Apple and Meta have spent years fighting over privacy and app tracking. Now, the two companies are working together to bring one of the largest open-weight AI models to the Mac.
Meta’s newly released Muse Spark Open has 816 billion parameters, and running it locally requires two Mac Studios equipped with 512GB of unified memory each.
The companies collaborated to optimize the model for Apple’s MLX framework, letting developers run it on their own hardware instead of relying entirely on cloud servers.
It’s an unusual partnership between two technology giants that have rarely seen eye to eye.
Why Meta’s new AI model needs two Mac Studios
Muse Spark Open is designed for demanding coding and AI agent tasks, including software development projects that require multiple steps of reasoning.
The model uses a mixture-of-experts architecture, activating 42 billion of its 816 billion parameters at a time. It also supports a 1 million-token context window, allowing it to work with exceptionally large amounts of information in a single session.
Running something that large on a desktop computer presents an obvious challenge: memory.
Meta’s solution involves connecting two Mac Studios powered by Apple’s M5 Ultra chip, each configured with 512GB of unified memory.
Together, the machines provide roughly 1TB of memory, enough to hold the model locally. Apple’s unified memory architecture lets the GPU access that memory directly, while the M5 Ultra’s 1.2TB/s memory bandwidth helps move the data needed during inference.
Apple CEO John Ternus celebrated the collaboration on X, calling it “amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk.”
Meta CEO Mark Zuckerberg also thanked Ternus for working with the company.
Running it locally won’t be cheap
There’s a significant catch for developers hoping to recreate the setup. Apple doesn’t currently sell a Mac Studio with 512GB of unified memory. That configuration is expected to arrive later this month.
The existing 256GB model already costs more than $10,000, meaning a two-machine setup with twice the memory could require an investment of tens of thousands of dollars.
Fortunately, owning two Mac Studios isn’t necessary to use Muse Spark Open. Several cloud API providers offer the model, including Groq, OpenRouter, and Together AI.
Developers can also download its weights through platforms such as Hugging Face and Google Cloud Model Garden. The weights are free to download under a license that permits limited commercial use.
Apple and Meta find common ground in AI
The collaboration is particularly notable given the companies’ history. Apple’s privacy changes have repeatedly put it at odds with Meta’s advertising business, while the two companies have pursued competing approaches to personal computing.
Yet the growing demand for powerful AI models that can run outside traditional cloud infrastructure appears to have created an opportunity for cooperation.
Meta’s AI chief Alexandr Wang acknowledged the unusual relationship with a joke, saying the companies were excited to see “mommy and daddy” getting along again.
For Apple, the partnership also highlights a potential selling point for its highest-end Macs: the ability to run enormous AI models locally, provided developers can afford the hardware.
The 512GB Mac Studio configuration is expected to become available before the end of October.
