Apple and Meta — two of the biggest data-harvesting companies on earth — are joining forces on artificial intelligence, and the pitch to ordinary Americans is a tradeoff: you can run a powerful AI model privately on your own hardware, but you'll need north of $20,000 in Mac Studios to do it.
Meta unveiled Muse Spark Open on Thursday, an 816-billion-parameter open-weight model that the company calls the "most capable US open-weight model to date." Apple optimized the model to run on its Mac Studio with the M5 Ultra chip, using its MLX array framework to split the model across two or four machines. The model runs locally — no cloud, no API calls, no middleman harvesting your prompts. That is a real privacy win, and it's the kind of computing sovereignty the founders would have appreciated.
But the price of admission is steep. A Mac Studio with the required M5 Ultra chip and 256GB of memory starts at $10,799, according to CNBC. You need at least two of them to hold the entire model. A 512GB version arrives later this month at an even higher price. This is not a tool for the common man — it's a rig for developers and well-funded shops.
The collaboration itself is a seismic shift in Silicon Valley politics. Apple and Meta have despised each other for years. Apple's 2021 privacy changes kneecapped Meta's ad-targeting business; Meta estimated the move would cost it $10 billion in revenue that year. CNBC notes that John Ternus taking the reins at Apple appears to have thawed the relationship. Both companies declined to say whether this is the start of something deeper. Translation: there are almost certainly more deals coming.
Meta AI chief Alexandr Wang said the partnership brings "leading agentic capabilities to your desktop for the first time on the Apple M5 Ultra Mac Studios." Meta's blog post touts that two Mac Studios "behave like one supercomputer with a terabyte of unified memory." Apple published new benchmarks and technical documentation on MLX's ability to split open-source models across machines, and emphasized that local execution lets developers "experiment with the latest models and methods in a private environment."
The open-weight angle matters beyond Silicon Valley. CNBC reports that open-source AI has become a flashpoint in Washington, where politicians worry that free Chinese models could undercut American AI. The bipartisan instinct, naturally, is to control the flow — which is exactly the wrong instinct if you believe in free speech and open access to technology. Open-weight models like Muse Spark Open, downloadable from Hugging Face without extra configuration, are the counterweight to the closed oligopoly of OpenAI and Anthropic. They also challenge Nvidia's dominance in AI chips, since Apple's silicon is proving it can handle serious model workloads outside the data center.
The open question is whether two companies built on surveillance can be trusted stewards of a technology that promises local, private computing. Apple brands itself a privacy company; Meta built an empire on tracking everything you do. That they've found common ground here is either a sign of genuine alignment on open-source principles — or simply a marriage of convenience against shared rivals. The model runs on your desk, not their servers. That part is verifiable. The rest you'll have to watch.
