The Decade-Long GPU Bet
When NVIDIA launched the CUDA platform in 2006, this strategic bet was of limited commercial value at the time, the great majority of customers still seeing the GPU only as a gaming graphics card, the application scene of general-purpose computing of scant demand, and the company had internal debate over whether the direction was worth continued investment. Only after the rise of deep learning, when neural-network training fit highly with the GPU parallel architecture, did this investment at last see an explosive market return nearly a decade later, and NVIDIA thereby became the most core compute-infrastructure supplier behind this wave of AI.