At its “Advancing AI” event, AMD fired a shot across the AI datacenter bow, unveiling plans for its EPYC “Verano” CPUs, Instinct MI500X accelerators, and next-gen rack-scale systems—all aimed directly at NVIDIA’s Vera Rubin-based NVL576 AI super-racks.
What To Expect In 2027:
-
EPYC Verano CPUs – AMD’s next-gen server chips built for AI-scale compute, likely powered by a Zen 6 or Zen 7 core architecture on TSMC’s cutting-edge N2P or A16 node with backside power delivery.
-
Instinct MI500X GPUs – AMD’s answer to NVIDIA’s Rubin AI chips, rumored to use CoWoS-L packaging and designed for high-density, high-performance AI workloads.
-
Second-gen AI rack – A post-Helios architecture with denser compute blades, faster I/O, and full-stack AMD silicon, set to rival or exceed Rubin-powered systems in scale and efficiency.
AMD’s 2026 Helios rack—powered by the 256-core EPYC “Venice” CPU, MI400X accelerators, and 800GbE Vulcano NICs—already targets Rubin’s NVL144 performance. But in 2027, AMD is shifting from “challenger” to “contender,” building an ecosystem that scales from silicon to system.
NVIDIA’s Rubin VS AMD’s Answer
While NVIDIA accelerates on a six-month cadence, AMD is playing the long game with yearly, stable rollouts that ensure real-world deployment and system maturity. And it’s working.
Lisa Su didn’t mince words:
“We are already deep in the development of our 2027 rack-scale solution… Verano CPUs and Instinct MI500X-series GPUs will push the envelope on performance, efficiency, and scalability.”
NVIDIA’s Vera Rubin might own the spotlight now, but AMD is assembling a rack-ready arsenal that could flip the datacenter power dynamic. It’s not just about matching Rubin’s brute force—it’s about building a full-stack AMD alternative, from CPU to GPU to networking, optimized and battle-tested.
AMD vs. NVIDIA: The AI Rack War Begins
In this billion-dollar battleground, performance alone won’t cut it. Ecosystem matters. Efficiency matters. Delivery matters.
And AMD’s message is clear: We’re not here to play.
