AI workload impact on data center efficiency

The data center and IT sectors are increasingly focusing on efficiency and cost optimization driven by the rise of artificial intelligence workloads.

Initial developments show a transition in how data center efficiency is measured. The industry is moving from the standard Power Usage Effectiveness (PUE) metric toward a more granular “energy per inference” metric to better understand the power requirements of specific AI tasks. While a PUE of 1.20 is considered excellent, the shift toward “energy per inference” addresses the need for specific insights into AI task execution as data rates scale to 800G and 1.6T.

To manage rising power demands, technologies like Linear Pluggable Optics (LPO) are being used to reduce power draw in optical interconnects by shifting signal conditioning to host switch silicon. This can reduce the power draw per 800G link from approximately 13W-16W to between 7W and 9W. Looking further ahead, the industry is exploring optical-electrical convergence to combat massive consumption; for instance, NTT’s IOWN initiative aims to reduce network power consumption by 100 times by replacing electrical transmission with light.

IT leaders are also prioritizing infrastructure and cost optimization. This includes “storagemaxxing” and using tools like Amazon S3 Storage Lens and AI coding agents to manage datasets and reduce storage costs. As the International Energy Agency predicts global data center electricity consumption could reach 1,000 TWh by 2030, these advancements in storage lifecycle management and optical computing are becoming critical to managing the AI lifecycle and broader technological scaling.