Origins
NVIDIA was founded in 1993 to build graphics chips for personal computers. Its GeForce line became one of the best-known names in PC gaming. In the mid-2000s it introduced CUDA, a programming platform that let developers use graphics processors for general calculations, not only for drawing images.
Why a graphics chip suits data centres
A central processing unit (CPU) handles a wide variety of tasks one after another very quickly. A graphics processing unit (GPU) has many simpler cores that work in parallel. Drawing a 3D scene means doing the same maths on millions of pixels at once, and so does training a machine learning model. That overlap is why GPUs moved from gaming PCs into data centres.
How the business makes money
- Chip and system sales. GPUs sold to PC makers and graphics card makers, plus complete accelerator systems and boards for data centres.
- Networking. High-speed interconnect and networking equipment that links many processors together, a business expanded by its acquisition of Mellanox.
- Software and platforms. CUDA and related libraries are free to use but tie developers to NVIDIA hardware, and some enterprise software is sold by licence or subscription.
NVIDIA reports its results in segments and also breaks revenue down by market platform, such as data centre and gaming. The annual report shows the current split.
What protects it
- Software ecosystem. Years of CUDA libraries and developer knowledge make it costly to switch to another chip.
- Full-stack approach. It sells chips, boards, networking and software designed to work together.
Dependencies
As a fabless company, NVIDIA depends on outside foundries, especially TSMC, and on advanced packaging capacity. Demand is also concentrated among a relatively small number of very large customers, and export rules can restrict where some products are sold. Its risk factors section covers each of these.
This page explains how a company is organised. It is not investment advice and says nothing about whether a share is worth buying or selling.