
Photo by Huy Phan on Pexels
In a plot twist that rivals the most dramatic Silicon Valley soap operas, arch-rivals Meta and Apple have found something to agree on: open-weight artificial intelligence is actually pretty cool, even if it means sleeping with the enemy.
Meta dropped a fresh heavyweight into the ring on Thursday: Muse Spark Open. Coming on the heels of the somewhat tragic, universally side-eyed “dud” that was Llama 4, Meta desperately needed a redemption arc. Enter Superintelligence Labs and a model sporting 816 billion parameters and a 1-million-token context window. It’s officially the U.S.’s most capable open-weight model.
Instead of renting cloudy server space from OpenAI or Anthropic, companies can now download Muse Spark Open and run it on their own iron. You want to feed it confidential medical records or unreleased movie footage without handing your secrets to a third party? Go right ahead. You can gorge on tokens until your hardware begs for mercy, completely free of per-token API fees.
The Mac Studio Mini-Data Center
The real kicker isn’t just that Meta built a massive open model—it’s where it was demoed. During a media briefing, the magic happened on a pair of Apple Mac Studio computers cabled together into a tidy, desk-friendly cluster.
Historically, Meta and Apple treat each other with the warmth of rival medieval kings. Apple’s privacy crusades have long acted as an industrial-grade wrecking ball against Meta’s targeted-ad business model. Yet, local AI has forged an unexpected alliance.
Apple’s high-end silicon—featuring Neural Engines baked right into the M1 Max and Ultra chips—wasn’t originally designed with autonomous software-writing LLMs in mind back in 2022. But Apple leaned into the shift. Today, AI enthusiasts are snapping up Mac Minis for agents like OpenClaw, while industrial-strength setups daisy-chain a couple of Mac Studios together with 1 TB of RAM. It costs tens of thousands of dollars, sure, but it fits neatly next to your stapler instead of requiring a loud, dedicated server rack.
Apple’s Clever Play
Apple didn’t build a frontier, general-purpose LLM of its own. While I’ll happily take a jab at Apple’s walled garden when it comes to certain things (seriously, give me back my wired earbuds and a physical headphone jack without dongle gymnastics), you have to admit their hardware engineering is elite—and I say this as someone who owns several iPhones alongside my Android gear.
By teaming up with Meta and optimizing tools like their MLX software framework for Mac Studio clusters, Apple has pulled off a masterclass. They don’t need to win the frontier lab race against Google or OpenAI. If developers and enterprises use Apple silicon to run heavy-duty open-weight models like Muse Spark Open (or even China’s Qwen), Apple wins by simply selling the hardware.
Pundits spent years crying that Apple was “behind in AI.” Turns out, they were just waiting for everyone else to build the software so they could sell the ultimate machines to run it. Outfoxed again.
Source reference: fastcompany
Disclaimer: This article was generated and formatted with the assistance of Artificial Intelligence.