Photo caption: Artificial Intelligence
The cost of building the world’s most advanced AI data centre could reach $200bn within the next five years, according to a new study by researchers from Georgetown University, Epoch AI, and the RAND Corporation.
The report, which examined the global growth of AI supercomputers from 2019 to 2025, warns that the scale of computing required to stay competitive in artificial intelligence is accelerating rapidly, and so are the costs.
Driven by massive investments, companies have been scaling up their AI supercomputers to handle increasingly complex training runs.
“If the observed trends continue, the leading AI supercomputer in June 2030 will need two million AI chips, cost $200bn, and require 9GW of power,” the report stated.
Power consumption and hardware costs are doubling at a striking pace. According to the study, hardware costs for top AI systems have grown by a factor of 1.9 each year, while power demands have increased by 2.0 times annually.
While breakthroughs in chip manufacturing and mega-investments such as the $500bn Project Stargate may help keep hardware and funding on track, the escalating energy requirements pose a more difficult challenge.
“Powering the next frontier of AI will stretch current infrastructure to its limits,” the researchers said, suggesting that decentralised training methods might be needed to spread the load across multiple global locations.
Much of the momentum is being driven by the private sector. Since 2019, corporate-run AI systems have grown at an annual rate of 2.7 times, outpacing public sector systems, which have expanded at 1.9 times per year.
As a result, the private sector’s share of global AI computing power is projected to jump from 40 per cent in 2019 to 80 per cent by 2025.
The study painted a stark picture of the AI-driven future—one not only marked by smarter machines but also by extraordinary financial, energy, and infrastructure demands.
