PwC puts the AI build out at 31.6 trillion dollars to 2050 and makes the Middle East its fastest growing region

By The Edge Research Team

A capital expenditure forecast for data centres published on 2 September 2026 runs to 2050 and covers 46 countries and territories. Its central figure is 31.6 trillion dollars of cumulative investment between 2026 and 2050, and the number carries two qualifications that matter more than the headline.

It is capital expenditure only, covering buildings and structures on one side and the servers, graphics processors, storage and networking hardware installed in them on the other. The methodology defines those two components and no others, so on our reading running costs, and electricity in particular, sit outside the total. And every figure is stated in real 2025 dollars at 2025 exchange rates, so it is not a nominal total inflated by 24 years of price growth.

The modelling was done by Oxford Economics in three stages: demand anchored to a logistic adoption curve and converted into megawatts of information technology load drawing on sources including the International Energy Agency and the Uptime Institute, supply distributed across countries by a regression on 11 variables, of which six are named, among them electricity prices, transmission and distribution losses, hyperscale cloud presence, semiconductor trade, information technology service exports and cybersecurity maturity, then capital expenditure derived from the Turner and Townsend data centre construction cost index for 2025 to 2026.

Spending keeps rising every year, and on our calculation the rate at which it rises more than halves

Three points are quoted in the text for the annual path: about 800 billion dollars of data centre capital expenditure in 2026, 1.1 trillion in 2030, and 1.8 trillion in 2050. The underlying chart carries the same series unrounded, at 828.4 billion, 1.056 trillion and 1.808 trillion.

The report’s own reading of those points is that this cycle does not behave like earlier ones. “Unlike traditional infrastructure booms, which taper off after the initial build out, AI infrastructure investment is expected to accelerate as chips and other ICT equipment require upgrades every few years.” Annual spending, it says, accelerates over time, and the reason is the refresh: a data centre is described as essentially a chip replacement subscription with a building wrapped around it, where you build the box once and refill it every four to six years.

That is a claim about the level of annual dollars, and on the published path it is correct. Spending in 2050 is 2.18 times spending in 2026 on our calculation, and it is still climbing at the end of the horizon.

Run the same series as compound rates, however, and a second pattern appears that the report does not print. On the unrounded chart values, annual spending compounds at 6.25 percent between 2026 and 2030, then at 2.73 percent between 2030 and 2050, on our calculation. The rate across the second stretch is 3.53 percentage points slower than across the first, a fall of 56 percent in the pace of growth, and the whole 24 year period averages 3.31 percent. This is our arithmetic on the forecast’s own numbers, not a finding it makes.

The choice of inputs matters here and we state it because it changes the answer. Computed instead on the three rounded figures quoted in the prose, the same exercise gives 8.29 and 2.49 percent and a fall of 70 percent, on our calculation. The unrounded series is the sounder basis and it is the one carried above; a reader working from the quoted round numbers alone would overstate the deceleration by roughly a quarter.

Both things are true at once and the distinction is the practical one. A supplier reading the level sees a market that never stops growing. A supplier reading the rate sees the difference between a market expanding above 6 percent a year and one expanding below 3 percent, on our figures, which carries very different implications for capacity planning and for how much of any given year’s growth is available to be won.

What changes underneath is the composition. Information and communications technology equipment rises from 70 percent of capital expenditure in 2026 to 93 percent by 2050, which leaves buildings and structures falling from 30 percent to 7 percent as the complement, on our calculation. By mid century the money is going into installing and replacing hardware rather than constructing the shell around it, on a refresh cycle put at 4 to 6 years for servers and processors.

A separate ratio in the report measures the same dependence over the life of a site rather than at a point in time: every dollar of construction capital expenditure is said to commit the market to roughly 12 dollars of later spending on information and communications technology equipment. That is a lifetime commitment against a single year’s build, and it is not the same arithmetic as the 93 to 7 split above.

The Middle East is small in dollars, first in growth, and most exposed to a chip shock

The 31.6 trillion dollars splits across 5 regions. The Americas take 16.5 trillion dollars, of which the United States alone accounts for 15.1 trillion, or 47.7 percent of the global total on our calculation. Asia Pacific takes 8.2 trillion dollars, Europe 5.6 trillion, the Middle East 1.1 trillion and Africa 255 billion.

The Middle East is 3.5 percent of the global total and 4.3 times the size of Africa’s share, on our calculation. That share is described as “roughly in line with the region’s contribution to GDP” and, more usefully, as the fastest growing in the forecast on a compound basis. The reasons given are a low existing installed base and an ability to compress building timelines “by aligning energy, capital, planning, and developer pipelines through a single coordinated front door.”

That speed comes with a specific exposure. The regional build is graphics processor heavy and aimed at attracting internationally mobile workloads as well as serving regional demand, “an approach that delivers the largest CAGR in the central case but concentrates the region’s exposure to anything that disrupts global chip supply.”

The report then quantifies exactly that. Under a scenario in which export controls tighten, global cumulative investment falls to about 25.5 trillion dollars, 6.1 trillion below the central case and 19.3 percent lower on our calculation. The Middle East is the most proportionally exposed region in the study, with cumulative capital expenditure down 29 percent, which takes the region to about 781 billion dollars on our calculation, concentrated in Saudi Arabia, Qatar and the United Arab Emirates.

A separate digital sovereignty scenario, in which investment shifts toward countries with strong domestic demand, costs the region far less: 1.1 trillion dollars becomes 1.0 trillion, a fall of 9 percent. Globally that scenario barely dents the total, taking cumulative capital expenditure from 31.6 trillion dollars to 29.5 trillion, a reduction of 6.7 percent, and the report calls it “redistribution rather than reduction.” Africa gains the most in proportional terms, rising about 12 percent to 284 billion dollars, Asia Pacific rises 7 percent and Europe 2 percent, while the Americas fall 17 percent, the United States alone losing around 2.9 trillion dollars. Within the Gulf the same scenario names Kuwait and Oman among the markets that gain from onshoring domestic workloads.

The contrast between minus 29 and minus 9 percent is the most useful pair of numbers in the study for a Gulf reader, because it separates the risk the region actually carries from the one it does not. In dollars, sovereignty costs the Middle East about 100 billion dollars of cumulative capital expenditure and a chip supply shock costs it about 319 billion dollars, a gap of roughly 219 billion on our calculation.

One feature of the export control case is easy to miss and cuts against reading it as a permanent loss. Annual capital expenditure falls to roughly half the central scenario by 2030, but by 2050 it runs 8 percent above the central case. The shortfall is front loaded: the cycle is delayed and then partly recovered, and the 6.1 trillion dollar cumulative gap is the area between those two paths rather than a level the market never reaches.

Europe, meanwhile, is judged to be “punching below its economic weight” at 17.7 percent of global capital expenditure, against Asia Pacific at 25.9 percent, or 1.46 times the European figure on our calculation.

The range is wider than the headline suggests

The 31.6 trillion dollar figure is the central case in a cone running from about 22 trillion dollars to about 50 trillion. The upside is 1.58 times the central case and the low end is 0.70 times it, on our calculation, so the published range spans 2.27 times from bottom to top. The Americas carry the largest absolute swing in either direction, rising to 27.1 trillion dollars under faster adoption, and Asia Pacific the largest proportional one, rising 69 percent and falling 34 percent. Europe moves least on the upside at 23 percent.

No probability weighting is attached to any of it. The central scenario “reflects the most likely trajectory under current assumptions about AI adoption, supply chain conditions, and the policy environment” while the alternatives “describe coherent departures from that trajectory rather than ranked risks.” The upside case assumes every country follows the historic growth rate of uptake recorded by the world’s current leader, the United Arab Emirates, drawing on Microsoft’s 2025 adoption work.

Two further caveats come from the study rather than from us. The model assumes no structural break in chip pricing. And quantum computing sits outside the modelled range entirely rather than inside it as a weighted scenario, on the grounds that a shift there would change the workload mix the whole forecast rests on.

What the report does not say

Power is the constraint the study returns to throughout, and it is also the number it withholds. There is no projected electricity demand in terawatt hours, no share of total generation and no required generating capacity anywhere in the publication. The only capacity unit used is megawatts of information technology load, and that is a modelling input drawn from the International Energy Agency and the Uptime Institute, not a power demand forecast.

What is offered instead is qualitative and specific: “Transmission capacity, substation availability, and multiyear transformer lead times are becoming the bottlenecks that determine which projects break ground and when.” Operators bringing their own generation to sites helps individual projects, it says, but does not remove the need for grid build out at market scale.

The publication contains no electricity demand figure of any kind.

Cumulative data centre capital expenditure between 2026 and 2050 in the central scenario, in real 2025 dollars. Shares are ours, computed against the regional total of 31.655 trillion dollars, which is the sum of the 5 published rows. The report’s own unrounded regional series sums exactly to its global total; the 55 billion dollar gap against the 31.6 trillion headline is an artefact of the one decimal rounding used in this table, not a discrepancy in the forecast. The United States alone accounts for 15.1 trillion of the Americas figure, which is 47.7 percent of the global total on our calculation and matches the report’s own description of around 48 percent.

The published scenarios, ranked by global capital expenditure. The two adoption figures for the Middle East are read from the report’s regional chart rather than its text. The export control figure is ours, applying the 29 percent regional reduction the report states to its own 1.1 trillion dollar central case. No probability weighting is attached to any scenario.

Cumulative capital expenditure to 2050 under the scenario in which chip export controls tighten. All figures are published except the Middle East, which is ours from the stated 29 percent reduction. In proportional terms the reductions are Americas 16 percent, Asia Pacific 22 percent, Europe 23 percent, Africa 24 percent and the Middle East 29 percent, so the region carrying the largest relative loss is the second smallest in absolute dollars.

The 3 points on the annual spending path as the report’s own chart carries them unrounded, in real 2025 dollars; its text quotes the same three points as about 800 billion, 1.1 trillion and 1.8 trillion dollars. The compound growth rates are ours, computed between these endpoints, and across the full period from 2026 to 2050 the average is 3.31 percent a year. Computed on the rounded figures instead, the same three rates read 8.29, 2.49 and 3.44 percent.

Sources: PwC, Oxford Economics.

Get Complimentary Insights

Follow us