AI chip supply reportedly falls short in meeting data center demands
In a recent analysis, London Economics International has shed light on the growing demand for electricity from data centers in the U.S., particularly those powered by AI. The report, commissioned by the Southern Environmental Law Center, suggests that projections around U.S. data center growth and electricity demand may be overstated due to global constraints in semiconductor chip production.
The analysis reveals that the demand for electricity from data centers could account for up to 12% of the U.S.' total electricity demand by 2028, depending on the scenario. This growth is driven by the exponential growth of AI and other tech services. By 2035, power demand from AI data centers in the U.S. is projected to reach 123 GW, up from 4 GW in 2024.
However, the report also highlights potential challenges. Constraints in semiconductor chip production could limit the ability to meet this demand, thereby affecting the rate of growth in electricity demand. The trend towards bigger data centers to meet AI needs could exacerbate this demand, despite efforts towards energy efficiency.
The report's conclusions are supported by other observations. Astrid Atkinson, a former Google senior director of software engineering, stated that utilities are seeing five to 10 times more interconnection requests than data centers that will actually be built. This indicates a potential overestimation of data center growth.
AI chip manufacturing capacity has grown about 6.1% annually over the last decade. However, for the forecasts to hold water, the U.S. would require more than 90% of the world's new supply of semiconductor chips from 2025 through 2030, which is unrealistic.
The implications of these findings are significant. If AI chip manufacturing capacity were to rise significantly, to 10.7%, it could supply an incremental 63 GW of data center-related demand globally through 2030. However, the U.S. currently accounts for slightly less than 50% of global semiconductor chip demand, suggesting that other regions in the world are also projecting strong demand.
The U.S. is rushing to add new generation and transmission to power various sectors, including transportation, buildings, and industry. However, data centers are expected to require an outsized portion of the resources. London Economics projects that the U.S. data center demand from the power market would require 90% of global chip supply through 2030.
The Southern Environmental Law Center warns that speculative infrastructure investment creates significant economic risks for ratepayers, who ultimately bear the financial burdens. Regulators are urged to prioritize transparent, realistic, and evidence-based analyses to ensure infrastructure developments are necessary and beneficial.
In conclusion, while the demand for electricity from data centers, particularly those powered by AI, is expected to grow significantly, the current projections do not account for potential limitations due to semiconductor chip production constraints. Therefore, the projected demand might be overstated if these constraints significantly impede the rapid expansion of AI data centers. However, the primary drivers of increased electricity demand remain the exponential growth of AI and data center infrastructure.
[1] Source: Schneider Electric's 2030 AI power demand scenarios [2] Source: London Economics International analysis for the Southern Environmental Law Center [3] Source: Energy Star, U.S. Environmental Protection Agency
- Given the growing demand for electricity from AI data centers, there could be a substantial increase in the need for both AI chips and energy within the technology industry, particularly in finance and energy sectors.
- The report indicates that that the trend towards larger data centers to meet AI needs could strain semiconductor chip production capacity in the technology industry, which could potentially affect the growth in electricity demand across various industries, including artificial-intelligence and others reliant on technology services.
