AI data-center investment is already enormous, and one like-for-like measure suggests its contribution to U.S. economic growth has exceeded a comparable measure at the height of the dot-com boom. But there is no sound single-number ranking of AI infrastructure against railways, electrification, or the internet: published figures measure different geographies, time periods, spending categories, and outcomes.
What the comparison can—and cannot—tell you
The short answer is that today’s AI-related buildout is a major investment wave, but claims that it is definitively larger than every earlier technology boom go beyond what the available comparisons establish. The strongest historical comparison is narrower: Federal Reserve Bank of St. Louis researchers estimate that selected U.S. investment categories associated with AI contributed more to real GDP growth in the first three quarters of 2025 than comparable information-technology categories did in 2000.
As an Amazon Associate I earn from qualifying purchases.
That is a comparison of contributions to economic growth, not a tally of all dollars spent on AI versus all dollars spent on the internet, railways, or electrification. Those comparisons require choices about what counts as investment, which countries and years to include, how to adjust for inflation, and whether to count equipment, buildings, networks, or related research. Different choices can produce very different totals.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallIt also helps to separate four things that are often blurred together: money already spent, a company’s capital expenditure, investment forecasts, and economic results. A forecast is not an invoice; a spending total is not a count of completed facilities; and infrastructure built does not automatically generate a particular level of profit or productivity.
How large is current data-center investment?
The biggest recent figures refer to different pools of spending. The International Energy Agency (IEA) estimated in its 2025 Energy and AI report that global data-center investment reached half a trillion dollars in 2024, nearly twice its 2022 level. That estimate covers the data-center sector globally; it is not an AI-only tally. In the same year, data centers consumed 415 terawatt-hours (TWh) of electricity, approximately 1.5% of global electricity use.
A separate IEA estimate measures capital expenditure by five large technology companies rather than investment across the whole data-center sector. In its 2026 Key Questions on Energy and AI report, the agency said the five companies’ combined capex exceeded $400 billion in 2025 and was expected to increase by a further 75% in 2026. The IEA said the 2025 figure exceeded global investment in oil and natural-gas production. It should not be read as a complete worldwide AI data-center total: it covers a selected group of companies, and company capex is not interchangeable with the IEA’s global sector estimate.
A February 2026 Federal Reserve note adds a U.S.-specific projection: data-center spending in the United States alone was expected to exceed half a trillion dollars in 2025. This is a forecast for U.S. spending, not the IEA’s realized global estimate for 2024 or the five-company capex figure. The figures are useful evidence of scale, but they cannot be added together or treated as three estimates of the same quantity.
| Measure | Figure and period | What it covers |
|---|---|---|
| Global data-center investment | $500 billion in 2024; nearly double the 2022 level | IEA estimate for the global data-center sector, not just AI facilities. |
| Data-center electricity use | 415 TWh in 2024; about 1.5% of global electricity | IEA estimate of energy consumption, not investment spending. |
| Capex by five large technology companies | More than $400 billion in 2025; a further 75% increase expected in 2026 | IEA estimate and projection for five companies, not total global data-center investment. |
| U.S. data-center spending | Expected to exceed $500 billion in 2025 | Federal Reserve forecast in a February 2026 note. |
These measures answer separate questions: how much the global sector invested in a completed year, what a small group of companies spent or was expected to spend, and what U.S. spending might reach. Keeping their scope and dates attached is more informative than combining them into a headline total.
What the dot-com comparison actually shows
A comparison of GDP-growth contributions
The most direct quantified comparison comes from St. Louis Fed authors, who examined selected investment categories in 2000—the height of the dot-com-era investment boom—and the first three quarters of 2025. For 2025, their AI-related grouping includes information-processing equipment, software, research and development, and data centers. Their comparable 2000 grouping includes information-processing equipment, software, and research and development. A separate data-center investment series was not available for 2000.
The authors estimate that these categories contributed 0.97 percentage points to U.S. real GDP growth in the first three quarters of 2025, compared with 0.81 percentage points for the comparable IT categories in 2000. Without the 2025 data-center category, the 2025 estimate falls to 0.90 percentage points. The categories accounted for 39% of total GDP growth in the first three quarters of 2025, or 36% excluding data centers, compared with 28% in 2000.
This result says that selected investment categories made a larger contribution to growth in the measured 2025 period than their counterparts did in 2000. It does not show that AI spending is a larger total-dollar investment than the dot-com buildout, nor that the two periods produced equivalent profits, assets, or long-term benefits. The categories are proxies for AI-related investment: not every dollar spent on software, equipment, or R&D in them is necessarily AI-related.
Free tools Windows power users keep installed
One-click scans. No signup required.
There are additional limits to the time comparison. The relevant Census data-center construction series begins in 2014, so the 2000 comparison cannot include a separately measured data-center category. The St. Louis Fed authors also imputed September 2025 data-center spending using July and August figures to estimate the third quarter. These qualifications do not erase the comparison, but they matter when interpreting a result that includes data centers for 2025 and not for 2000.
A broader view of investment as a share of the economy
A July 2026 Federal Reserve Board note offers a different historical yardstick: U.S. investment in intellectual property and equipment as a share of GDP. The share reached 11.5% in 2000. After rising sharply in 2025, it stood in the first quarter of 2026 only slightly below that earlier peak. The note describes the recent acceleration in this broad investment measure relative to GDP growth as comparable to the 1990s acceleration, while emphasizing that the current period began from a higher base level.
This series is broader than AI or data centers. It can show that investment in equipment and intellectual property is unusually strong relative to the economy, but it cannot isolate AI infrastructure’s share or establish that this boom is a replay of the 1990s. Use it as context for investment intensity, not as an AI spending total.
Can AI infrastructure be ranked against railways, electrification, or the internet?
PwC commissioned Oxford Economics to model data-center capital expenditure across 46 countries and territories and five regions. Its 2026 central scenario projects $31.6 trillion of global data-center capex through 2050, with a broad range of roughly $22 trillion to $50 trillion. These are long-range scenario estimates, not spending already committed or a prediction that one precise total will occur.
PwC characterizes its central scenario as orders of magnitude larger than the combined cost of U.K. and U.S. railway construction or the internet build-out. That is the consulting firm’s comparison, not a harmonized historical ledger in which spending across each era has been consistently measured on the same geographic, temporal, and asset basis. The report itself recognizes that comparisons across eras are imperfect. A multi-decade forecast in current investment categories cannot be ranked cleanly against an older infrastructure total without aligning the duration, geography, price basis, and kinds of assets counted.
There is no single established historical series here that ranks today’s AI data-center investment against railways, electrification, and the internet using a common method. The careful conclusion is not that the AI buildout is smaller or larger; it is that the available scenarios convey its potential scale, while the evidence does not support a definitive all-time ranking.
Why this investment cycle may have a different rhythm
A data center combines assets with different useful lives. The site, building shell, and some supporting infrastructure may remain in place much longer than the computing hardware inside them. Servers, storage, networking equipment, CPUs, and GPUs need replacement or upgrades as workloads and technology change.
In its 2026 modeling, PwC says servers and GPUs typically turn over every four to six years. It projects ICT equipment’s share of data-center capex to rise from 70% in 2026 to 93% in 2050. These are the report’s general turnover framing and forecast shares, not a rule that every operator replaces every component on a fixed schedule. They nevertheless describe an important difference from an investment wave dominated by one-time construction: continued hardware refresh can require substantial recurring capital even after many buildings have been completed.
That distinction cuts both ways. A continuing refresh cycle could sustain investment beyond the initial construction rush. It could also expose investors to repeated replacement decisions, changing chip capabilities, utilization risk, and the possibility that equipment becomes economically outdated before the surrounding facility does. Total projected capex over decades therefore includes more than building a stock of infrastructure once; it can include replacing and upgrading equipment over time.
Best Value
- Family farms not data design for people against AI server farms, data center expansion, rural land buyouts, corporate agriculture, and industrial tech development replacing farmland and open space. Rural conservation and anti data center message.
- AI protest design for farmers, land conservation supporters, anti AI activists, sustainability groups, environmental advocates, rural communities, and people opposing server farm construction, power grid strain, and farmland destruction.
- Lightweight, Classic fit, Double-needle sleeve and bottom hem
What could prevent investment from becoming useful capacity?
Spending plans do not guarantee a working data center. The IEA’s 2026 report describes constraints involving electricity supply, grid connections, advanced chips, and financing. Those bottlenecks can affect whether projects are built on schedule, how much compute they can deliver, and whether the installed capacity is used as planned.
Energy demand illustrates the scale of the conversion challenge. The IEA reports that global data-center electricity consumption grew 17% in 2025, while electricity use in AI-focused data centers grew 50%. It projects total data-center consumption to rise from 485 TWh in 2025 to 950 TWh in 2030, with consumption in AI-focused data centers tripling. Those are energy estimates and projections, not investment totals. They help explain why power supply and grid access have become central to the buildout rather than incidental operating details.
The IEA also says investment has grown too large to be funded from company balance sheets alone and will depend on capital markets. That ties construction to expected returns and broader financing conditions. If power connections, chips, or financing are unavailable, announced capex may not translate on time into completed capacity. Even a completed facility is not proof of profitable utilization or economy-wide productivity gains.
Recommended Free Tools
What the history of overinvestment can—and cannot—warn about
The previous IT investment boom provides a mechanism for risk, not a reliable script for the future. A Federal Reserve note observes that the earlier boom ended in the early 2000s with unproductive capital, including “dark fiber”—network capacity that had been laid but was not productively used. This is a reminder that investors can build infrastructure ahead of demand, and that assets can remain underused when expectations do not materialize.
It does not establish that AI data centers will repeat that outcome. Demand, technology, financing structures, power constraints, and the composition of assets differ across cycles. The relevant question is whether completed capacity finds sustainable use and returns, not whether a new technology boom resembles an earlier one in a headline spending figure. A large investment wave can create valuable long-lived infrastructure and still include projects that prove uneconomic.
A practical way to read the next AI-investment headline
When a new number appears, identify the measure before deciding what it says about the size of the boom. A few checks prevent most misleading comparisons:
- Is it observed spending, capex, an announcement, or a forecast? The IEA’s 2024 global investment estimate, its five-company capex figure, and PwC’s through-2050 scenario are different kinds of evidence.
- What geography and entities does it cover? A global sector estimate, a U.S. forecast, and spending by five large companies do not have the same denominator.
- What is counted? Buildings, data-center equipment, software, R&D, and broad IP-and-equipment investment are not interchangeable categories.
- What is the comparison measuring? Total dollars, GDP share, GDP-growth contribution, and electricity demand answer distinct questions; one cannot stand in for another.
- Are the time span and price basis comparable? A 25-year forecast should not be ranked against a historical total without accounting for duration, geography, and how prices are treated.
- Did the investment produce capacity and use? Grid access, chips, financing, completion, utilization, and returns determine what spending becomes in practice.
The IEA’s 2025 Energy and AI report puts the energy connection succinctly: “There is no AI without energy.” For investment comparisons, the companion point is that money alone does not describe the outcome. The strongest conclusion supported by the available measures is that AI-related investment has become exceptionally prominent and is contributing substantially to current U.S. growth; whether its total scale ultimately surpasses earlier infrastructure revolutions cannot be settled by the headline figures now available.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
