PwC Puts $31.6 Trillion Price Tag on Global AI Data Center Build-Out Through 2050

The number that keeps surfacing in every AI infrastructure conversation just got a long-range anchor. PwC released its inaugural Global Data Centre Outlook on September 2, projecting $31.6 trillion in cumulative data center capital expenditure across 46 countries and territories through 2050, with a plausible upside of nearly $50 trillion if AI adoption moves faster than the central forecast.

The modeling, which PwC commissioned from Oxford Economics, is the first long-range capex forecast of its kind to extend through mid-century, according to a press release reviewed by Bloomberg. The baseline scenario puts annual data center spend at roughly $800 billion this year, rising to $1.1 trillion by 2030 and $1.8 trillion by 2050.

The regional split matters as much as the total. According to PwC's Global Data Centre Outlook, the United States is expected to capture close to half of cumulative global investment, at $15.1 trillion, reflecting the country's position at the center of the advanced-chip ecosystem. Asia Pacific follows at $8.2 trillion, Europe at $5.6 trillion, the Middle East at $1.1 trillion, and Africa at $255 billion. China and India are forecast to drive the largest share of incremental Asia Pacific demand, per the Outlook.

One number inside the report deserves more attention than the headline: by 2050, ICT equipment, meaning chips, servers, and networking gear, will account for 93% of annual data center investment, up from 70% today. That shift is the actual structural argument PwC is making. Unlike a fiber rollout or a highway system, where construction is the majority of the cost and the asset then sits for decades, AI infrastructure requires hardware refreshes roughly every four to six years. The spending never plateaus. It compounds.

That framing directly challenges the common investor question about whether current hyperscaler capex is a bubble with a natural ceiling. PwC's position, supported by the Oxford Economics modeling, is that the ceiling keeps moving because the hardware does.

The report also ran two alternative scenarios. In a tighter export-controls environment, where semiconductor supply chains face meaningful disruption, cumulative global investment through 2050 would fall by close to 20%, settling around $25.5 trillion, before recovering as supply chains adapt, according to the Outlook. A digital-sovereignty scenario, in which countries push AI spending into domestic infrastructure, reshuffles where the capital lands rather than shrinking the pool.

The one constraint PwC calls out above all others is power. The Outlook identifies electricity availability as the decisive factor shaping where investment flows, ahead of capital access, labor, or regulatory environment. Affordable, reliable, and increasingly low-carbon power at scale is, the report notes, the hardest requirement for most markets to meet.

That's the part worth watching on the infrastructure side. The capital forecasts in a report like this are educated projections, not contracts. But the power constraint is already showing up in real siting decisions, permitting backlogs, and utility negotiations happening right now. The $31.6 trillion number tells you roughly how much the industry intends to spend. The grid tells you where it actually can.

Sources cited:
- PwC Global Data Centre Outlook (press release) (https://www.pwc.com/gx/en/news-room/press-releases/2026/global-investment-in-ai-infrastructure.html)
- PwC Global Data Centre Outlook (report page) (https://www.pwc.com/gx/en/1/services/consulting/technology/data-centre-outlook.html)
- Quartz (https://qz.com/pwc-data-center-ai-investment-forecast-2050-090226)
- Taipei Times (https://www.taipeitimes.com/News/biz/archives/2026/09/04/2003863640)
- TheStreet (https://www.thestreet.com/technology/pwc-ai-data-center-capex-forecast-power-constraint)
- IEEE ComSoc Technology Blog (https://techblog.comsoc.org/2026/09/03/pwc-global-ai-data-center-spending-to-hit-31-6tn-by-2050-role-of-full-stack-orchestration-layer/)