AMD targets 4x increase in AI energy efficiency by 2026

AMD has announced that it is on track to achieve a fourfold increase in AI energy efficiency by 2026 compared to 2024 levels. This progress supports the company’s long-term goal of a 20x increase in rack-scale efficiency for AI training and inference by 2030.

The efficiency improvements focus on a holistic rack-scale approach rather than individual chips. This includes the co-optimization of accelerators, CPUs, memory, interconnects, and software. By improving performance-per-watt across the entire system, AMD aims to address the power and cooling constraints currently facing data centers.

Recent developments include the launch of the Helios rack-scale compute platform, which integrates 72 MI455X GPUs. While individual components like the MI455X offer significantly higher performance than previous generations, the company is prioritizing the ability to scale AI workloads efficiently across the entire system to reduce the total cost of ownership and energy consumption.