CNBC·2025-11-21 13:00Core Insights - Nvidia reported significant profits driven by its graphics processing units (GPUs) that excel in AI workloads, indicating a strong market demand for AI-related technologies [2] - The landscape of AI chips is evolving, with custom application-specific integrated circuits (ASICs) gaining traction among major tech companies, potentially reducing reliance on Nvidia GPUs [3][14] Group 1: Nvidia and GPU Market - Nvidia has transitioned from primarily gaming GPUs to AI workloads, shipping approximately 6 million current-generation Blackwell GPUs in the past year [5][12] - The shift towards AI began around 2012 with the introduction of AlexNet, which demonstrated the effectiveness of GPUs in training neural networks [6][7] - Nvidia's GPUs are paired with CPUs in server rack systems for data centers, optimizing them for both training and inference phases of AI computation [8][9] - Nvidia's GPUs are sold to cloud providers like Amazon, Microsoft, and Google, who then rent them to AI companies, with notable deals including a $30 billion agreement with Anthropic [11][12] Group 2: Custom ASICs and Competitors - Major tech companies are developing custom ASICs, such as Google's Tensor Processing Units (TPUs) and Amazon's Trainium, which are designed for specific AI tasks and can outperform general-purpose GPUs in certain scenarios [3][14][18] - Custom ASICs are seen as a long-term investment for large cloud providers, offering better control over workloads despite higher upfront costs compared to GPUs [16][19] - Google has been a pioneer in custom ASICs, releasing its seventh-generation TPU in November 2025, while Amazon's Trainium boasts 30% to 40% better price performance compared to other hardware [17][19]