Apple launches M6 and M5 Ultra chips in Mac Studio and Mac mini, targeting AI developers

Reviewed byNidhi Govil

33 Sources

Share

Apple unveiled refreshed Mac Studio and Mac mini desktop computers powered by new M6 and M5 Ultra chips, optimizing for local AI inference with massive unified memory bandwidth. The M6 debuts as Apple's first 2nm chip while the M5 Ultra becomes the company's most powerful processor ever, featuring quad-die architecture with up to 512GB memory capacity.

Apple Introduces M6 and M5 Ultra Chips for Mac Studio and Mac mini

Apple announced updated Mac Studio and Mac mini

1

2

desktop computers on Tuesday, introducing two new Apple M-series chips: the M6 and M5 Ultra. These refreshes represent a significant push toward local AI inference and on-device AI processing, addressing growing demand from developers and AI enthusiasts who have been daisy-chaining these machines to run large language models locally.

Source: Digit

Source: Digit

The Mac mini now starts at $899 with the M6 chip

4

, marking a $200 price increase from previous models, while Mac Studio configurations begin at $2,499 with the M5 Max and $5,499 with the M5 Ultra

5

. Both machines are available for pre-order now and ship September 22 with macOS 27 Golden Gate preinstalled, though the highest-end M5 Ultra configuration with 512GB memory won't arrive until late October.

M6 Chip Debuts with 2nm Process and Enhanced Neural Engine

The M6 chip represents Apple's first 2nm process

3

chip in the M-series lineup for Mac computers. Built with a 12-core CPU complex featuring two "super cores," four performance cores, and six efficiency cores, the M6 is the first Apple system-on-a-chip to use all three core types simultaneously. Apple claims the M6 delivers 20 percent faster multithreaded performance than the M5 and 40 percent faster than the M4

1

.

Source: Wccftech

Source: Wccftech

The chip includes a 12-core GPU, adding two more GPU cores than its predecessors, and features a dual 16-core Neural Engine that doubles peak AI performance

2

. Unified memory bandwidth reaches up to 170GB per second, though memory capacity maxes out at 32GB. According to Apple, the M6 delivers nearly 30 percent increase in peak GPU compute for AI workloads compared to the M5, significantly improving prompt processing speeds when using on-device large language models.

M5 Ultra Becomes Apple's Most Powerful Chip Ever

Apple describes the M5 Ultra as "the most powerful chip ever"

2

, featuring a groundbreaking quad-die architecture. The chip fuses two dual-die M5 Max chips using UltraFusion technology, achieving inter-die bandwidth exceeding 4.4TB per second

3

. This configuration delivers up to 36 CPU cores (12 super cores and 24 performance cores) and 80 GPU cores, with unified memory bandwidth reaching 1.2TB per second—50 percent more than the M3 Ultra released just 18 months earlier.

Source: MacRumors

Source: MacRumors

The M5 Ultra supports up to 512GB of unified memory, a specification clearly targeting AI workloads rather than typical consumer needs. Apple claims the chip offers 4.3x faster AI compute than previous generations

5

, alongside 1.8x faster graphics performance and 1.3x faster CPU speed compared to the M3 Ultra. Storage speeds have doubled to 15GB/s across both Mac mini and Mac Studio models.

Optimizing for Local AI Inference and Developer Workflows

These refreshes directly address the growing popularity of Mac Studio and Mac mini for local AI inference, a use case that gained momentum after macOS 26.2 shipped in December. That update enabled "low-latency communication between Thunderbolt 5 hosts for use cases including distributed AI inference using MLX"

1

, Apple's open source array framework designed to optimize machine learning workflows for M-series chips' unified memory architecture.

Developers and researchers have been connecting multiple Mac minis or Mac Studios to run inference on local large language models that exceed the capacity of single devices, providing alternatives to specialized hardware with Nvidia GPUs. During briefings, Apple discussed connecting up to four Mac Studios over Thunderbolt 5 to run frontier-level LLMs with over a trillion parameters locally

3

, allowing the systems to share memory while running on a standard wall outlet.

Many developers are shifting workflows to incorporate AI coding agents more heavily, but cloud model costs remain steep. Open-weight models like recent Qwen and DeepSeek releases can accomplish similar tasks locally without token charges

1

. However, most consumer hardware still lags behind in model size capacity, making these chained Mac configurations increasingly attractive for AI-ready development environments.

Additional Specifications and Availability

Both machines include Apple's N1 chip supporting Wi-Fi 7 and Bluetooth 6

1

. The Mac mini ships with 2.5Gb Ethernet as standard, with optional 10Gb upgrades. The M6 Mac mini starts at $899 with 16GB memory, while M5 Pro configurations begin at $1,699

4

. Mac Studio with M5 Max starts at $2,499, and M5 Ultra configurations start at $5,499, though prices escalate significantly with higher specifications.

These updates arrive after months of limited availability for both desktop computers, which have been difficult to purchase throughout 2026

3

. The Mac minis have grown particularly popular among tech enthusiasts for running local AI agents, with units selling at marked-up prices on platforms like eBay due to rising demand

4

.

Today's Top Stories

© 2026 TheOutpost.AI All rights reserved