Apple just shipped a going-away present to itself: a Mac mini fast enough to run AI models without phoning a data center, timed for what may be Tim Cook's last major hardware launch as CEO before he hands the keys to incoming CEO John Ternus on September 1.

Two New Chips, One Very Local Ambition

Apple unveiled a refreshed Mac mini built around two new chips: the M6, with a 12-core CPU and 12-core GPU, and the beefier M5 Pro, scaling up to an 18-core CPU and 20-core GPU. Both chips build Neural Accelerators directly into every GPU core, and Apple says the M6 model delivers up to 4x faster AI performance and 2x faster graphics and storage than the outgoing M4 version, while the M5 Pro claims up to 8.5x faster LLM prompt processing versus the M2 Pro.

Pricing starts at $899 for the M6 model and $1,699 for M5 Pro, with preorders open now and shipping beginning September 22.

The Quiet Pitch: Skip the Cloud Bill Entirely

The headline spec bump matters less than what Apple is positioning the machine to do: run and fine-tune large models entirely on-device, and even cluster multiple Mac minis together over Thunderbolt 5 to split a bigger model across several machines. That's Apple making a direct pitch to developers and small businesses who'd rather own the hardware than rent GPU time by the hour.

It's also Apple quietly sidestepping the "our AI depends on someone else's data center" problem that's been dogging every company bolting a chatbot onto its product. A sub-$900 box that can run inference locally, with no per-token bill and no outage risk from a third-party API, is a genuinely different pitch than "just trust the cloud."

Somewhere, a startup's cloud compute budget just breathed a small sigh of relief.

If you're weighing whether an AI feature belongs on your server, in the cloud, or on a box like this one, that's exactly the kind of AI integration strategy conversation we have with clients every week — reach out and let's map out yours.

Source: CNBC