Meta's New AI Model Runs on Macs — If You Have 1TB of RAM

Meta and Apple are collaborating on a new open-weight AI model, Muse Spark Open, but running it locally requires an extreme hardware setup that puts it out of reach for nearly all users. According to Engadget, running the massive 816-billion parameter model requires a full terabyte of RAM, a detail that reframes the entire announcement from a consumer-facing development into a niche tool for heavily-funded researchers.

The partnership, announced by Meta, focuses on making its latest and most powerful open-weight model run efficiently on Apple silicon. The Verge reports the collaboration leverages Apple's latest MLX framework, a software library designed for AI on Apple chips. While the promise of running a state-of-the-art model on a desktop computer is compelling, the reality is far from accessible.

A Partnership of Convenience

On the surface, the news is a straightforward collaboration between two tech giants. Meta announced Muse Spark Open, which it claims is the “most capable US open-weight model to date,” according to reports from both The Verge and Engadget. Simultaneously, TechCrunch and CNBC confirmed the partnership with Apple, which aims to make the model easier to run locally. This isn't happening in the cloud; the processing is done on the machines themselves.

The specific configuration, as described by multiple outlets, involves clustering two Mac Studio computers together. CNBC notes the collaboration is focused on making the model run on these “clustered Mac Studio systems.” This setup is a direct showcase for Apple’s unified memory architecture, which allows RAM to be shared efficiently across the system-on-a-chip. It’s a powerful demonstration of what Apple silicon can do at the highest end, but it's a scenario that applies to a tiny fraction of even professional Mac users.

The 1TB Asterisk — 'Desktop AI' for Who, Exactly?

Meta’s marketing suggests a move toward democratizing powerful AI. But the hardware requirements tell a different story. The 1TB RAM figure reported by Engadget is the single most important detail of this launch. It’s a barrier so high that it effectively excludes anyone outside of large corporations or elite research institutions. The cost of two maxed-out Mac Studios pushes well into five figures. This isn't for a developer in their garage.

This reveals the true audience for Muse Spark Open. It is not for consumers, hobbyists, or even most independent developers. It is a tool for the AI elite. The phrase “runs on your desktop” is technically true but practically misleading. It creates an impression of accessibility that simply does not exist. The pattern indicates a strategic positioning rather than a broad product release. Meta gets to plant its flag in the high-end research community, and Apple gets to prove its most expensive hardware is essential for the future of AI development.

The choice to optimize for Apple's MLX framework is also telling. This is Apple's answer to Nvidia's CUDA, the dominant platform for AI training and inference. By working with Meta to ensure a major new model runs well on MLX, Apple is signaling to the entire AI community that Mac hardware is a viable, and perhaps superior, alternative to the custom-built Linux machines that have dominated the field.

Meta’s 'Open' Gambit Meets Apple’s Walled Garden

This collaboration is a fascinating intersection of two opposing philosophies. Meta has aggressively pursued an “open” strategy with its AI models, releasing the weights to the public to foster a community and create a counterweight to the closed, proprietary models from OpenAI and Google. Apple, on the other hand, is famously a closed ecosystem. The two coming together suggests a shared, pragmatic goal: challenging the current leaders in the AI space.

For Apple, this is a low-risk, high-reward move. It doesn’t have to invest billions in creating its own 800-billion-plus parameter model from scratch. Instead, it can partner with Meta to ensure that a leading open model runs best on Apple's most expensive hardware. This makes the Mac Studio an even more compelling purchase for AI research departments, directly tying Apple's hardware sales to the AI boom without having to compete on the foundational model front.

For Meta, the benefit is equally clear. It gets to showcase its model on premium, desirable hardware and tap into Apple's influential developer ecosystem. It’s a powerful endorsement. Together, these reports point to a calculated alliance aimed at carving out a specific niche in the AI arms race. It’s not about bringing AI to every Mac user’s desktop today. It's about establishing the Mac as the premier platform for the developers who will build the AI of tomorrow.