Apple and Meta have teamed up to enable Meta’s Muse Spark Open, a large-scale AI model, to run efficiently on Apple’s Mac Studio desktops, marking a rare collaboration between the two tech giants since 2017.
- Muse Spark Open runs on dual Mac Studios with M5 Ultra chips
- Model uses 42 billion active parameters and requires high memory capacity
- Local model operation ensures user privacy, with no data sent to Meta
What happened
Apple and Meta collaborated to optimize the Muse Spark Open AI model to run on Apple’s Mac Studio desktops. This groundbreaking effort allows the model, which contains 816 billion parameters, to be processed locally using Apple’s M5 Ultra chips and the MLX software framework. Meta CEO Mark Zuckerberg and Apple's new CEO John Ternus acknowledged the partnership publicly on social media, highlighting the technical achievement.
The Muse Spark Open model requires substantial computing resources, running across two linked Mac Studio machines to hold the entire model and provide efficient performance for tasks like coding, video analysis, and running AI agents. Each machine can access up to 512GB of memory, highlighting the system's high-end hardware requirements.
Why it matters
This collaboration marks a significant thaw in the relationship between Apple and Meta, who have not had a public partnership since 2017. Their cooperation signals a renewed willingness to work together in AI innovation despite past tensions, including fallout from Apple’s 2021 privacy changes that impacted Meta's revenues.
Running the Muse Spark Open model locally on Mac hardware enhances user privacy by keeping sensitive prompts and data off Meta’s servers. This approach also advances desktop AI capabilities, allowing enterprises and developers more direct and efficient access to large-scale machine learning models without reliance on cloud infrastructures.
What to watch next
The cost and hardware requirements for running Muse Spark Open remain steep, with base Mac Studio setups exceeding $10,000, and higher-memory variants arriving soon. Observers should monitor whether this partnership leads to broader consumer or commercial deployments or shapes future hardware and AI software integration strategies.
It remains unclear if Apple and Meta will build on this renewed collaboration or if it will be a one-off project. Industry watchers will also be attentive to announcements regarding enhancements to the MLX framework and whether this local AI execution model expands to other devices and AI applications.