A new forecast from PwC projects that global investment in AI data center infrastructure will reach $32 trillion by 2050, a sum the report says exceeds the capital required for past transformative networks like railways, electrification, and the internet. The spending cycle, driven by rapid turnover in GPU hardware, is expected to reset every four to six years, according to the report. The scale of this build-out is tied directly to the pace of chip development from Nvidia, AMD, and other manufacturers, which release new generations every two to three years.
The report breaks down regional spending, with the United States projected to account for $15.1 trillion of the total. The Asia-Pacific region, including China and India, is expected to spend $8.2 trillion, while Europe is forecast to invest $5.6 trillion. Smaller allocations include $1.1 trillion for the Middle East and $255 billion for Africa. Beyond chipmakers, the report identifies networking equipment and copper materials for internal data center power as sectors that will benefit from the massive outlay.
PwC’s forecast is not without caveats. The report flags power availability, data sovereignty requirements, and chip availability as factors that could slow the build-out. In the U.S., data centers are projected to consume 20% of the nation’s total power supply by 2035, prompting operators to turn to natural gas turbines for on-site generation. That demand has already led to jet engine shortages, with SpaceX beginning in-house turbine blade manufacturing to cut delivery delays of up to 18 months.
Geopolitical tensions pose another risk to the forecast. Trade bans on rare earth elements and high-end chips could reduce the global investment projection by 20%, according to the report. The report also acknowledges broader concerns that the current AI boom may be a bubble, noting that some AI tech companies carry what it calls “hidden debt” worth around $1.65 trillion. Costs are rising as firms like OpenAI seek a path to profitability.
Despite those risks, investment continues apace. Nvidia is partnering with several firms to build a $500 billion AI infrastructure fund to support further investments. Meanwhile, one Google architect has said that a data center GPU’s service life is only about one to three years, a factor that some experts worry could turn GPU depreciation into a major financial crisis for hyperscalers. The report’s authors frame the current cycle as a defining era, noting that the prior infrastructure booms each required enormous capital and reshaped their periods.
More hardware news from TechManNews.
