Meta says it worked with Apple to make Muse Spark Open run locally on M5 Ultra Mac Studios, letting users pair two to four Studios to code or do agentic work

You can now run a massive AI model locally on your Mac Studio, potentially reducing costs and increasing privacy for complex coding and agentic tasks.

Meta and Apple have announced a collaboration allowing developers to run Meta's Muse Spark Open AI model locally on high-end Mac Studio computers. This partnership enables the use of an 816-billion-parameter, open-weight model on Apple's M5 Ultra chip-equipped machines. Users can pair between two and four Mac Studios to perform tasks like coding and agentic work. Running the model locally offers enhanced security and cost savings compared to cloud-based solutions. This collaboration signifies a potential shift in Silicon Valley alliances, bridging a historical gap between the two tech giants. Meta's AI chief, Alexandr Wang, highlighted that Apple's Mac Studios offer sufficient memory for practical local execution of such large models. Developers can access the model via platforms like Hugging Face, with optimizations for Apple's silicon and its MLX framework.

This collaboration marks a significant move towards on-device AI processing, suggesting a market trend where powerful, open-weight models are becoming accessible outside of massive data centers. It could drive demand for high-end local hardware and influence how AI development tools are distributed and utilized, particularly for developers seeking greater control and reduced cloud dependency. The partnership also signals Apple's ambition to become a key player in the AI hardware ecosystem beyond its current chip dominance.

The ability to scale by pairing multiple Mac Studios could appeal to professionals needing substantial local compute power without the overhead of cloud infrastructure. This on-premise AI capability, especially with an open-weight model, may spur innovation in secure, privacy-focused AI applications. It positions Apple's silicon as a viable alternative to specialized AI hardware for certain intensive tasks, potentially challenging existing market dynamics dominated by cloud providers and GPU manufacturers.