AMD has fundamentally redefined the capabilities of local artificial intelligence by launching the Ryzen AI Max PRO 400 series. Announced during the IFA 2026 opening keynote in Berlin on September 4, 2026, this new flagship silicon pushes the memory ceiling to unprecedented levels for x86 client processors. The headline feature is support for up to 192GB of unified LPDDR5X-8533 memory, with as much as 160GB allocatable directly as GPU VRAM. This capacity allows a single desktop or workstation to run large language models with 300 billion parameters entirely locally, eliminating the need for cloud GPU rentals or recurring API costs.
Breaking the Memory Barrier
The transition from the previous generation represents a significant leap in hardware capability. The Ryzen AI Max PRO 300 series, which topped out at 128GB of LPDDR5X-8000 unified memory, already attracted attention from AI developers seeking to reduce inference costs. The new 400 series increases this limit by 50%, reaching 192GB. This expansion places a mainstream Windows-compatible chip in territory previously reserved for high-end cloud infrastructure or Apple’s unified-memory Macs. While Apple’s M3 Ultra configurations still offer higher maximum memory at 512GB, AMD’s solution provides the advantage of running standard Windows and Linux AI tooling rather than a closed ecosystem.
Technical Specifications and Lineup
The Ryzen AI Max PRO 400 series, internally referred to as “Gorgon Halo,” consists of three halo-tier processors built on Zen 5 CPU cores paired with an RDNA 3.5 integrated GPU. The lineup is designed for commercial AI PCs and what AMD terms “agent computers,” machines built to run autonomous AI agents locally.
- Ryzen AI Max+ PRO 495: The top-tier model features 16 cores and 32 threads, a boost clock up to 5.2GHz, and an NPU rated at up to 55 TOPS.
- Ryzen AI Max PRO 490: This mid-range option offers 12 cores and 24 threads with an NPU performance of roughly 50 TOPS.
- Ryzen AI Max PRO 485: The entry-level SKU provides 8 cores and 16 threads, also with a 50 TOPS NPU.
All three models share the same maximum 192GB unified memory ceiling and 160GB VRAM allocation limit. The configurable TDP range spans from 45 to 120 watts, allowing flexibility for different chassis designs. Memory bandwidth on the top-end configuration reaches approximately 270 to 273GB/s, a critical metric for token-by-token LLM inference where bandwidth often determines generation speed.
Practical Impact for Developers
For technical readers, the primary benefit is the ability to deploy complex models without external dependencies. Running 300-billion-parameter models at INT4 quantization locally means that sensitive data remains on-premise, and operational costs are reduced to electricity and hardware depreciation. This is particularly relevant for enterprises and developers who require low-latency responses and strict data privacy. The shift from cloud-dependent inference to local execution changes the economic model of AI deployment, making high-end model testing and production viable on a single workstation.
AMD described the Ryzen AI Max PRO 400 as the world’s first x86 client processor capable of running 300B-plus-parameter models locally.
This distinction separates the new chip from every prior Windows-compatible AI PC component. It signals a maturation of the AI PC category, moving beyond simple NPU-assisted tasks to full-scale model hosting. The 50% increase in memory capacity is not just a spec bump; it is a functional threshold that enables a new class of local AI applications.
What to Watch Next
As of late September 2026, AMD has not disclosed specific pricing for the processors, leaving final system costs to individual OEMs. Readers should monitor upcoming system launches from major PC manufacturers to see how they configure these chips. The choice between the 495, 490, and 485 SKUs will depend on the balance between CPU core count, NPU performance, and the specific memory requirements of the intended AI workload. Additionally, the performance of the RDNA 3.5 integrated GPU in handling large model weights will be a key factor in real-world throughput. The industry will be watching to see if this local-first approach gains traction among enterprise users who have traditionally relied on cloud services for heavy AI workloads.
In conclusion, the Ryzen AI Max PRO 400 series marks a pivotal moment for local AI. By providing the memory headroom to run massive language models on a single x86 chip, AMD has empowered developers and enterprises to take control of their AI infrastructure. This move challenges the dominance of cloud-based inference and offers a compelling alternative for those prioritizing privacy, cost-efficiency, and low latency.