How AMD's new Helios AI infrastructure challenges NVIDIA's dominance
What's the story
Advanced Micro Devices (AMD) has unveiled its Helios rack-scale artificial intelligence (AI) solution and sixth-generation EPYC central processing units (CPUs). The new systems are powered by the latest Instinct MI455X graphics processing units (GPUs). The launch is a direct challenge to NVIDIA's Vera Rubin infrastructure, marking a major move in the rapidly growing AI compute market.
Shipping details
Helios system now in full production
At its Advancing AI conference in San Francisco, AMD CEO Lisa Su announced that the Helios system is now in full production.
Hardware partners are expected to start shipping the new systems by the end of Q3 2026.
The Helios system combines sixth-generation Epyc processors, Instinct MI455X GPUs, Pensando networking, and ROCm software to meet the growing demands of agentic AI deployments.
Performance comparison
Helios offers significant advantages over NVIDIA's Vera Rubin system
The Helios system is directly positioned against NVIDIA's Vera Rubin NVL72 system.
According to Su, Helios offers 15% more AI compute, 50% more memory, and 50% more scale-out bandwidth than its NVIDIA counterpart.
It also provides up to 30% more tokens per dollar, making it a cost-effective solution for large models and scalability across thousands of racks.
CPU performance
Venice optimized for enterprise, cloud, HPC, and AI workloads
AMD also unveiled its Venice family of four server CPUs, optimized for enterprise, cloud, HPC, and AI workloads.
The company claims its 256-core Venice CPU delivers 2.2 times the throughput of an 88-core Vera CPU.
A 96-core Venice chip also provides a 20% better per-core performance compared to NVIDIA's Arm-based Vera.
Market forecast
AMD revises AI infrastructure market forecast
AMD has revised its 2028 forecast for the AI infrastructure market from $500 billion to $1.4 trillion by 2030.
The company expects the server CPU market, currently at $25 billion, to reach $200 billion by 2030.
This growth is mainly due to the increasing demand for orchestration in agentic AI workloads.
Adoption and plans
Microsoft and Anthropic to adopt Helios for AI workloads
Microsoft is an early adopter of Helios, planning to use the system for AI workloads like frontier model AI inference and Azure AI services.
Anthropic has also announced plans to deploy up to 2 gigawatts (GW) of the GPUs within Helios solutions.
As for its future roadmap, AMD plans to launch seventh-generation CPUs (Florence) in 2028 and eighth-generation CPUs in 2030.