Apple and Meta's AI alliance flips their long feud

The unlikely alliance

Apple and Meta spent years publicly sparring. Tim Cook called out Meta over privacy in 2018. Mark Zuckerberg told Joe Rogan Apple had not invented anything great since the iPhone. Apple’s 2021 privacy changes were estimated to cost Meta $10 billion in a single year. The two companies went nearly a decade without a public partnership.

Now they are working together on AI.

The partnership, announced Thursday, lets developers run Meta’s Muse Spark Open model directly on Apple’s high-end Mac Studio computers — no cloud required. It is a genuinely novel move. Two rivals with fundamentally different business models have found common ground in open-weight local AI, and the implications reach far beyond a single software release.

Why this matters more than it looks

The most interesting thing about this deal is not that Apple and Meta are cooperating. It is what the cooperation reveals about where the AI industry is actually heading.

For years, the dominant narrative has been that AI lives in the cloud. OpenAI, Google, Anthropic — all of them built empires on centralized compute. Meta pivoted toward that same model under Alexandr Wang, who joined after Meta invested more than $14 billion in his former company, Scale AI. Wang’s Muse Wonder, released earlier this week, is proprietary. It is built for scale, for enterprise, for the kind of agentic work that requires constant connectivity.

But Muse Spark Open is different. It is open-weight, meaning developers can download its parameters and run it locally. And with Apple’s involvement, it runs on hardware most people cannot afford — an M5 Ultra Mac Studio with 256GB of memory starts at $10,799, and Meta’s developers already bought hundreds of Macs just to experiment with local AI earlier this year.

This is a bet on a different AI future: one where powerful models live on your desk, not in someone else’s data center.

Who wins, who loses

Apple wins the most obvious way. The partnership is a showcase for its silicon. Nvidia has dominated the AI chip conversation for years because data centers run on its hardware. Apple is now proving that its M5 Ultra can handle models with 816 billion parameters — a remarkable feat for a chip designed for consumer desktops. The new benchmarks and MLX framework documentation Apple published alongside the announcement are not marketing copy. They are arguments in a much larger competition about where the next wave of AI inference will happen.

Meta wins in a quieter way. Wang acknowledged that Muse Spark Open is not the most advanced model his team has built. It is smaller than Muse Wonder. But open-weight releases like this create ecosystems. Developers tinker with it at small scale, then scale up — and when they do, they are more likely to reach for Meta’s proprietary offerings. It is a funnel strategy disguised as openness.

OpenAI and Anthropic lose relative ground. Their models live in the cloud by design. They cannot run locally on a Mac Studio. If the industry shifts even partially toward on-device AI — and Apple’s track record with Mac Pro and Studio machines suggests it believes the shift is coming — then the incumbents who bet everything on cloud inference face a structural headwind.

The China angle nobody is talking about

There is a layer to this partnership that goes unmentioned in the press releases. Washington has grown increasingly anxious about Chinese AI models flooding the open-source market at low or zero cost. American politicians worry that free models from China could undercut domestic competition and erode US leadership.

Apple and Meta’s collaboration is an implicit answer to that concern. By making the most capable open-weight model viable on American hardware, they are drawing a line: powerful open AI belongs on US silicon, not on whatever commodity rig a developer in Shenzhen is running. It is not stated that way, but the geopolitical subtext is impossible to miss.

The price of entry

The Mac Studio is not a mass-market product. Even the base configuration at $10,799 places this partnership firmly in the realm of professional developers and well-funded startups. Meta’s own engineers bought hundreds of these machines for experimentation — a signal that they see local AI as a serious frontier, not a gimmick.

Apple’s strategy here is deliberate. It is not trying to put Muse Spark on every MacBook. It is establishing the Mac Studio as the premium platform for on-device AI development. That is a niche, yes, but it is also a defensible one. The developers who cut their teeth running large open-weight models on M5 Ultra chips are the ones who will shape how AI is built going forward.

What comes next

n
Neither company has said whether this marks the beginning of a deeper relationship. Meta declined to comment on whether more collaborations are planned. Apple has not issued a statement beyond the partnership announcement itself.

What is clear is that the AI landscape is fragmenting. Cloud-based models and on-device models are no longer competing for the same users. Meta is pursuing both paths simultaneously — Muse Wonder for the cloud, Muse Spark Open for the edge. Apple is positioning its hardware as the bridge between them.

The rivalry between Cook and Zuckerberg is not over. But the next chapter of Apple-Meta relations may be written not in courtrooms or press conferences, but in the quiet hum of Mac Studios running open-weight models at 2 a.m.

That is a partnership worth watching.