Meta and Apple just buried the hatchet — 816B-parameter AI is coming to your desktop

What happened#

Meta released Muse Spark Open today — which it calls the most capable U.S. open-weight AI model to date — and announced it worked directly with Apple to make the model run locally on the desktop. Not in a data center. On a desk.

The specs are serious:

Spec	Muse Spark Open
Total parameters	816 billion
Active parameters	42 billion (mixture-of-experts)
Context window	1 million tokens
Modalities	Text, image, video, audio (native)
Built for	Coding + long-horizon agentic work
License	Free download, limited commercial use

The trick is Apple's unified memory architecture. Each M5 Ultra chip lets its GPU address up to 512GB of unified memory at 1.2 TB/s of bandwidth — so two Mac Studios behave, in Meta's words, "like one supercomputer with a terabyte of unified memory." Apple's MLX framework distributes the workload across the machines over Thunderbolt 5 with RDMA. You need either two 512GB Mac Studios or four 256GB ones.

The catch: it costs as much as a car#

This is frontier-scale AI on a desk — but the desk is expensive. A single 256GB Mac Studio starts at $10,799, the 512GB version ships later this month, and a four-machine configuration runs past $40,000. The weights are free on Hugging Face (under Meta's Muse Model License); the hardware very much is not. API access is also available through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.

Why this is a big deal#

Apple and Meta have spent years as public enemies — privacy, advertising, app-store policy. Today, Mark Zuckerberg publicly thanked Apple CEO John Ternus, who posted back congratulating the team. Meta's AI chief Alexandr Wang called it "an exciting moment for both companies."

The strategic read: this is the open-weight race going local. Chinese open models (Qwen, DeepSeek, Kimi) still lead on some coding benchmarks — Meta acknowledged as much — and the American answer is now a model you can run in your office without renting cloud compute for every query. Apple gets a marquee AI showcase for its newest silicon; Meta gets a flagship consumer stage for its open models.

Our take#

Two things are true at once. First, the engineering is genuinely impressive: an 816B-parameter model running on desktop hardware would have been absurd five years ago. Second, the price tag means this is a developer and enterprise story today, not a consumer one. The signal for investors: the local-AI stack — memory-rich silicon plus open weights — is becoming a real product category, and the two companies best positioned to own it just decided to share it.

This is a news summary, not financial advice. Nothing here is a recommendation to buy, sell, or hold anything.