trunk/46746185c803a05ff6dc310236896f888e62832e
Apple's hidden AI infrastructure project just leaked in a PyTorch commit.
Deep Dive
A cryptic commit to the public PyTorch repository has revealed Apple's internal AI compiler project, codenamed "AOTI" (Ahead of Time Inductor). The commit, tagged by an Apple engineer, fixes deserialization issues for the proxy executor, confirming active development. This leak provides the first concrete evidence of Apple's behind-the-scenes efforts to build a high-performance AI inference engine, separate from its public MLX framework, to compete directly with NVIDIA's CUDA and OpenAI's Triton.
Why It Matters
This leak exposes Apple's serious, secretive push to build the AI hardware/software stack needed to power future devices and compete with NVIDIA.