AMD announced the Ryzen AI Max 400 series, and the headline number is genuinely staggering: 192GB of unified memory in a chip small enough to fit inside a mini PC.
Not much has changed from the last generation chip, but even so, if you’re all for running large AI models locally, the AI Max 400 is definitely worth checking out.
What has actually changed from the last generation?
The Ryzen AI Max 400, codenamed Gorgon Halo, carries forward the same Zen 5 CPU architecture, RDNA 3.5 graphics, and XDNA 2 neural engine from the previous generation chips.
However, the flagship Ryzen AI Max+ Pro 495 squeezes out a clock speed bump of 100 MHz over its predecessor, pushing the boost ceiling to 5.2 GHz. The mid and lower-tier variants, including the Pro 490 and Pro 485, clock in at 5 GHz; they don’t get any upgrades in the area.
To me, it sounds like AMD has simply increased the 192GB memory ceiling on the Gorgon Halo, versus the 128GB cap on the Strix Halo chips, as both of them are quite similar in every other way.
So, does the 192GB unified memory matter?
Yes, but for a relatively small number of users, who are running LLMs locally on their devices, perhaps for a small business or research, where memory could be the real bottleneck for an otherwise capable system.
AMD claims the Gorgon Halo is the first x86 chip capable of handling an LLM with up to 300B+ parameters, entirely on-device, and to make sure that the claim holds up, it can allocate 160GB of the total 192GB as VRAM.
That is enough headroom to run AI models that otherwise require cloud compute or plenty of powerful GPUs. Naturally, AMD is positioning the Ryzen AI Halo box as something around the “token economy” argument, claiming that one unit can save up to $750 per month on equivalent cloud API costs.
The catch here is timing. OEM systems from brands like Asus, HP, and Lenovo land in Q3 2026. Pre-orders for the Ryzen AI Halo box, which ships with last-gen Strix Halo at $3,999, open in June.
Meanwhile, Gorgon Halo systems have no confirmed date yet. And with the global memory crisis already forcing Apple to pull high-memory Mac Studio configurations, AMD’s 192GB aspirations may be harder to ship at scale.
