> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Apple builds its new Mac desktops around AI models that run locally
- URL: https://www.metatalks.ai/apple-unveils-the-m6-and-m5-ultra-desktop-chips-built-to-run-ai-models-on-the-machine-itself/
- Published: 2026-08-27T11:41:00.000Z
- Updated: 2026-08-27T11:40:59.000Z
- Author: Al
- Tags: News, On-device AI, Semiconductors, #newswire

**Apple** [announced two desktop chips](https://www.apple.com/newsroom/2026/08/apple-unveils-a-more-powerful-mac-mini-featuring-the-all-new-m6-and-m5-pro/?ref=metatalks.ai) **on August 25, the M6 for a new Mac mini and the M5 Ultra for a new Mac Studio, and built its case for both around models that run locally. The Studio takes up to 512GB of unified memory, and the company says two of the machines can be wired together to pool it.**

The M5 Ultra moves 1.2TB/s of memory bandwidth, half again as much as the M3 Ultra, and Apple says a 512GB pool will hold a model with hundreds of billions of parameters entirely on device. The Studio starts at $5,499 with that chip, up from $5,299 for the M3 Ultra version.

The M6 in the Mac mini is aimed lower and more explicitly at agents: Apple describes its 32GB of unified memory as room to run language models on device for “secure and private agentic tasks.” Peak GPU compute for AI is close to 30 percent above the M5's, and Apple's own benchmark has an M6 mini clearing prompts in LM Studio up to 4.8 times faster than the M4 mini.

The entry mini now costs $899, $100 more, which Apple ties to memory prices. Both machines ship on September 22.

Chaining Macs is not new. macOS 26.2 shipped in December with [low-latency links](https://developer.apple.com/documentation/macos-release-notes/macos-26%5F2-release-notes?ref=metatalks.ai) between Thunderbolt 5 hosts for distributed inference through MLX, Apple's array framework, and developers and researchers have since daisy-chained minis and Studios to run models no single consumer machine can hold — a cheaper route than specialised Nvidia hardware. Wiring two Ultras together extends that to the top of the range.

Neither desktop gains features beyond the chips. Every performance figure so far is Apple's own; no independent benchmarks have been published.