AMD Came from Behind to Mount a Challenge in the AI Chip Wars (www.wsj.com)

AMD announced a landmark partnership with OpenAI in which the AI lab will buy "tens of thousands" of AMD chips to power about 6 gigawatts of computing capacity for inference — the runtime work that lets models like ChatGPT respond to user queries. The deal sent AMD shares up roughly 24% and helped lift the company’s market value above $330 billion, underscoring how AMD has reoriented from consumer GPUs and PC processors into high-performance data-center accelerators tailored for the AI era. Technically and strategically, the agreement is a major validation of AMD’s push into server-grade AI silicon and a direct challenge to Nvidia’s dominance in AI accelerators. Procuring massive inference capacity from AMD suggests growing multi-vendor deployments for model serving, which can ease supply constraints, pressure pricing, and spur software and systems work to optimize runtimes across different accelerator architectures. For the AI/ML community, that means potential diversification of hardware choices for inference fleets, new opportunities (and engineering work) to adapt frameworks and kernels, and a more competitive market that could accelerate innovation in both chips and infrastructure.