AMD Unveils Ryzen AI Halo and Ryzen AI Max PRO 400 Chips

AMD is making a serious play to pull AI computing away from the cloud and place it squarely onto your desk. The company just unveiled its new Ryzen AI Halo developer platform alongside the Ryzen AI Max PRO 400 Series processors. These chips are built specifically for what the tech industry calls Agent Computers, which are local systems powerful enough to plan actions, understand complex prompts, and run intense workflows with very little human intervention.

The big hurdle for local AI has always been memory capacity and latency. AMD is tackling this by packing an incredible amount of unified memory right onto the chip, bypassing the need for expensive multi-GPU enterprise setups just to test local models.

Local Environments Built for Developers

The journey starts with the Ryzen AI Halo developer platform. Launching as a compact, validated system designed for local AI creation, it gives developers a way to build and test applications without dealing with constant cloud subscription costs or data privacy worries.

The initial developer kit is powered by the Ryzen AI Max+ 395 processor and supports up to 128GB of unified system memory. This gives it enough muscle to run models with up to 200 billion parameters right on the device. It plays nicely with the tools developers are already using, including PyTorch, vLLM, llama.cpp, and Ollama. It's fully optimized for AMD's ROCm software stack, letting you move a project seamlessly from a Linux prototype to a final Windows deployment on a single machine.

Pre-orders for this initial platform start in June 2026, and it will be sold exclusively through Micro Center.

Pushing the Limits With Ryzen AI Max PRO

AMD is taking things a step further later in the year. In the third quarter of 2026, the company will update the Ryzen AI Halo platform with the newly announced Ryzen AI Max PRO 400 Series processors.

Built on the Zen 5 architecture, these chips combine high-performance CPU cores with RDNA 3.5 graphics and an XDNA 2 NPU capable of up to 55 TOPS of local AI performance. The real kicker is the memory configuration. These top-tier models scale up to 192GB of unified system memory, allocating up to 160GB specifically for VRAM. This massive memory pool makes them the first x86 client processors capable of running a 300 billion parameter model entirely on your local machine.

The lineup includes three specific chips. The flagship Ryzen AI Max+ PRO 495 packs 16 cores and 40 graphics compute units. Right behind it is the Ryzen AI Max PRO 490 with 12 cores and 32 graphics units, followed by the 8-core Ryzen AI Max PRO 485. All three feature a configurable TDP ranging from 45W to 120W, making them versatile enough for mobile workstations and small form-factor desktop builds.

Hardware Coming Later This Year

You won't have to wait too long to see these enterprise chips out in the wild. Major OEM partners, including HP and Lenovo, are already working on bringing these processors to commercial AI PCs and workstation-class laptops.

These upcoming systems aim to consolidate heavy data science workloads, 3D rendering, and local agentic AI execution into a single architecture. If you're looking to upgrade your development setup, expect to see the first retail systems hitting the market during the third quarter of 2026.

Source