Meta and Apple, Longtime Foes, Team Up to Put Open-Weight AI on Macs

Mark Zuckerberg has battled Apple AAPL 0.43%increase; green up pointing triangle ever since the early days of the iPhone. Now AI has brought the two fierce rivals together.

Meta META 0.36%increase; green up pointing triangle Platforms announced Thursday that it’s releasing a new open-weight AI model and collaborating with Apple to run it locally on the company’s high-end Mac computers. The tech giants’ partnership gives developers another American-made AI option to power research projects, app development, video production and more.

“This is an exciting moment for both companies, where we’re finding ways to work together that are productive,” Meta’s AI chief, Alexandr Wang, said in an interview. “We think that a lot of the world will run on open models, and they should be American.”

As part of the collaboration, Apple updated its MLX software toolkit to enable Meta’s model run well on Macs.

Open-weight models have their core components released to the public, meaning anyone can download them. Many popular, low-cost Chinese models are open-weight, including Alibaba’s Qwen, DeepSeek and Moonshot’s Kimi K3. Other American open-weight offerings include Inkling from Thinking Machines and Nemotron from Nvidia, which released three new models on Thursday.

Meta said its new model, called Muse Spark Open, was built for coding and agentic work but acknowledged that Chinese open models are still superior for software development and some other tasks. It’s the second open-weight model Meta has released since the company overhauled its AI efforts last year.

Apple’s Mac Mini and Mac Studio computers have proven popular among AI developers looking to run sophisticated models locally in their home or office. Apple’s computers are highly efficient, especially in how they use computer memory necessary to run models.

The Meta-Apple collaboration marks a new chapter for two companies that have long been foes.

Meta’s apps run on iPhones, giving Apple power over how Meta operates—leverage it used in 2021 to limit the company’s ability to track people for ad targeting. At the time, Meta publicly said the policy changes would cost it more than $10 billion in lost sales in 2022. When Apple’s former chief executive, Tim Cook, made comments critical of Meta in 2018, Zuckerberg said privately that he wanted his team to “inflict pain” on Apple, The Wall Street Journal previously reported.

Macs are popular for running AI models because they pack lots of specialized computer memory that is shared between chips inside the machine. Multiple macs can be linked together to share AI workloads with the help of Apple’s MLX software.

Muse Spark Open’s heavy memory demands make it “a poor fit for most hardware,” Wang said, but well suited to run on Macs.

Customers will need to buy Macs collectively packing at least 1 terabyte of memory to run the new model. The total cost is now upwards of $40,000.

Even at that price, the computers could pay for themselves for app developers and others who otherwise would have to rent a similar amount of computing power from cloud providers, which can cost thousands of dollars per month.