NVIDIA's business model has evolved from a focused graphics chip company into a full-stack computing platform business that generates revenue across hardware, software, and services. Understanding this evolution — and the deliberate architectural choices that enabled it — is essential to evaluating NVIDIA's current market position and future trajectory. The hardware business remains the largest revenue contributor, structured across two primary segments: Data Center and Gaming, with smaller contributions from Professional Visualization and Automotive. The Data Center segment, which encompasses AI training and inference GPUs, networking products (acquired through the Mellanox acquisition in 2020), and cloud computing infrastructure, has become the dominant revenue driver, representing approximately 80% of total revenue in fiscal year 2024 following the AI demand explosion. The GPU product line for data center is sold at dramatically different price points than gaming GPUs. An H100 SXM5 module sells for approximately $30,000-$40,000 per unit, compared to a consumer gaming GPU that might retail for $500-$1,500. Enterprise and cloud customers purchasing clusters of thousands of H100s for AI training are spending tens of millions of dollars per order, and the largest hyperscaler deployments involve billions of dollars in GPU infrastructure investment. This pricing dynamic, combined with the AI-driven demand surge, explains how NVIDIA's data center revenue grew more than tenfold in two years. The Gaming segment — NVIDIA's original and historically largest business — sells GeForce GPUs to PC gamers through add-in board partners like ASUS, MSI, and Gigabyte. While gaming revenue has been overshadowed by data center growth, it remains a multi-billion dollar business that provides important technology development leverage: gaming GPU architectures share the same underlying design as data center GPUs, allowing NVIDIA to amortize R&D investment across multiple product lines and maintain a consumer brand presence that supports talent recruitment and ecosystem development. The software layer is where NVIDIA's most durable competitive advantage resides, and increasingly where it is building additional revenue streams. CUDA, the foundational GPU programming platform, is provided free of charge — a deliberate strategic choice to maximize ecosystem adoption and create switching costs. The bet has paid off enormously: the CUDA ecosystem encompasses millions of trained developers, hundreds of thousands of applications, and decades of optimized libraries and frameworks. A competitor that builds a technically superior GPU still faces the nearly insurmountable challenge of convincing the AI research and development community to rewrite their software stack from CUDA to an alternative. Beyond CUDA, NVIDIA has built an extensive suite of software tools — collectively branded as the NVIDIA AI Enterprise software suite — that are available on a subscription basis. These include cuDNN (deep neural network library), TensorRT (inference optimization), NeMo (large language model training framework), and RAPIDS (data science acceleration). The transition toward software subscription revenue is strategically significant: software carries near-100% gross margins compared to hardware margins in the 60-70% range, and subscription revenue is recurring and predictable in ways that hardware sales are not. The networking business, acquired through the $7 billion Mellanox acquisition in 2020, has proven strategically prescient. As AI clusters scale to thousands of GPUs, the interconnect network — the infrastructure that allows GPUs to communicate with each other during training — becomes a critical performance bottleneck. NVIDIA's InfiniBand networking products, which Mellanox pioneered, provide the highest-bandwidth GPU interconnect in the market and are the preferred networking architecture for the largest AI training clusters. The Mellanox acquisition effectively gave NVIDIA end-to-end control over the entire AI infrastructure stack, from the GPU chip to the network fabric. The platform business model — hardware plus software plus ecosystem — creates compounding returns. Each new AI researcher who learns to code on CUDA deepens the ecosystem. Each optimized library reduces the barrier to adoption for the next researcher. Each enterprise that deploys NVIDIA infrastructure at scale creates integration dependencies that increase switching costs. NVIDIA does not merely sell GPUs; it sells access to a computational ecosystem that has been built and refined over nearly two decades, and the value of that ecosystem grows with every new participant.