Oracle Cloud to deploy 50,000 AMD AI chips, signaling new Nvidia competition (www.cnbc.com)

Oracle Cloud Infrastructure said it will deploy 50,000 AMD Instinct MI450 GPUs beginning in the second half of 2026, signaling a major push to offer AMD-based alternatives to Nvidia in data-center AI. The MI450 is AMD’s first AI accelerator designed for rack-scale aggregation—72 chips can be linked to act as a single system—making it suitable for both large training clusters and production inferencing. OCI executives stressed the importance of AMD’s software stack and expect early, strong uptake in inference workloads rather than displacing Nvidia outright in high-end training. The announcement matters because it underscores a broader industry shift toward multi-vendor GPU strategies: OpenAI has announced large, multi-gigawatt deals with AMD (and maintains deep ties with Nvidia), and Oracle itself has a high‑value cloud pact with OpenAI. For AI/ML teams, this could mean greater supply diversity, potential cost competition, and more choices in performance/power tradeoffs—especially for inference at scale. If AMD’s rack-scale approach and software maturity deliver, cloud providers may begin to chip away at Nvidia’s dominant data-center share, while enterprises gain more flexibility for deploying large models and production pipelines.