AMD Launches Ryzen AI Halo Developer Platform to Run 200-Billion-Parameter Models Locally

AMD has announced Ryzen AI Halo, a compact developer platform it describes as its first system purpose-built for running, testing and building agent-based and generative AI applications entirely on local hardware — no cloud connection required. The platform is powered by the Ryzen AI Max+ 395 processor and is aimed squarely at developers, researchers and creators who want to work with large AI models without depending on remote compute or paying recurring cloud fees.

The headline specification is striking: AMD says the system can handle models of up to 200 billion parameters locally. That puts it in territory previously associated with server racks rather than a compact desktop unit.

What’s Inside the Machine

According to AMD’s published materials, Ryzen AI Halo ships with up to 128GB of unified system memory — meaning the CPU, GPU and NPU all draw from the same memory pool, rather than the GPU being limited to its own separate VRAM allocation. GPU performance is rated at 60 FP16 TFLOPS, and the NPU delivers up to 50 TOPS. Both Windows and Linux are supported, with AMD’s ROCm software stack included to give developers a full local AI development environment from day one.

The platform is designed to work with widely used AI developer tools. AMD specifically names PyTorch, vLLM, llama.cpp, Ollama, ComfyUI and LM Studio as supported frameworks, which covers a broad range of use cases from running local language models to image generation workflows.

Pre-orders are scheduled to open in June 2026, according to AMD’s materials. A price in GBP has not been confirmed in AMD’s published sources at the time of writing.

A Second, More Powerful Version Is Already Planned

AMD has also confirmed that a follow-up version of the Ryzen AI Halo platform is planned for the third quarter of 2026. That later system will use Ryzen AI Max PRO 400 Series processors and will push the unified memory ceiling considerably higher — up to 192GB of unified memory and 160GB of VRAM. AMD hasn’t published full specifications for that variant yet, but the roadmap suggests the company sees this product line as an ongoing platform rather than a one-off release.

Why Local AI Matters to Developers

Running large AI models locally has a practical appeal that goes beyond raw performance numbers. Cloud inference costs can mount quickly when developers are iterating on models or running inference at volume; a local system with fixed hardware costs changes that equation. There’s also the question of latency — a model running on-device responds faster than one that has to send data to a remote server and wait for a reply.

Industry commentary has positioned Ryzen AI Halo as a high-specification local AI workstation that undercuts some competing developer AI systems on price, though verified GBP pricing from AMD itself isn’t yet available. The combination of ROCm support, broad software compatibility and the Ryzen AI Max+ 395’s unified memory architecture is what AMD is betting will attract developers who’ve been building on cloud infrastructure and want to bring workloads in-house.

It’s worth being clear about what this product isn’t. Ryzen AI Halo isn’t a mainstream consumer desktop or a gaming PC. It’s a developer workstation in compact form, and AMD’s framing throughout is directed at technical users building and testing AI applications rather than end consumers running finished products.

Unanswered Questions

Several details remain unconfirmed. AMD’s published pages refer to Windows and Linux support broadly, though a post on NeowinFeed specifically referenced Windows 11 support; AMD’s own materials don’t narrow the Windows version down further. UK pricing hasn’t been published. And while AMD has outlined the Q3 2026 roadmap for the PRO 400 Series variant, no launch date or pricing for that system has been given.

Whether ROCm’s software ecosystem — which has historically lagged behind Nvidia’s CUDA in developer adoption — will be a friction point for developers evaluating the platform is a question AMD will need to answer as Ryzen AI Halo moves closer to availability.

Lisa Su, AMD’s chief executive, has spoken publicly about the company’s broader AI ambitions, saying: “We are focused on making AI accessible across every computing environment — from the data centre to the edge.”

What This Means for Kent Residents

For Kent-based developers, AI researchers or small businesses already experimenting with local language models through tools like Ollama or LM Studio, Ryzen AI Halo represents a potential on-premises alternative to cloud AI services — though UK pricing and availability from retailers haven’t yet been confirmed by AMD. More broadly, UK consumers and professionals interested in running large AI models locally without ongoing subscription costs will want to watch for pre-order details when June 2026 arrives.