From large training clusters to small edge inference servers, AI is becoming a larger percentage of data center workloads. This represents a shift to higher rack power densities. AI start-ups, enterprises, colocation providers, and internet giants must now consider the impact of these densities on the design and management of the data center physical infrastructure.
A contribution from our premium sponsor Schneider Electric.
In recent years, we have witnessed an extraordinary acceleration in the growth of artificial intelligence (AI), transforming the way we live, work, and interact with technology. Generative AI (e.g., ChatGPT) is a catalyst for this growth. Predictive algorithms are having an impact on industry sectors ranging from healthcare and finance to manufacturing, transportation and entertainment.
The data requirements associated with AI are driving new chip and server technologies resulting in extreme rack power densities. At the same time, there is massive demand for AI. Together these present new challenges in designing and operating data centers to support this demand.
AI growth projection
Schneider estimates that AI represents 4.5 GW of power consumption (i.e., demand) today and projects this to grow at a CAGR of 25% to 33%, resulting in a total consumption of 14 GW to 18.7 GW by 2028. This growth is two to three times that of overall data center power demand CAGR of 10%. See Table 1 for more details.
One key insight is that inference loads will increase over time as more newly trained models are transitioned to production. The actual energy demand will heavily depend on technology factors including successive generations of servers, more efficient instruction sets, improved chip performance, and continued AI research.
Read more about important AI attributes and trends that create challenges for data centers in the white paper "The AI Disruption: Challenges and Guidance for Data Center Design".