AMD has unveiled Ryzen AI Halo, a compact AI development system built around its Strix Halo platform and directly targeting Nvidia’s DGX Spark and Apple’s Mac Mini systems. According to AMD, the new Ryzen AI Halo system is designed for AI development, local inferencing, and generative AI workloads, while retaining full x86-64 compatibility with both Windows and Linux. AMD says the platform can operate as a standalone AI workstation or as part of a small multi-system cluster connected to larger workstations.
“AI is no longer confined to the cloud. It is now something developers can build, train, and run locally,” said Jack Huynh, SVP and GM, Computing and Graphics Group. “With the Ryzen AI Halo and Ryzen AI Max PRO 400 Series, we are delivering the performance, memory, and open AI software stack that developers and enterprises need to bring the next generation of agentic AI systems to life right on their desks.”
The heart of the AMD Ryzen AI Halo is the Ryzen AI Max+ 395 processor, combining a 16-core Zen 5 CPU with a large RDNA 3.5 integrated GPU and providing 50 TOPS of AI performance from the NPU. The system also comes with 128GB of LPDDR5X memory through a quad-channel 256-bit memory interface, as well as a 2TB NVMe SSD. According to AMD, the large unified memory pool enables support for AI models with up to 200 billion parameters using suitable quantization methods.
AMD also compared Ryzen AI Halo against Nvidia’s DGX Spark and Apple’s Mac Mini with M4 Pro in several AI workloads. According to AMD, the system delivers up to 7% higher token throughput in GPT OSS 120B and up to 12% better performance in Qwen 3.5 122B compared to DGX Spark. The company also reported smaller gains in less memory-intensive models, including a 4% lead in Qwen 3.6 35B and up to 14% higher performance in GLM 4.7 Flash 30B workloads.
Against Apple’s Mac Mini with M4 Pro, AMD highlighted the advantage of Ryzen AI Halo’s 128 GB memory configuration, noting that the Mac Mini’s 64 GB memory limit prevents it from running larger models such as GPT OSS 120B and Qwen 3.5 122B locally. AMD also claims the platform can deliver up to 4x higher performance in certain generative AI workloads compared to Apple’s compact desktop system.
AMD was keen to note the software side of the platform with tools such as the AMD Ryzen AI Development Center and AI Playbooks, designed to simplify deployment of AI workflows on both Windows and Linux. The company says several AI workflows come pre-installed, including image generation, local LLM deployment, AI-assisted coding, and workflow automation tools, with additional Playbooks planned for monthly release.
The company is also introducing the AMD AI Developer Program, which includes cloud credits, training resources, developer community access, and optimized support for Ryzen AI Halo users.
The Ryzen AI Halo system is priced at $3,999, with pre-orders opening in June 2026. The updated version, with the new Ryzen AI Max+ PRO 495 SKU, is coming at a later date.
