Meta released Muse Spark Open, an 816 billion-parameter model it calls the most capable U.S. open-weight model to date, and said it worked with Apple to tune the model for two Mac Studio desktops. Apple published updated software and benchmarks the same day. Meta said the model and Apple’s desktops “together” are “the best way to run a powerful model on your own desktop.”
Their relationship wasn’t always this friendly. In 2021, Apple began requiring apps on iPhones to ask permission before tracking users across other apps and websites, which cut off much of the data Meta used to target and measure ads. Meta said the change would cost it about $10 billion in 2022, CNBC reported at the time. Apple spent the same years marketing privacy as a reason to buy its devices, and the two companies’ chief executives traded public criticism over whose business model was better for users.
A Frontier-Class Model Now Fits on a Desk
Models this size normally live in data centers. Apple’s team showed that two Mac Studios wired together can hold Muse Spark Open and answer at a usable speed, and that four of them run it faster than most people can read. Each machine plugs into a standard wall outlet. Daniel Gross, a former Apple executive now at Meta, helped broker the collaboration, Axios reported.
Each side gets something concrete. Meta gets developers building on its model instead of on OpenAI’s, Anthropic’s or China’s open models, and it can point to Apple’s hardware as proof the model works outside its own data centers. Apple gets a showcase for its most expensive desktop and an AI story that fits its privacy pitch, since it never hosts the model or touches the data, The Wall Street Journal reported. Keeping prompts and files on the user’s own machine also removes the objection that divided the companies for years.
Microsoft Ships a Laptop Built for the Same Job
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Apple isn’t alone in treating the desk as the new home for AI. Microsoft’s Surface Laptop Ultra ships today starting at $2,599, built with an Nvidia chip and enough memory to run large models with no cloud connection, Microsoft said when it opened preorders Oct. 7.
Microsoft also claimed the machine outperforms Apple’s MacBooks on AI work, Bloomberg reported.
Satya Nadella, Microsoft’s CEO, framed the launch as bringing “unmetered intelligence to every desk and every home.” Copilot on Windows can now use models on the PC itself for everyday tasks and send harder ones to the cloud, Nadella said. Meta is also bringing its Muse personal agent to Windows, Axios reported.
Businesses Weigh Owning AI Instead of Renting It
The pitch lands with companies already unhappy with their AI bills. Legal AI company Harvey, healthcare startup Abridge and FinTech Ramp are building on or weighing open-weight models after usage-based pricing from OpenAI and Anthropic squeezed margins. AT&T cut the cost of coding and other advanced AI tasks by as much as 56% by routing questions to cheaper models.
PNC Financial Services, CH Robinson, Siemens and Tinder are also considering using open-weight models to reduce AI bills, PYMNTS noted.
Running those models on a company’s own machines is the next step, and Apple is selling into it. Its Mac Studio pitch promises users can run large models “without counting tokens or worrying about rising cloud costs,” the company said when it introduced the M5 Ultra in August. For a bank or hospital, that also means customer data never leaves the building.
For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.
Add as Preferred Source
