Apple Acquires SAM Team
Apple reportedly acquired Meta’s SAM team to bring native on‑device segmentation into macOS 26, with demos showing SAM‑class models running in real time on M2 hardware via MLX. That strengthens Apple’s playbook for privacy‑first, local multimodal models across creative and manufacturing workflows. (x.com)
An X post by developer @me_bruno_dev claims Apple acquired Meta’s SAM research team and shows short demos of SAM‑class segmentation running via MLX on M2 hardware (x.com). As of April 2, 2026, major technology outlets that cover acquisitions and team moves have not reported a confirmed Apple purchase of Meta’s SAM group; recent reporting instead chronicles Meta hiring Apple AI staff and the wider talent flow between the companies. (bloomberg.com) (techcrunch.com). Multiple community projects have already ported SAM‑family models to Apple’s MLX runtime with explicit optimizations for M1/M2/M3 chips and interactive demos, demonstrating that SAM‑class inference is feasible on current Apple Silicon. (github.com) (huggingface.co). MLX is Apple’s native array and ML framework with WWDC documentation and a Metal‑backed execution path that emphasizes unified memory and device‑native kernels—capabilities MLX uses to reduce data movement on Apple Silicon. (developer.apple.com) (mlx-framework.org). Independent benchmarks and developer writeups show MLX often outperforms PyTorch’s MPS backend on many tensor ops and that real‑time, interactive segmentation prototypes have been demonstrated on M2‑class machines by contributors. (towardsdatascience.com) (deekshith.me). SAM‑family models are already being integrated into edge and industrial workflows for auto‑labeling and on‑device vision; platforms like Roboflow and AnyLabeling document SAM3/SAM integrations and automated labeling pipelines that are relevant to visual inspection and manufacturing automation. (blog.roboflow.com) (anylabeling.nrl.ai).
