Apple and Meta announced a collaboration on October 15, 2026, to help developers run Meta’s Muse Spark Open AI model locally on high-end Mac Studio computers. The partnership pairs Meta’s open-weight model with Apple’s MLX software and M5 Ultra hardware, but it does not disclose a broader commercial agreement or prove specific security or cost savings.
1
3
What Apple and Meta announced
Muse Spark Open is an 816-billion-parameter model built with a mixture-of-experts architecture. Reports describe 42 billion parameters as active for each token, alongside a one-million-token context window. Meta released the model’s weights under a license with limited commercial terms, so “open-weight” should not be read as meaning unrestricted use.
1
11
16
The companies’ work is intended to let developers run the model on linked Mac Studios using Apple’s MLX framework. Reports describe two-machine setups as well as configurations of up to four computers. The model is aimed at coding and agentic tasks; coverage also mentions video analysis among potential uses.
1
3
4
The setup is substantial. One report puts the model’s compressed NVFP4 weights at about 493GB, before additional memory needed during inference. Apple’s reported tests used multiple Mac Studios, and the model did not fit on a single machine in that tested configuration.
3
8
A reported starting price for an M5 Ultra Mac Studio with 256GB of memory is $10,799. That is the price of one computer, not a quoted price for a complete multi-machine setup; the reported model-running configuration uses high-memory machines.
6
33
What local execution does—and does not—mean
Running a model locally means its computation can take place on the user’s own computers rather than relying only on remote servers. That may appeal to developers who want to work directly with model weights and local hardware. The announcement and reporting describe local operation, but the available evidence does not establish a quantified security advantage, guarantee that sensitive data cannot leave a device, or show how local running compares in total cost with cloud use.
1
3
Likewise, the reported footprint and multi-machine setup help explain the hardware requirements, but they are not enough on their own to establish a particular speed or performance advantage. Benchmarks depend on the configuration and measurement conditions; the materials available here do not provide a sufficiently clear basis to treat a specific tokens-per-second figure as a general result.
A notable collaboration between longtime rivals
The companies’ leaders publicly welcomed the work. Mark Zuckerberg credited Apple and CEO John Ternus with helping Muse Spark Open run on Mac Studio, while Ternus congratulated the team. Reporting describes the engineering effort as months in the making.
10
34
The announcement stands out because coverage described it as Apple and Meta’s first public collaboration since 2017, after years of tension between the companies. That history makes the cooperation notable, but it does not establish that they have agreed to a wider partnership.
10
The public reports reviewed here do not specify financial terms, exclusivity, duration, or an expansion plan. They also do not substantiate the often-repeated claim that Apple’s 2021 privacy changes cost Meta $10 billion in revenue the following year, so that figure should not be treated as established by this announcement.
Dan Ives’s $75-per-share thesis is a forecast
On CNBC’s “Fast Money,” Yorkville Ives analyst Dan Ives argued that Apple and Meta need each other and suggested the announcement could be the beginning of broader cooperation. Coverage of his view says he estimated $75 per Apple share in incremental AI value and cast Apple as a potential “toll collector” for consumer AI.
18
23
51
That is an analyst’s interpretation of possible future value—not a disclosed payment, guaranteed stock gain, or commitment from either company. The confirmed announcement is narrower: Apple and Meta worked together to make Meta’s model run locally on Mac Studio hardware. Whether that engineering collaboration leads to more products or a commercial arrangement remains unknown.
1
23