IMPLICATOR
Apple and Meta, rivals that have fought over privacy for years, worked together for months to run Meta’s new open-weight Muse Spark Open locally on M5 Ultra Mac Studio desktops, announced Oct. 15. Apple updated its MLX software so two linked machines with 512GB of memory each, or four with 256GB, can hold the model. The companies had struck no public partnership since Apple removed system-level Facebook integration from the iPhone in 2017.
What Changed
- Apple and Meta worked together for several months so Meta's open-weight Muse Spark Open runs locally on linked M5 Ultra Mac Studios, announced Oct. 15.
- The companies had struck no public partnership since Apple removed system-level Facebook integration from the iPhone in 2017. Meta estimated in early 2022 that Apple's 2021 privacy changes would cut its revenue that year by $10 billion.
- Apple's own benchmarks recorded four Mac Studios generating 36.73 tokens per second. No independent measurement was available at launch.
- The two-machine setup needs 512GB Mac Studios that have not shipped. The 256GB model costs at least $10,799.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
How the collaboration began
Meta AI chief Alexandr Wang credited Daniel Gross, a former Apple director now on Meta’s AI team, with bringing the companies together over several months.
“We’re excited for mommy and daddy to be playing nice again,” Wang said. Meta bought hundreds of Macs earlier this year for its Superintelligence team to experiment with local AI and agents.
Mark Zuckerberg wrote on X that the model “runs great on new Mac Studios thanks to work with @johnternus and Apple.”
Apple CEO John Ternus replied: “Congrats to the team! Amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk.”
Ternus succeeded Tim Cook last month; Cook is now executive chairman.
FREE · ABOUT FIVE MINUTES
Follow the deals that put AI on your own desk.
Implicator cuts the flood to the developments worth your attention, with concise reporting on what changed and why it matters. Delivered every weekday at 4:45 a.m. Pacific, 7:45 a.m. Eastern.
From San Francisco. No spam. Unsubscribe anytime.
Years of privacy disputes
During the Cambridge Analytica scandal in 2018, Cook was asked what he would do if he were Zuckerberg. “I wouldn’t be in this situation,” Cook said. Zuckerberg called the comments “extremely glib.”
Apple’s 2021 privacy changes limited digital advertising targeting. In early 2022, Meta estimated they would cut its revenue that year by $10 billion. In January 2025, Zuckerberg told Joe Rogan that Apple “hasn’t invented anything great since the iPhone.”
What the companies get
Because the model runs on developers’ own machines, Meta does not receive their prompts. “Meta has no control or visibility to any of that,” Wang said.
Apple published its own benchmarks on launch day, Oct. 15. The machines communicate over Thunderbolt connections, with MLX distributing the model and its calculations across the linked desktops. Apple recorded four Mac Studios generating 36.73 tokens per second at an 8,192-token context.
The speed measurements covered short runs, generating 128 tokens per run. At a 65,536-token context, the four-desktop setup took 40.6 seconds to produce the first token. No independent measurement of the setup was available at launch.
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
Hardware costs
The Muse agent and Meta’s top Muse Wonder models still run only in the cloud. Muse Spark Open’s 4-bit weights alone occupy about 493GB, leaving insufficient operating memory on one machine.
The two-machine setup requires 512GB Mac Studios that have not shipped. They are due later this month, without an announced price. The available 256GB model costs at least $10,799.
Macworld writer Jason Cross expects the 512GB Mac Studios to cost at least $15,000 each, putting the two-desktop requirement at “over $30,000 of Mac Studio hardware.” He wrote: “For all but a small subset of researchers, you’d be much better off paying to run it on a cloud server.”
Representatives of both companies declined to say whether this starts a deeper relationship. Neither has said whether Meta models will reach Apple Intelligence.
“Where it makes sense to collaborate and partner, we want to be able to do so,” Wang said.
Frequently Asked Questions
What did Apple and Meta announce?
On Oct. 15, Meta released Muse Spark Open, an open-weight model it worked on with Apple so it runs locally on linked M5 Ultra Mac Studios. Apple updated its MLX software so two machines with 512GB of memory each, or four with 256GB, can hold the model.
Why is the partnership unusual?
The companies have fought over privacy for years. Apple's 2021 privacy changes limited ad targeting, and Meta estimated in early 2022 that they would cut its revenue that year by $10 billion. The two had struck no public partnership since Apple removed system-level Facebook integration from the iPhone in 2017.
Who brought the companies together?
Meta AI chief Alexandr Wang credited Daniel Gross, a former Apple director now on Meta's AI team, with bringing the companies together over several months. Representatives of both companies declined to say whether this starts a deeper relationship.
How fast does the model run on Mac Studios?
Apple's own benchmarks recorded four Mac Studios generating 36.73 tokens per second at an 8,192-token context. At a 65,536-token context, the four-machine setup took 40.6 seconds to produce the first token. No independent measurement was available at launch.
What does the hardware cost?
The two-machine setup needs 512GB Mac Studios that have not shipped and have no announced price. The 256GB model costs at least $10,799. Jason Cross of Macworld expects the 512GB machines to cost at least $15,000 each, putting two of them above $30,000.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Free AI briefing · Weekdays
The AI stories that matter, sourced and explained.
Join the Morning Briefing. It goes out every weekday at 4:45 a.m. Pacific, with later sends for the East Coast, Berlin and Tokyo.
Free when you sign up: The Paperclip Compendium, our tested guide to running AI agents.
Free. Unsubscribe in one click.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
