Google open-sourced the GPU engine running Local AI across millions of devices. It’s called ML Drift. It gives developers one GPU layer across: 🍎 Metal + 🌐 WebGPU + 🤖 OpenCL / OpenGL ES + 🖥️ Windows + Linux via Dawn And it’s built for LLMs. ML Drift changes its kernels depending on what the model is doing such as 📥 prefill → compute-optimized ✍️ decode → bandwidth-optimized 🧠 custom KV-cache layout 💾 memory reuse Google reports up to 12% lower memory overhead in its Gemma tests. Google says ML Drift runs today across millions of devices in Chrome, YouTube Shorts, Photos and Meet. Chrome uses it to accelerate Gemini Nano locally. it’s Apache 2.0 so developers now have one open GPU foundation for running AI across phones, browsers, Macs and PCs. This is something to test! Link in ALT.