Apple has taken its most aggressive step yet into the local AI compute market, unveiling the M6 chip in a redesigned Mac mini and the M5 Ultra in a refreshed Mac Studio. Announced on August 25, 2026, the releases pair Apple's first 2-nanometer processor with its first quad-die architecture — a combination the company says is designed to run large AI models entirely on device.
The launch positions Apple squarely in the fast-growing market for desktop machines that can run, fine-tune, and serve AI models without sending data to the cloud — a category that has exploded as open-weight models have grown more capable. For more context on this story, see our ongoing more AI stories.
M6: Apple's First 2nm Chip
The M6, which powers the new Mac mini, is built on state-of-the-art 2nm process technology and advances nearly every compute block in the SoC. It features a new 12-core CPU complex — two more cores than the M5 — composed of 2 super cores, 4 performance cores, and 6 efficiency cores. Apple claims the world's fastest single-threaded CPU performance, up to 1.2x faster multithreaded performance than the M5, and up to 2.4x faster than the M1.
For AI workloads specifically, the M6 introduces a Dual 16-core Neural Engine delivering up to 2x the peak compute of previous generations, alongside a larger 12-core GPU with a Neural Accelerator in each core. According to Apple, that design delivers a nearly 30 percent increase in peak GPU compute for AI compared to the M5, and more than 8x compared to the M1.
The chip supports up to 32GB of unified memory and up to 170GB/s of memory bandwidth — a 10 percent increase over the M5 and a 2.5x increase over the M1. Apple says system frameworks can automatically use both Neural Engines simultaneously, so applications see faster on-device model execution for everything from coding agents to image editing.
M5 Ultra: A Quad-Die First for Apple Silicon
The Mac Studio's M5 Ultra is Apple's most powerful chip to date and the company's first quad-die architecture. It uses next-generation UltraFusion technology to connect two dual-die M5 Max chips into a single unified processor, with inter-die bandwidth exceeding 4.4TB/s and connection density more than 6x higher than before.
The spec sheet reads like a direct answer to AI workstation demands: an up-to-36-core CPU, an up-to-80-core GPU with a Neural Accelerator in every core, and a 32-core Neural Engine. Apple rates the GPU at up to 4.5x the peak AI compute of the M3 Ultra and more than 6x the M1 Ultra, with up to 40 percent faster graphics performance than the M3 Ultra thanks to second-generation Dynamic Caching, hardware-accelerated mesh shading, and third-generation ray tracing.
The headline number for AI developers is memory. With up to 512GB of unified memory and 1.2TB/s of bandwidth — 50 percent more than the M3 Ultra — Apple says the Mac Studio can store huge datasets entirely in local memory and "run huge LLMs with hundreds of billions of parameters entirely on device."
"Today, we're debuting the next giant leap in performance and AI compute for Apple silicon with the incredibly advanced M6 and the most powerful M-series chip yet, M5 Ultra," said Sri Santhanam, Apple's vice president of Silicon Engineering Group, in the announcement.
Pricing and Availability
As TechCrunch reported, the new M6 Mac mini starts at $899 with 16GB of RAM and 256GB of storage, with pre-orders open now and shipping beginning September 22 alongside macOS 27 and the upgraded Siri AI. Apple says the new Mac mini's AI performance is up to 4x better than the M4 model, with storage and graphics twice as fast. A second configuration pairs the Mac mini with an M5 Pro chip, starting at $1,699 with 24GB of RAM and 512GB of storage, a 15-core CPU and 16-core GPU at base, upgradeable to 18 CPU cores and 20 GPU cores. Both models include Wi-Fi 7, Bluetooth 6, and a 2.5Gb Ethernet port with a 10Gb option.
The Mac Studio line now starts at $2,499 for an M5 Max configuration with 36GB of unified memory and 512GB of storage, while the M5 Ultra model starts at $5,499 with 96GB of unified memory and 1TB of storage — expandable up to that 512GB ceiling.
A Bet on Local AI Demand
The timing reflects a real shift in how developers work. As TechCrunch noted, Mac minis have grown popular over the past year among enthusiasts running local AI agents, with machines reselling at marked-up prices on eBay earlier this year due to demand. Apple's developer story leans into that trend: frameworks including Core AI, Core ML, Metal, and Xcode tap directly into the new hardware, letting developers run and fine-tune large models locally, use Apple Foundation Models, or plug in their own proprietary models.
The company demonstrated the point with third-party tools: generating AI images locally in Draw Things on the M5 Ultra, and running local AI models in LM Studio Bionic to trigger complex MATLAB simulations on the Mac Studio.
Apple discontinued its $599 entry Mac mini earlier this year, and until now the $799 M4 model was the base of the line — a sign the company sees AI-capable hardware, not budget pricing, as the franchise's future.
Why It Matters
The M6 and M5 Ultra land amid intensifying competition in local AI compute. Nvidia is pushing its DGX Spark desktop for agentic workloads, chip rivals are racing toward 2nm-class processes, and open-weight models keep raising the memory requirements for serious local inference. A desktop that can hold a massive model entirely in unified memory — with no token bills and no data leaving the device — is quickly becoming the baseline expectation for AI developers rather than a niche luxury.
For Apple, the announcements also mark a philosophical statement: that the iPhone maker intends to compete on raw AI silicon at the high end, not just on efficiency and integration. The first benchmark results and hands-on reviews will arrive when the machines ship on September 22, but on paper, the M5 Ultra's 512GB memory ceiling puts Apple in territory once reserved for workstation-class GPU rigs.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
