Nvidia was founded in 1993 to build graphics chips for the emerging market of 3-D video games. For years that was the whole story: faster, prettier games, driven by a chip — the GPU — designed to do the same simple maths on thousands of pixels at once.
The accident that became the future
That design — massive parallelism — turned out to be exactly what a completely different problem needed: training neural networks. In 2006 Nvidia released CUDA, a way to program its GPUs for general-purpose maths, not just graphics. Researchers quietly began using gaming cards to do science.
The deep-learning spark
The moment everything changed came in 2012, when AlexNet — a neural network trained on a pair of Nvidia GPUs — crushed the field at the ImageNet image -recognition contest. It proved that GPUs made deep learning practical, and demand for Nvidia hardware shifted from gamers to AI labs.
Powering the AI era
Today Nvidia's data-centre GPUs (the A100 and H100 and their successors) are the default hardware for training large language models, and its CUDA software has become a deep competitive moat. A company that started by rendering game worlds now sits at the centre of the AI economy — one of the clearest examples of how a technology built for one purpose can end up reshaping another entirely.