Meta AI chief Alexandr Wang has revealed that former Apple executive Daniel Gross played a key role in bringing the iPhone-maker and Meta together for an AI collaboration. After years of disagreements, the two companies are now working together to run Meta's new Muse Spark Open AI model on Apple's high-end Mac Studio computers. Their collaboration allows Muse Spark Open, which has 816 billion parameters, to run across two or four connected Mac Studios powered by Apple's M5 Ultra chips, using Apple's MLX framework. Alexandr Wang told Axios that Gross, who now works at Meta, helped the two sides work together. Wang revealed that the two companies had been working together for several months.
“I think he played a really big role in kind of getting the two companies to work nicely together,” Alexandr Wang told the publication. “We're excited for mommy and daddy to be playing nice again,” he said.
Tim Cook-Mark Zuckerberg once clashed over privacy
The rivalry between Apple and Meta (then Facebook) dates back to 2018 during the Facebook's Cambridge Analytica scandal. The then Apple CEO Tim Cook was asked what he would do if he were Mark Zuckerberg. Tim Cook then replied saying “I wouldn't be in this situation”.
Later, Mark Zuckerberg called Cook's comments "extremely glib.” Again in 2021, when Apple introduced new privacy protections, Zuckerberg told employees that Facebook needed to "inflict pain" on Apple over the dispute, per a report by the Wall Street Journal then.
Meta's most capable open AI model now runs on Apple's Mac Studio
Muse Spark Open is an 816-billion-parameter model built on a mixture-of-experts design. Meta calls it the most capable US-developed open-weight model to date. It is aimed at long coding sessions, multi-step reasoning and video analysis. The model is free to download from platforms like Hugging Face, and its licence allows commercial use with some limits.
The setup links two or four Mac Studios running Apple's M5 Ultra chip. Apple's MLX framework splits the model across those machines. Apple has also put out new benchmarks and technical notes on how MLX handles open models of this size. Since everything stays on the desk, Meta does not see the prompts or files developers feed it. "Meta has no control or visibility to any of that," Meta AI chief Alexandr Wang said.
It does not come cheap. An M5 Ultra Mac Studio with 256GB of memory costs over Rs 9 lakh at US prices, and the model needs at least two. A 512GB version is due later this month.