Economic Tailwinds from AI’s CapEx
This year, the markets have been driven by the administration’s tariff policies. When higher than expected tariffs have been threatened or imposed, common stocks and fixed-income securities have lost value. On the other hand, lower tariff rates or delays in the assessment of these taxes have caused the markets to rebound. While these policy changes are extremely important, the secular growth in capital expenditures necessary to build out infrastructure for Artificial Intelligence (AI) is not getting the coverage that it deserves.
The United States is primarily a service-based economy with over 75% of economic activity tied to services rather than goods. The broad dominance of the service economy includes the economic value of the output and the workforce providing services. This service economy includes information technology, professional services, education, finance, government, and healthcare.
Of these sectors, the information technology sector dominates the services economy given the growth of its revenues and prospects for future sales expansion. The last technology boom was driven by the Internet and involved laying miles of fiber optics, installing switches and routers, and deploying millions of servers.
The current technology sector is rapidly building the infrastructure necessary to power AI. This growth requires the buildout of data centers tasked with running AI models. AI’s capabilities result from a combination of data inputs, neural networks and computing powers. The Scaling Hypothesis posits that increasing these three inputs results in greater intelligence. This intelligence increases at a rate greater than the sum of the parts. Why this happens is unknown to the AI researchers themselves. Compute for AI comes from the growth of data centers where each data center itself acts like one giant computer running the AI algorithms.
Following is a table from Bloomberg that shows the growth in capital spending by the largest technology companies in the US. Capital expenditures for this cohort are expected to grow from $169 Billion in 2022 to nearly $370 Billion in 2026, suggesting an annual growth rate of almost 22% per year.
These data centers are very power hungry. In 2024, US data centers required about 25 gigawatts of energy. According to a McKinsey & Company article*, this demand could grow to more than 80 gigawatts by 2030. For context, one gigawatt provides enough electricity for about 900,000 households. Therefore, these data centers alone will require the same energy as if the US grew by roughly 50 million households over the next five years.
The cost to build one gigawatt (GW) of new power generation and transmission capacity can range from $1.5 billion to $3 billion, according to Barron's**. Therefore, the capital expenditures associated with building enough electrical capacity to run AI will cost an additional $75-$150 billion.
Earlier this year, the Chinese AI company DeepSeek made waves when its model demonstrated impressive performance using less powerful hardware. These results suggest that DeepSeek’s training protocols were better and that the company’s models were more efficient. Therefore, there is the potential that future models will require less computational power – including fewer data centers – to still produce incredibly intelligent systems.
It is impossible to estimate the economic impacts of the administration’s tariff policies. The policies are in a state of flux within this administration and future presidents may return to historically lower tariffs depending on how the politics evolve.
On the other hand, the economic stimulus from the AI data center build is a secular trend that will power the US economy for the next few years. It should be noted that the above impacts only relate to the infrastructure spending itself. The productivity gains associated with AI could easily be multiples of the capital expenditures themselves.
Disclosures: For informational purposes only. The information has been obtained from sources believed to be reliable. Opinions are subject to change. Carnegie Investment Counsel (“Carnegie”) is a registered investment adviser under the Investment Advisers Act of 1940. Registration as an investment adviser does not imply a certain level of skill or training. For a more detailed discussion about Carnegie’s investment advisory services and fees, please view our Form ADV and Form CRS by visiting: https://adviserinfo.sec.gov/firm/summary/150488. You may also visit our website at: https://www.carnegieinvest.com
**Source: https://www.barrons.com/articles/natural-gas-power-plants-buyout-1fcb1d34