Source: Abhineet Kaul, et al., “AI Beyond Data Centers: On-Device, On-Demand, Everywhere,” (Access Partnership, October 13, 2025).
Commentary: Generative artificial intelligence (AI) is projected to be a generational productivity and growth catalyst, with experts estimating that the technology will lead to a 7 percent increase in worldwide gross domestic product (GDP) growth and a 1.5 percent or more increase in productivity over the next decade. However, technology also requires significant energy to perform the inferences that enable it to produce text, images, and videos. At the current pace of AI growth, compute demand is projected to increase by a factor of 125 by 2030, requiring more and larger data centers that consume large amounts of energy. In fact, data center capacity will need to grow by 130 percent by 2030 to meet this demand, the equivalent of building 1,000 hyperscale data centers. What’s more, the resources needed to build and power these data centers will be costly, requiring an additional $2.8 trillion in infrastructure investment on top of what major tech firms have already committed (estimated at $3 to $8 trillion). On-device AI—a lightweight model that runs on a user device rather than relying on a data center—is a solution to this data center bottleneck. These smaller AI models are estimated to reduce AI's total power consumption by about 90 percent.
