The AMD Advancing AI 2026 event concluded in San Francisco with a series of new datacenter hardware announced, mostly addressing the AI markets – covering everything from rack-scale infrastructure to edge robotics and developer software. For the silicon nerds out there, the chipmaker introduced its Instinct MI400 series GPUs and sixth-generation EPYC processor lineup for the occassion.
AMD Advancing AI 2026 Announcements
The headline hardware launch at the event is the AMD Helios rack-scale platform, which is effectively the primary competitor to NVIDIA’s NVL72 equivalents (latest generation being GB200 NVL72). The company described Helios as the industry’s first and most powerful open rack-scale AI solution, and it is now in production, being deployed by various AI companies at multi-gigawatt scale, with names such as Anthropic (makers of Claude) and Microsoft among those.
Next, let’s address the Instinct GPU: the new flagship datacenter GPU from Team Red, specifically the Instinct MI455X, packs 320 billion transistors through a multi-chiplet package surrounded by 432GB of HBM4 memory. While the chipmaker will replace the consumer-facing RDNA architecture with UDNA moving forward, this one is running on CDNA 5 architecture, so it’s hard to say if any of its design decisions will get carried over to the next-gen Radeons.
For CPUs, the 6th Gen EPYC processors lineup – split into 4 subcategories with SP7, SP8, 3D V-Cache and LP versions – will preview what would eventually be the next-gen Ryzen lineup. Inside these new datacenter GPUs are the new “Zen 6” CCD, and while AMD hasn’t explicitly confirmed this, the core count of the EPYC 9006 LP series “Verano” (a 72-core processor 6×12 layout makes more sense than a 9×8 layout) and the die shots in the marketing materials seem to suggest the rumors of the new CCD being 12-core units are likely true.
On the physical AI front, AMD announced the Ryzen AI Embedded X100 series processors, while new Kria AI System-on-Modules (SOMs) and a dedicated Robotics Developer Platform were also introduced to accelerate development in this space. Also introduced is ROCm.ai, which is AMD’s AI-native software platform built to unify AI development across its hardware.
Lastly on client computing, AMD only mentioned the Ryzen AI Max+ PRO series (codename “Gorgon Halo”), being the only new consumer lineup that is designed for local AI workloads. Like the Gorgon Point that acts as the refresh of Strix Point, Gorgon Halo is a minor update to Strix Halo lineup, this time bringing maximum memory to 192GB instead of 128GB, and a slight bump in clock speeds. The PRO series is identical to the ones announced in May this year, only that these models get enterprise manageability features for fleet use.
Pokdepinion: Would we even get to see consumer Zen 6 by year end? With how quiet it has been lately, I have my doubts.
