Global investment in data centres could reach USD 31.6 trillion by 2050 as the adoption of artificial intelligence (AI) turns computing infrastructure into a multi-decade capital spending cycle, according to PwC.
The consulting firm’s Global Data Centre Outlook said annual spending on data centre capital is expected to rise from USD 800 billion in 2026 to USD 1.1 trillion by 2030 and USD 1.8 trillion by 2050.
The forecast, modelled by Oxford Economics across 46 countries and territories, puts the central estimate at USD 31.6 trillion, with an upside of nearly USD 50 trillion if AI adoption accelerates.
The scale of the investment reflects a difference between the AI infrastructure boom and earlier infrastructure waves such as railways, electrification and the internet.
While those build-outs tended to require heavy spending upfront before investment tapered as networks matured, AI infrastructure will require repeated spending because the technology inside data centres becomes obsolete much faster than the buildings themselves.
PwC estimates servers, graphics processing units, storage, networking equipment and other information and communications technology equipment will account for 93% of total data centre investment by 2050, compared with 70% today. GPUs and servers generally need to be refreshed every four to six years, meaning facilities could require several rounds of technology investment over their operating lives.
That makes the AI build-out not a property or construction story but a continuing technology and capital-allocation cycle. PwC estimates that every USD 1 spent on construction effectively commits the market to about USD 12 of future ICT equipment expenditure.
The United States is positioned to capture the largest share of the spending. PwC forecasts USD 15.1 trillion of cumulative investment in the US through 2050, equivalent to 48% of the global total. The country’s lead reflects its role in advanced semiconductors, as well as its concentration of hyperscalers, AI model developers, capital and talent.
Asia-Pacific is expected to attract USD 8.2 trillion, with China and India providing major sources of incremental demand. Europe is forecast to receive USD 5.6 trillion, while the Middle East accounts for USD 1.1 trillion and Africa for USD 255 billion.
The regional picture, however, could change sharply depending on technology supply, government policy and the speed of AI adoption. PwC’s central forecast sits within a broad range of roughly USD 22 trillion to USD 50 trillion through 2050.
One of the biggest constraints is not capital but electricity. PwC stated that access to affordable, reliable, and increasingly low-carbon power at scale is crucial for the future of data centres.
The transmission capacity, substations, and transformers required for these projects can have lead times of several years, which may cause delays.
Operators are considering on-site generation, but PwC said such a solution does not remove the need for grid investment. Connectivity, security, access to GPUs and technology ecosystems, policy certainty and community consent will also determine which markets can convert demand into actual capacity.
The power constraint could become particularly important as AI workloads require greater computing density and electricity consumption than traditional cloud services. Training large models can be located in markets offering cheap, reliable power and access to chips, while inference workloads are more sensitive to latency, data access, privacy and sovereignty requirements.
Trade restrictions pose another risk. PwC modelled a scenario in which tighter export controls disrupt global semiconductor supply chains. Under that scenario, annual investment falls to roughly half the central forecast by 2030 before recovering as supply chains adjust. Even after the recovery, cumulative spending through 2050 would reach only USD 25.5 trillion, roughly USD 6 trillion below the baseline.
The impact would be uneven. The Middle East would be among the most exposed regions, with cumulative investment falling by 29% in the export-control scenario, particularly affecting Saudi Arabia, Qatar and the UAE, where projects depend heavily on access to advanced chips and internationally mobile AI workloads. Africa’s cumulative investment would fall from USD 255 billion to about USD 193 billion.
Digital sovereignty creates a different outcome. If governments and regulated industries increasingly require sensitive workloads to be hosted domestically, PwC expects investment to be redistributed rather than dramatically reduced. Global cumulative spending would decline only slightly, to about $29.5 trillion, while countries with strong domestic demand but relatively limited existing capacity would gain investment.
Asia-Pacific would be the largest absolute beneficiary in that scenario, with cumulative investment rising 7% above the central forecast. India, Vietnam, Indonesia, the Philippines and Thailand would benefit as domestic demand encourages local capacity. Africa would also gain proportionally, with investment rising 12% to roughly USD 284 billion.
For investors, the implications extend beyond traditional data centre real estate. The changing capital stack means returns and risks will increasingly depend on chip replacement cycles, power availability, equipment procurement, cooling requirements and the ability to secure long-term customers.
The market is also becoming more diverse. Alongside hyperscalers, demand is coming from neocloud providers, AI model developers, inference platforms, enterprises and governments. Each has different requirements and credit risks, potentially creating more specialised data centre markets.
PwC said that about 30% of workloads currently have localised requirements and that this share is growing, reinforcing the case for distributed capacity. At the same time, AI training can gravitate towards locations with abundant electricity, advanced chips, technical talent and established ecosystems.
Community resistance is another emerging constraint. Bloomberg, citing Data Center Watch, reported that at least 75 projects worth USD 130 billion were halted or put on hold because of community opposition in the first quarter of the year. Concerns over electricity bills, water, land use and pressure on local grids are increasingly turning data centre development into a political issue.
The investment outlook therefore hinges on more than the availability of money. Governments that expand grids, accelerate planning, improve regulatory certainty and secure low-carbon power could attract a disproportionate share of the AI economy. Markets that fail to address those bottlenecks risk losing projects despite strong demand.
PwC’s forecast ultimately indicates a structural shift in the financing and valuation of data centre assets.
The buildings may last for decades, but the computing technology inside them will need continuous renewal.
As AI adoption expands, the resulting investment cycle could make data centres one of the world’s largest and most persistent channels for technology infrastructure capital globally through the middle of the century.