AI Data Center Spending Could Reach $31.6T by 2050
Global data center capital spending could reach $31.6 trillion from 2026 through 2050 as artificial intelligence drives demand for computing capacity, according to a PwC outlook. The report’s central finding is that the spending cycle would continue long after facilities are built.
PwC commissioned Oxford Economics to model spending across 46 countries and territories. Slower AI adoption would bring cumulative spending to about $22 trillion, while faster adoption could push it close to $50 trillion. The estimates are expressed in 2025 U.S. dollars.
In the central scenario, annual spending climbs from roughly $800 billion in 2026 to $1.1 trillion in 2030 and $1.8 trillion in 2050. The increase reflects recurring purchases of servers, graphics processing units and other technology, which PwC expects operators to replace every four to six years.
Buildings Last Longer than Equipment Inside
PwC estimates that every $1 of construction spending effectively entails about $12 of future spending on information and communications technology. Equipment rises from 70% of annual data center capital spending in 2026 to 93% by 2050 in its model.
“The bulk of the spend doesn’t go towards the buildings,” the report says.
That division complicates how investors assess a data center. Its building and utility connections may have long useful lives, while its computing equipment could require several replacements over the same period. More powerful hardware may also call for upgrades to electrical systems and cooling. Lease terms, customer obligations and the ownership of equipment will affect who bears those costs.
PwC urges investors to assess property, power infrastructure and semiconductor exposure as distinct layers of an investment. It also cautions that demand from large cloud companies, specialized computing providers, enterprises and governments comes with different financial and operational risks. Strong demand for capacity alone does not establish that each project will earn an adequate return.
Power Access Could Decide Where Projects Go
Electricity is the principal physical constraint in PwC’s outlook. Transmission capacity, substation availability and transformer delivery times can delay projects even where financing and potential customers are available.
“Power sits at the top of the list,” the report says in assessing the forces that determine where investment flows.
The Americas account for $16.5 trillion of projected spending through 2050 in the central scenario. The United States alone represents $15.1 trillion, or about 48% of the global total, reflecting its concentration of cloud providers, AI developers and advanced-chip activity.
PwC projects $8.2 trillion for Asia-Pacific, $5.6 trillion for Europe, $1.1 trillion for the Middle East and $255 billion for Africa. Europe’s share is constrained by power, planning and regulatory challenges, the report says, although Nordic countries may benefit from renewable-heavy grids and lower cooling needs.
Computing workloads could pull investment in different directions. Training AI models can take place farther from users if a site has sufficient power and access to chips. Running models for customers may require facilities nearby to meet response-time, privacy or data-sovereignty requirements. PwC estimates that about 30% of workloads currently have localized requirements.
Trade and Sovereignty Alter Forecast
PwC tested what would happen if restrictions on advanced-chip trade intensified. Under that scenario, cumulative spending falls to $25.5 trillion, roughly $6 trillion below its central forecast. Annual spending drops to about half the central scenario’s level by 2030 before supply chains begin to adjust.
A separate scenario assumes more critical workloads must be served within their home countries. Global spending declines less sharply, to $29.5 trillion, but its location changes. U.S. cumulative spending falls by about $2.9 trillion, while Asia-Pacific spending rises 7% above the central projection as countries build more domestic capacity.
PwC assigns no probabilities to its alternative scenarios, and its model assumes no fundamental break in chip pricing. Its forecast describes the potential scale and distribution of investment, not the profits that investment will produce. For individual deals, power delivery dates, customer credit, lease terms and responsibility for replacing equipment remain critical to the outcome.