Updated:
AI infrastructure faces a revenue test: power supply and cash flow set the timetable
Electricity connections, operating receipts and financing schedules determine whether data-centre investment can support a lasting business. IEA scenarios and SEC accounting guidance explain what the figures can establish.
Research and compilation
On this page
Revised October 15, 2026: expanded the evidence on electricity supply and cash flow, and clarified the limits of long-term forecasts.
PwC’s central scenario projects US$31.6 trillion in cumulative global data-centre investment through 2050.1
For an individual project, the more immediate questions are when it can start delivering paid services and how it will meet payments before then. Electricity availability determines whether equipment can operate; financial statements distinguish money earned from customers from money supplied by lenders or shareholders. Together, those two perspectives offer a way to examine the distance between construction spending and a self-supporting business.23

Figure: AI-generated conceptual illustration of data centers, power facilities and infrastructure under construction.
A measured starting point and a conditional outlook
The International Energy Agency’s April 2025 report estimated that global data-centre investment reached about US$500 billion in 2024, nearly twice its 2022 level. It estimated electricity consumption at 415 terawatt-hours in 2024, about 1.5% of the world total. These are estimates for a past year; they cover data centres generally, rather than AI alone.2
The same report’s Base Case projected electricity use of about 945 TWh in 2030. AI was the main growth driver, alongside other digital services. That projection is a scenario published in 2025, not a measurement of demand already delivered or an updated reading of 2026 activity. Electricity consumption also cannot, by itself, reveal customer revenue, margins or returns on the capital used to provide the service.23
The uncertainty is material. Across the IEA’s cases, data-centre electricity demand in 2035 ranged from 700 to 1,700 TWh. Adoption, hardware and model efficiency, and energy-sector bottlenecks all influenced the results. Those cases describe alternative assumptions, not a confidence interval or a guarantee of demand for a particular operator.2
The power connection has its own timetable
A modest global share can conceal a large local constraint. The IEA found that nearly half of US data-centre capacity was concentrated in five regional clusters. It also estimated that around 20% of planned projects could face delays unless electricity-system risks were addressed. This is a conditional assessment of exposure to delay, not a count of projects already postponed.2
In advanced economies, building transmission lines can take four to eight years, according to the report. It also described lengthening waits for transformers and cables. Buying servers and completing a building therefore do not establish when the entire facility will receive usable power. The relevant evidence is the connection schedule, available capacity and the dependencies that must be completed before service begins.2
The financial implication is conditional but concrete: if customer receipts begin later while construction or financing payments still fall due, a project needs cash to bridge that interval. Whether it has enough depends on its own payment terms and funding arrangements. The IEA’s system-wide findings do not establish that a named company is short of cash or that every delayed project will fail.23
Revenue, profit and cash answer different questions
The US Securities and Exchange Commission’s financial-statements guide explains that an income statement reports revenue and the expenses deducted to arrive at profit. A cash-flow statement instead tracks cash entering and leaving the business. Profit and cash generation are related, but they are not interchangeable.3
That distinction matters when assessing expensive equipment. Depreciation spreads an asset’s cost across the periods in which it is used; purchasing machinery can require a cash outflow under investing activities. In the cash-flow reconciliation, depreciation is added back because it is non-cash for that period. The adjustment does not return the money previously spent on equipment, nor does it pay for the next purchase.3
The guide separates operating, investing and financing cash flows. Borrowing or issuing shares can increase cash available for construction, while repaying loans uses cash. Applied to an infrastructure operator, this means a growing cash balance is not enough to demonstrate that customers are funding expansion. The source of the cash and the timing of future obligations also matter.3
Efficiency changes the calculation, not just the electricity bill
The IEA’s High Efficiency Case put 2035 data-centre electricity demand about 20% below its Base Case. This comparison holds to that report’s assumptions; it is not a forecast of a matching reduction in industry revenue. More efficient models or hardware can alter the electricity needed for a workload without determining what customers will pay for it.2
The report also identified siting facilities where grid capacity is available and using servers, onsite generation or storage more flexibly as ways to ease integration. Such flexibility has trade-offs: the IEA noted that interrupting a highly capital-intensive AI facility can be costly. Spare computing capacity and backup assets create options, but they do not make electricity constraints disappear.2
For a project evaluation, these findings point to a linked set of questions: what service can be delivered within the power available, what operating costs accompany it, and what receipts remain after those costs? A reduction in energy per task is useful evidence, but its commercial effect depends on volume, pricing and the costs of providing the service. This is an application of the energy and accounting evidence, not a new market-size forecast.23
What would demonstrate a sustainable operating model?
A useful review would connect the commissioning record to the financial record: capacity actually receiving power, services sold, revenue and operating expenses, cash generated by operations, purchases of long-lived assets, and money raised or repaid. These measures should refer to the same business and comparable periods. Industry-wide electricity scenarios cannot substitute for a project’s accounts.23
The SEC guide also stresses financial-statement notes and management’s discussion of trends, risks and uncertainties. Those disclosures help explain accounting choices and why reported profit differs from cash flow. They provide context for evaluating expansion; no single figure settles the question.3
The practical test is whether a facility can deliver a paid service on a timetable that its funding can support, and whether operations can sustain the continuing costs of that service. Power delays can weaken that case, while efficiency and more flexible operation may improve it. The evidence supports examining those conditions project by project; it does not justify a blanket verdict on the commercial value of AI infrastructure.23
Source notes
-
2026-09-02 · Where $31.6 trillion of capex flows in the era-defining AI build-out · PwC. ↩
-
2025-04-10 · Energy and AI: Executive summary · IEA, Paris. CC BY 4.0. Selected text findings have been condensed and adapted. This is a work derived by Global Bole News from IEA material and Global Bole News is solely liable and responsible for this derived work. The derived work is not endorsed by the IEA or its Member countries in any manner. No IEA photographs, logos, cover art or third-party graphical components are reproduced. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12
-
2007-02-05 · Beginners’ Guide to Financial Statements · US Securities and Exchange Commission. Government-authored educational material, adapted under the SEC’s website dissemination policy; no SEC endorsement is implied. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10


