Kyle Wiggers, AI Editor for TechCrunch, recently collaborated on a deep and broad study. This investigation unpacks the evolution and expansion of the data centers that form the backbone of AI’s growing data needs. Our detailed analysis unpacks over 500 distinct AI data center projects. Specifically, it forecasts major trends in computational performance, power consumption, and capital expenditures. These results predict an amazing change in the infrastructure required to support artificial intelligence technologies by 2030.
Perhaps most shocking of all, it demonstrates that data centers are undergoing a phenomenal increase in computational performance, doubling their capabilities every year. Unfortunately, this tremendous jump in performance comes along with a similarly volatile increase in power demand. From 2019 to 2025, the power demands of AI data centers are predicted to double annually. This year-over-year upsurge indicates rising demand for energy to power sophisticated AI computations.
The capital expenditures for these data centers have gone through the roof. During the same timeframe, hardware costs skyrocketed at a rate of 1.9 times per year. Increased capital expense This financial pressure tells you just one aspect of the growing expense of setting up AI data centers and keeping them running. The economic study predicts that by June 2030, setting up the world’s premier AI data center could reach around $200 billion.
Nevertheless, even major energy efficiency improvements will likely not be enough to offset the swelling power needs of these facilities. The study had some positive findings, including an impressive improvement in computational performance per watt, increasing 1.34X each year from 2019-2025. This progress is not nearly enough to keep up with the expected energy demands of new AI data centers.
OpenAI, meanwhile, has publicized similarly ambitious plans to raise the sun-scorched ground under its startups up to $500 billion. Their goal is to build out a huge network of connected AI data centers around the US and later, the world. These facilities are all about development and deployment of AI applications. Yet they can each soon contain hundreds of millions of chips and require energy on par with that of a medium-sized city.
The organization’s acknowledgement that the technology landscape is rapidly evolving remains critical. By 2030, experts expect a leading edge AI data center will be filled with an estimated 2 million AI chips and need about 9 gigawatts of capacity. This stark increase is a reminder of the pressing need for infrastructure that can handle the rapidly growing needs of AI.
