(https://www.cnbc.com/2026/07/20/amd-helios-microsoft-ai-nvidia.html) with Microsoft on the customer list—solid execution, no question. But Wall Street is reading this as a training-chip story when it's actually a margin-compression story. Inference workloads (running pre-trained models) have lower unit economics than training, and that's exactly what AMD is winning. Microsoft, Meta, OpenAI want diversity in supply chains, not cheaper training rigs.

Research these companies