Meta and Apple Join Forces to Bring Open-Weight AI Models to Mac for Local Execution
nashnova research
Meta released Muse Spark Open, an open-weight AI model, and partnered with Apple to run it locally on Mac Studio — the first public collaboration between two long-standing rivals, driven by an AI race that is redrawing Silicon Valley's alliances.
What does this model do, and what does it take to run it?
Muse Spark Open targets coding and AI-agent tasks. Meta calls it one of the strongest open-weight models developed in the U.S.
Running it requires two Mac Studios with Apple's M5 Ultra chip, linked together with at least 1 TB of combined memory — total hardware cost currently exceeds $40,000.
This means → this is not a consumer product. It is a local-compute solution for professional developers — and Apple's unified memory architecture (CPU and GPU share the same memory pool, eliminating data transfers) happens to be the ideal hardware fit for running large models.
Apple updated its MLX software framework — a machine-learning toolkit built for Apple silicon — specifically to support this model running locally.
Why won't Meta open-source its most powerful model?
Muse Spark Open falls short of Muse Wonder, the closed-source flagship released earlier this week. Meta kept the gap intentional.
Meta's chief AI officer Alexandr Wang said: "For both safety and capability reasons, we generally don't open-source our most capable models."
In plain terms = Meta is running a dual-track strategy: closed models serve its own products and paying customers; open models grab ecosystem share and developers. Two lanes, neither sacrificed.
Wang also stressed the geopolitical angle: "Most AI in the world will run on open models, and they should be American." He added that even if Chinese models borrow design principles from Muse Spark Open, "that's probably a good thing."
Apple and Meta have been rivals for years — why team up now?
The relationship has been hostile. In 2021, Apple's privacy changes hit Meta's ad business hard — Meta estimated the revenue loss at roughly $10 billion that year. Zuckerberg reportedly told staff to "inflict pain" on Apple.
This collaboration was brokered by Daniel Gross, a Meta AI researcher and former Apple director. The two sides had been working together for months.
This reflects a shift: the pressure of the AI race is strong enough to override old grudges. Apple needs strong models to fill out its ecosystem; Meta needs Apple hardware to prove open models can run locally. The intersection of interests outweighed years of hostility.
How does Meta make money from a free model?
Muse Spark Open is a free download, available on Hugging Face and accessible through Amazon SageMaker AI, Google Cloud Model Garden, and several other platforms.
Separately, Meta is building its own cloud business, planning to sell external access to AI models and compute infrastructure.
This means → the free model is a funnel. Developers experiment locally at no cost; when they need more compute, they move to Meta's paid cloud services. Hook them first, charge later.
What to watch next?
The key test: whether Apple integrates Meta's model into Apple Intelligence, giving Meta a core role inside Apple's own AI features.
If that happens, the relationship upgrades from a project-level collaboration to a platform-level tie. If it doesn't, this partnership looks more like a one-off transaction where each side took what it needed.
Meta still faces competition from Chinese open models on certain tasks — Alibaba's Qwen, DeepSeek, and Moonshot's Kimi K3 hold advantages in areas like software development.
市场有风险，内容仅供研究参考，不构成投资建议。