Modular Platform 25.4 launches with AMD Instinct GPU support

AI Impact Summary

Modular Platform 25.4 launched with significant AMD Instinct GPU support, including MI300X and MI325X, alongside a new Kubernetes-native scaling system called Mammoth for GenAI inference. This release delivers up to 53% throughput improvements on prefill-heavy BF16 workloads using Llama 3.1, Gemma 3, and Mistral models, and enables deployment across AMD and NVIDIA hardware without code changes. The integration of Mojo kernels directly into Python workflows via nightly builds further expands developer capabilities.