AMD's Ryzen AI Max 400 chip comes with 192GB of memory, but acquiring one can be quite challenging.

      AMD has introduced the Ryzen AI Max 400 series, featuring an astonishing 192GB of unified memory in a compact chip that can fit into a mini PC.

      While not much has changed from the previous generation chip, if you're interested in running large AI models locally, the AI Max 400 is certainly worth exploring.

      ✨ Personal AI is the upcoming computing platform. AI is evolving from something you merely access to something you create locally, at the edge, and across various systems. This unlocks new opportunities for developers: • @AMD Ryzen AI Halo, a local-first developer system, preorder… pic.twitter.com/iHf2NloDHv— Jack Huynh (@jackhuynh) May 20, 2026

      What, then, has changed from the last generation?

      The Ryzen AI Max 400, codenamed Gorgon Halo, continues with the same Zen 5 CPU architecture, RDNA 3.5 graphics, and XDNA 2 neural engine found in earlier models.

      Nonetheless, the top-tier Ryzen AI Max+ Pro 495 boasts a clock speed increase of 100 MHz compared to its predecessor, reaching a boost limit of 5.2 GHz. The mid and lower models, such as the Pro 490 and Pro 485, maintain a clock speed of 5 GHz without any enhancements in this area.

      It appears that AMD has primarily raised the memory limit of the Gorgon Halo to 192GB, compared to the 128GB limitation of the Strix Halo chips, as they are quite alike in other respects.

      So, is the 192GB unified memory significant?

      Yes, but for a relatively small group of users running LLMs locally on their devices, possibly for small businesses or research, where memory might be a crucial limitation for an otherwise capable system.

      AMD asserts that the Gorgon Halo is the first x86 chip able to manage an LLM with over 300 billion parameters entirely on-device, claiming it can allocate 160GB of its total 192GB as VRAM.

      This offers enough capacity to run AI models that would typically necessitate cloud computing or numerous powerful GPUs. Naturally, AMD is marketing the Ryzen AI Halo box as a solution to save around $750 per month on equivalent cloud API expenses.

      The key issue is timing. OEM systems from companies like Asus, HP, and Lenovo are expected to arrive in Q3 2026. Pre-orders for the Ryzen AI Halo box, which will include the previous-gen Strix Halo for $3,999, are set to begin in June.

      On the other hand, there is no confirmed date for Gorgon Halo systems. Additionally, with the ongoing global memory crisis already causing Apple to withdraw high-memory Mac Studio configurations, AMD's ambitions for 192GB may encounter challenges for large-scale production.