NVIDIA is an American semiconductor and computing company headquartered in Santa Clara, California, founded in 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem. Originally built around graphics processing units (GPUs) for video games and visual computing, NVIDIA became the dominant hardware supplier for training and running deep learning models, a position that made it central to the economics of the generative AI boom of the 2020s and, at points during 2024 and 2025, the most valuable publicly traded company in the world.
History
NVIDIA spent its first fifteen years primarily as a graphics chip company competing for the personal computer and console gaming markets. A pivotal strategic decision came with the 2006 introduction of CUDA, a software platform that let developers use NVIDIA GPUs for general-purpose parallel computing rather than only rendering graphics. Adoption was gradual through the 2000s, but the 2012 AlexNet result, which trained a winning ImageNet model on two NVIDIA GPUs, demonstrated that the same massively parallel architecture suited to rendering pixels was also well suited to the matrix multiplications underlying Deep learning. Over the following decade, as neural networks grew from millions to hundreds of billions of parameters, demand for NVIDIA's data-center GPUs grew correspondingly, and the company progressively redesigned its highest-end chips, from the Tesla and Volta lines through Ampere, Hopper, and Blackwell, around the specific demands of AI training and inference rather than graphics alone.
From graphics to AI
By the early 2020s, NVIDIA's data-center segment, selling GPUs to cloud providers and AI labs training large models, had overtaken its historical gaming business as the company's primary revenue driver. Its H100 and successor chips became the preferred hardware for training Frontier models at organizations including OpenAI, Anthropic, and Meta, and NVIDIA's earnings and stock price became a widely watched proxy for the health of AI industry investment more broadly. The company's market capitalization grew from tens of billions of dollars in the early 2010s to more than three trillion dollars by 2024, a rise closely tied to the capital expenditure boom in AI data centers.
The CUDA moat
A central reason for NVIDIA's durable advantage over rival chipmakers has been CUDA's software ecosystem: because most deep learning frameworks, including PyTorch and TensorFlow, and most researchers' tooling were built and optimized against CUDA over more than a decade, switching to alternative hardware such as Google's TPU or AMD's competing chips has carried significant software-compatibility costs, a dynamic widely described as NVIDIA's software "moat" on top of its hardware lead.
Market position and risks
NVIDIA's dominance has drawn regulatory scrutiny in multiple jurisdictions over potential anticompetitive effects, and its reliance on Taiwan Semiconductor Manufacturing Company for chip fabrication has been flagged as a geopolitical vulnerability given tensions around Taiwan. The January 2025 DeepSeek market shock, in which a comparatively low-cost Chinese model appeared to match Western frontier systems, briefly wiped out a record single-day amount of NVIDIA's market value on fears that AI training might require far less compute than assumed, illustrating how tightly the company's valuation had become tied to continued belief in Scaling laws-driven demand for ever more GPU compute, a dependency some analysts have compared to the boom-bust dynamics of earlier technology cycles predicted, if imperfectly, by observations like Moore's law (and AI compute).