in data-center CPUs, networking, enterprise software and its neocloud revenue-sharing arrangements.
Management said on the call that the company expects FY2028 revenue to grow by approximately 70%, while supply remains constrained. The company sizes AI infrastructure spending at $3.00 trillion to $4.00 trillion annually by the end of the decade. It also said Vera CPU opens a $200.0 billion total addressable market and that its revenue opportunity per gigawatt is roughly $40.0 billion with Vera Rubin.
In FY2026, Compute & Networking revenue grew 66.5% to $193.5 billion, and Graphics revenue grew 57.0% to $22.5 billion.
- $26.9BFY2022
- $27.0BFY2023
- $60.9BFY2024
- $130.5BFY2025
- $215.9BFY2026
Structural risks
- Export controls and China: U.S. export restrictions and licensing requirements limit products that NVIDIA can sell into China and other markets. The company says it was effectively foreclosed from China’s data-center compute market at the end of FY2026, and it excludes China data-center compute revenue from its forward outlook.
- Supply and demand matching: Long manufacturing lead times, uncertain component and capacity availability, product transitions and changing demand can create shortages or excess inventory. The company may make non-cancellable orders and prepayments before demand is certain.
- Supplier concentration: Foundries, memory suppliers, contractors and assembly and test providers are limited in number and geographically concentrated. A supplier loss, capacity shortfall, quality issue or supplier priority given to another customer can delay shipments or raise costs.
- Infrastructure constraints at customers: Customer deployments require data-center capacity, energy, capital, land, power, cooling and construction. Delays or shortages can reduce or defer purchases of AI infrastructure.
- Competition and technology change: Competitors may offer lower prices, better performance or features, while internal cloud hardware, new architectures, open-source AI deployed on rival platforms and changing standards can reduce demand.
- Margin and inventory pressure: Higher memory and component costs, tariffs, product mix changes, inventory provisions and warranty costs can reduce gross margin. FY2026 included a $4.5 billion H20 charge for excess inventory and purchase obligations.
Bottom line
Our reading
Our reading: NVIDIA is primarily a data-center AI infrastructure business, with Compute & Networking supplying most FY2026 revenue; its platform combines hardware, networking and software rather than selling a processor alone. Our reading: its CUDA ecosystem, system-level design and broad cloud, enterprise and developer reach support the platform, while its growth is constrained by external manufacturing capacity, customer infrastructure build-outs, intense competition and export restrictions. Our reading: China restrictions and customer concentration are material exposures alongside a supply chain centered on external partners.