Apple debuts first 2nm M6 and quad-die M5 Ultra in updated Mac mini and Mac Studio

Two new chips, two refreshed desktops

On August 25, 2026, Apple announced the M6, its first chip built on a 2-nanometer process, and the M5 Ultra, its most powerful silicon to date, paired with refreshed Mac mini and Mac Studio machines. Both desktops are available for pre-order immediately and go on sale September 22, according to Apple's newsroom.

M6 brings 2nm fabrication to the desktop

According to Apple, the M6 pairs a larger 12-core CPU — made up of two super cores, four performance cores and six efficiency cores — with a 12-core GPU that places a Neural Accelerator in every core, plus a Dual 16-core Neural Engine that doubles peak compute compared with the previous generation. Apple claims the fastest single-threaded CPU core available, up to 1.2 times the multithreaded throughput of the M5 and up to 2.4 times that of the M1.

The chip supports up to 32GB of unified memory with up to 170GB/s of bandwidth, which Apple says is 10 percent higher than M5 and 2.5 times M1. Peak GPU compute for AI workloads rises roughly 30 percent over M5 and more than 8x over M1.

In the Mac mini, Apple rates the M6 at up to 4x faster AI performance, 2x faster graphics and 40 percent faster CPU performance than the M4 model. One figure Apple highlighted: LLM prompt processing in LM Studio runs up to 13.5x faster than an M1-based mini and up to 4.8x faster than the M4 version. The mini also gains an M5 Pro option with up to an 18-core CPU and 20-core GPU, along with Wi-Fi 7, Bluetooth 6 and 2.5Gb Ethernet standard, with a 10Gb upgrade available.

M5 Ultra goes quad-die

The M5 Ultra is Apple's first quad-die M-series design, connected through a next generation of the company's UltraFusion packaging. It scales to a 36-core CPU and an 80-core GPU that, for the first time on an Ultra chip, includes Neural Accelerators. Memory bandwidth reaches 1.2TB/s — 50 percent more than M3 Ultra, per Apple — and inside the Mac Studio it can be configured with up to 512GB of unified memory, which Apple positions as enough to run very large language models entirely on the machine.

Apple's performance claims for the Studio include up to 4.3x the peak AI compute of M3 Ultra and 9.8x that of M1 Ultra, together with up to 1.8x faster graphics, 2x faster storage and 1.3x faster CPU performance. The enclosure also moves to Thunderbolt 5, Wi-Fi 7 and Bluetooth 6, and an M5 Max configuration offers an 18-core CPU, up to a 40-core GPU and up to 128GB of memory.

Clustering and a new AI software stack

Two details stand out for anyone running models locally. First, Thunderbolt 5 combined with RDMA support lets multiple Mac Studio units be clustered into a shared memory pool; Apple says a group of four machines reaches up to 3x the AI inference speed of a single system, enough to load the largest open-weight models available today. Second, Apple is introducing Core AI, a new framework for building, running and deploying models on Apple silicon, alongside its open-source MLX framework. Both desktops ship with macOS 27 and the next generation of Apple Intelligence, including a Siri AI assistant.

Apple executives framed the launch around local AI workloads. Johny Srouji, Apple's chief hardware officer, called the new Mac Studio the company's most powerful Mac yet, while Sri Santhanam, vice president of the Silicon Engineering group, described the M6 and M5 Ultra as the next major step in performance and AI compute for Apple silicon.

Why it matters

This is nominally a hardware launch, but the consistent theme is moving large-model inference from the cloud onto the desk. A single Mac Studio with 512GB of high-bandwidth unified memory can hold models that would not fit in consumer GPU configurations, and clustering several units over Thunderbolt 5 gives small teams a way to serve the biggest open-weight models without renting datacenter capacity. The M6, meanwhile, pushes on-device AI into a compact, relatively low-cost desktop that Apple is explicitly marketing for always-on agentic workflows. If the claimed gains hold up in independent testing, Apple is positioning the Mac as a serious local alternative to GPU cloud instances for development and inference — and, with Core AI and MLX, it is building the software layer to keep that work inside its ecosystem.