AI capital expenditure forecast to exceed the cost of building railways in both the U.S. and the U.K. — with the internet added on top
PwC expects global data center capital expenditure to reach $31.6 trillion between 2026 and 2050
PwC forecasts spending on artificial intelligence to reach almost $32 trillion in 24 years as technology companies race to increase the build-out of AI.
The Big Four firm’s outlook on global data centers from 2026 to 2050 is for capital expenditure to surge to $31.6 trillion in that period, with a potential upside of $50 trillion if the rate of AI adoption speeds up. The data is based on information from advisory firm Oxford Economics, which modeled data center capex in 36 countries and territories and five regions.
According to the report, unlike the development of railways or the internet, the infrastructure of AI stands out because its building cycle resets every four to six years. The expected spending figures on AI are larger than the combined figures for the expansion of railways in the U.S. and the U.K. and the cost of building the internet, even when adjusted for inflation, per PwC.
The report noted that what also distinguishes this spending cycle from other similar ones is that its due to increase over time, with the firm expecting capex to grow from $800 billion this year to $1.1 trillion in 2030 and $1.8 trillion in 2050.
Don’t Short Yourself
Free Weekly Newsletter
Don’t Short Yourself offers weekly money tips to help you earn it, stack it and grow it.
“That’s because the bulk of the spend doesn’t go towards the buildings,” the report said. “Rather, it funds what fills them: servers, storage systems, networking equipment, central processing units (CPUs), and, crucially, the graphics processing units (GPUs) that provide compute power for AI —which age out in a handful of years and will need to be replaced.”
PwC’s $31.6 trillion prediction sits within a range of $22 trillion and $50 trillion, where the actual outcome will be based on the acceleration and rate of AI adoption. But across scenarios, the firm sees capex rising considerably by 2050.
Goldman Sachs has boosted its forecasts for hyperscaler-spending on AI from $1.2 trillion to $1.7 trillion for 2027 and from $1.5 trillion to $2.1 trillion for 2029.
“That’s a significant increase over a multi-year period in hyperscaler expected spending,” Brian Singer, analyst at Goldman’s research arm, said in a recent episode of the investment bank’s Exchanges podcast.
It also raised its targets for data center power demand, with 2030’s expected figure of 83 gigawatts lifted to 108 gigawatts. Singer said that the jumps in demand for AI and spending on the the buildout of the tech were “hard to ignore.”
The AI trade has powered U.S. equities this year, with the S&P 500
rising 12% and a leading index of semiconductor stocks up 59%.
Facts Only
* Global data center capital expenditure is expected to reach $31.6 trillion between 2026 and 2050.
* Spending on artificial intelligence is forecasted to reach almost $32 trillion in 24 years.
* The cost of AI infrastructure exceeds the combined figures for railway expansion in the U.S. and U.K. and internet construction, adjusted for inflation.
* The forecast is based on data from Oxford Economics modeling 36 countries/territories and five regions.
* Data center spending can potentially reach $50 trillion under accelerated AI adoption rates.
* Spending on AI funds hardware like servers, storage, networking, CPUs, and GPUs, which have shorter replacement cycles than traditional infrastructure.
* Capital expenditure is expected to grow from $800 billion this year to $1.1 trillion in 2030, and $1.8 trillion in 2050.
* Goldman Sachs raised forecasts for hyperscaler spending on AI from $1.2 trillion to $1.7 trillion for 2027 and from $1.5 trillion to $2.1 trillion for 2029.
* Data center power demand for 2030 was expected at 83 gigawatts, which has been lifted to 108 gigawatts by Goldman Sachs.
Executive Summary
Global data center capital expenditure is forecast to reach $31.6 trillion between 2026 and 2050, with a potential upside of $50 trillion contingent on the acceleration of artificial intelligence adoption. This projection is based on modeling data center capex across thirty-six countries and territories and five regions by Oxford Economics. The spending on AI is predicted to reach almost $32 trillion over the same period. The cost of building AI infrastructure is estimated to exceed the combined costs of expanding railways in the U.S. and the U.K., and the development of the internet, even when accounting for inflation.
The spending cycle for AI infrastructure differs from physical infrastructure like railways or the internet because its building cycle resets every four to six years. This investment primarily funds supporting components such as servers, storage systems, networking equipment, CPUs, and GPUs, which require replacement in a short timeframe rather than long-term construction projects. Projections indicate that capital expenditure on this sector is expected to grow significantly, rising from $800 billion this year to $1.1 trillion in 2030, and further to $1.8 trillion by 2050.
Full Take
The narrative positions AI capital expenditure as an infrastructural shift characterized by rapid, cyclical reinvestment in compute components rather than long-term physical construction. The core implication is that the velocity of AI adoption creates a spending trajectory that dwarfs historical benchmarks for large-scale infrastructure development. This pattern suggests a fundamental decoupling between traditional capital expenditure models (like building railways or the internet) and the exponential demands of computational capacity, specifically driven by rapidly aging hardware components like GPUs.
The mechanism identified—spending on compute versus physical structures—suggests that the true bottleneck is not merely the physical placement of data centers but the continuous cycle of hardware refreshment necessary to sustain the AI build-out. The uncertainty regarding whether the $31.6 trillion figure or the $50 trillion ceiling is realized depends entirely on the pace and efficiency of this replacement cycle, which links directly to real-world technological capacity rather than mere financial projections. The pattern observed involves leveraging a seemingly concrete expenditure (data center build-out) to mask an underlying dynamic process (hardware iteration speed).
What are the hidden assumptions driving the accelerated growth rates cited by analysts? Are these forecasts based on optimistic adoption curves or do they account for potential systemic bottlenecks in the supply chain that could constrain the projected $50 trillion ceiling? Furthermore, if spending is overwhelmingly focused on hardware iteration rather than physical expansion, what does this imply about long-term infrastructure resilience and sustainability outside of pure financial metrics?
Sentinel — Human
The text reads like a synthesis of financial reports and expert commentary, exhibiting a structure and citation pattern consistent with human-authored analysis rather than pure machine generation.