NVIDIA designs and sells AI computing infrastructure, with the GPU as its core product. NVIDIA's GPUs are the foundational hardware for training and running large AI models, and the company sells full-stack systems that combine GPUs, CPUs, networking hardware, and software into integrated rack-scale platforms. The flagship GB200 NVLink 72 system connects 72 Blackwell GPUs and 36 Grace CPUs into a single rack-scale computer. NVIDIA's customers include hyperscalers like Amazon, Microsoft, and Google, frontier AI model builders like OpenAI and Anthropic, enterprises, sovereign governments, and consumers who buy GeForce GPUs for gaming PCs. The company reports two segments: Compute & Networking, which includes data center, networking, automotive, and software, and Graphics, which includes gaming and professional visualization. NVIDIA's business model centers on selling premium hardware justified by performance-per-watt leadership — since AI data centers monetize token output directly, customers pay a premium for hardware that produces more tokens per watt. NVIDIA also captures value across compute, networking, and software within a single deployment. NVIDIA's CUDA software platform, with over 7.5 million developers and 6,000+ applications, acts as a retention mechanism and key competitive moat. NVIDIA operates on an annual product cadence, shipping new GPU architectures roughly every year, with each generation targeting a roughly 10x reduction in cost per token. NVIDIA is fabless, relying on TSMC for manufacturing, giving it a largely variable cost structure and non-GAAP gross margins targeting the mid-70% range.