AI infrastructure spending forecasts compared: every major firm raised its 2030 number in 2026
Every major AI infrastructure spending forecast went up in 2026, and the biggest increases landed on 2029 and 2030. Dell’Oro Group nearly doubled its 2030 data center capex outlook in six months. Gartner raised its 2030 data center systems forecast 58% in a single quarter. IDC raised its 2029 AI infrastructure forecast 42% in nine months.
The firms still disagree on how big the build-out gets. For 2030, the published annual figures run from IDC’s $1.21 trillion to Citi’s $3.8 trillion, a 3.1x gap. Most of that gap comes from what each firm counts, not from different views of demand.
This post puts the public numbers from 12 firms side by side. They are Gartner, IDC, Dell’Oro, Goldman Sachs, Citi, Morgan Stanley, McKinsey, Bain, Forrester, JLL, the IEA and BloombergNEF. It shows what each one measures, how far each has moved, what the five largest buyers are actually spending, what the power grid can deliver and what the build-out has to earn back. Where a firm published only a range or a scenario, I say so.
Every revision this year went up
Dell’Oro’s February 2026 report put worldwide data center capex at $1.7 trillion by 2030. Its August report raised that to more than $3 trillion. “Our 2030 data center capex outlook has nearly doubled since the January 2026 forecast,” said Baron Fung, vice president of research at Dell’Oro. He cited higher hyperscale capex guidance, higher projections for data center power capacity and higher commodity costs.
Gartner made the largest single-quarter move. Its 3Q26 Market Databook raised 2030 data center systems spending by $794.9 billion, to $2.16 trillion. That puts the 2Q26 figure at $1.36 trillion, so the forecast rose 58% in one quarterly update. Gartner says hyperscalers, sovereign AI programs and neocloud providers are all expanding compute capacity faster than expected.
IDC has raised its 2029 forecast twice. Its October 2025 release projected $758 billion in AI infrastructure spending for 2029. In April 2026 it said 2029 would exceed $1 trillion. In July it put 2029 at $1.08 trillion and 2030 at $1.21 trillion.
Wall Street raised the current year
Wall Street moved the most on the current year. Goldman Sachs reported in December 2025 that the consensus estimate for hyperscaler capex in 2026 was $527 billion. By August 2026, Goldman put the consensus at about $800 billion, a 52% increase. Goldman’s December note also showed how consistently the consensus has been too low. At the start of both 2024 and 2025, analysts expected roughly 20% capex growth, and actual growth topped 50% both years.
Near-term numbers moved least. IDC raised its 2026 forecast from $487 billion to $497 billion between April and July. Gartner’s 2026 data center systems figure rose $56.8 billion this quarter. Purchase orders and earnings guidance anchor the current year. The outer years are where forecasters are changing their minds.
BloombergNEF and Citi raised theirs too
Two more forecasters moved in the same direction. BloombergNEF raised its forecast for US data center capacity in 2030 to 118 GW in July, 52% above its December 2025 outlook, and its 2035 forecast 83% to 194 GW. The announced US pipeline grew by another 101 GW over those seven months. Citi raised its 2026 to 2030 global AI capex estimate from $8 trillion to $8.9 trillion in March, and its AI revenue estimate for the same window from $2.8 trillion to $3.3 trillion.
Hyperscaler guidance is the anchor
Every forecast in this post leans on what five companies say they will spend. Alphabet raised its 2026 capex guide on July 22 to $195 billion to $205 billion, up $15 billion at both ends. Meta raised the low end of its range to $130 billion to $145 billion. Microsoft guided calendar 2026 to roughly $175 billion, and coverage of Goldman’s September note cites $175 billion to $190 billion. Management said about $25 billion of that reflects higher component prices rather than new capacity. Amazon gave no full-year guide, and the same Goldman coverage puts it at $200 billion to $220 billion. Oracle’s fiscal 2027 target is $50 billion to $70 billion.
Added up, the ranges come to $750 billion to $830 billion. Goldman’s smoothed calendar-year estimate for the five is about $800 billion, nearly double the roughly $413 billion they spent in 2025. The IEA notes that the capex of five technology companies is now larger than global investment in oil and natural gas production.
Backlogs and chip sales point higher
The order books explain why guidance keeps rising. Oracle’s remaining performance obligations reached $664 billion in its fiscal first quarter, after more than $30 billion of new AI cloud contracts, and it delivered 850 MW of new capacity. Citi puts combined hyperscaler backlog at about $1.7 trillion at the end of the second quarter, up 151% in a year. On the supply side, Nvidia reported $89.0 billion in data center revenue for the quarter ended July 26, up 117% from a year earlier.
The 2030 numbers measure different things
Each forecast counts something different. IDC’s $1.21 trillion counts only server and storage systems running AI workloads. Servers made up 97.6% of IDC’s AI infrastructure spending in the first quarter of 2026. IDC does not count networking, power or buildings.
Gartner’s $2.16 trillion covers all servers, external storage and enterprise network equipment, AI and general-purpose combined. Servers alone account for $1.97 trillion of it. Gartner’s dataset does not separate AI servers from other servers.
Citi and Dell’Oro set the top of the range
Citi’s $3.8 trillion is the largest annual figure. Its September research projects global AI-related capex rising from $1 trillion this year to $3.8 trillion by 2030, counting every spender rather than equipment categories. Morgan Stanley put 2026 at a little less than a trillion dollars of AI capex, close to Citi’s starting point.
Dell’Oro’s more than $3 trillion is the broadest equipment-and-facilities figure. It covers servers, storage, network and physical infrastructure bought by every class of buyer. Dell’Oro breaks out the top four US cloud providers, the top three Chinese cloud providers, AI model builders and neoclouds, colocation, telecom and enterprise. Fung said the top four US hyperscalers alone could account for about half of global capex.
Gartner publishes two AI infrastructure numbers
Gartner itself publishes two different AI infrastructure numbers. Its September AI spending forecast puts AI infrastructure at $1.48 trillion in 2026 and $1.98 trillion in 2027. That category adds AI-optimized IaaS, AI semiconductors and AI devices. The 2026 figure is $606 billion more than Gartner’s data center systems figure for the same year.
| Firm | What it measures | Latest long-range figure | Earlier figure | Scope |
|---|---|---|---|---|
| Citi Research | Global AI-related capex | $3.8T (2030) | $8.0T for 2026-2030, raised to $8.9T in March 2026 | AI-related capex by all spenders, from $1T in 2026 |
| Dell’Oro Group | Worldwide data center capex | >$3.0T (2030) | $1.7T for 2030 in Feb. 2026 | Servers, storage, network and physical infrastructure, all buyers |
| Gartner | Data center systems end-user spending | $2.16T (2030) | $1.36T for 2030 in 2Q26 | Servers, external storage, enterprise network equipment, AI and general-purpose |
| Gartner | AI infrastructure (AI spending forecast) | $1.98T (2027) | n/a | AI-optimized servers and IaaS, AI network fabric, AI semiconductors, AI devices |
| Goldman Sachs Research | Capex, five largest US hyperscalers | $1.4T (2028) | Consensus $527B for 2026 in Dec. 2025 | Total company capex for Amazon, Alphabet, Microsoft, Meta and Oracle |
| Goldman Sachs Global Institute | Baseline AI capex scenario | $1.6T (2031) | n/a | Compute, data centers and power, modeled from Nvidia data center revenue |
| IDC | AI infrastructure spending | $1.21T (2030) | $758B for 2029 in Oct. 2025 | Server and storage systems running AI workloads |
| McKinsey | Cumulative data center capex | $6.7T (through 2030) | n/a | AI and traditional data centers, including chips, buildings and power |
| Morgan Stanley | Global data center capex | $2.9T (2025-2028) | n/a | Data center construction, with a financing split by source |
| JLL | Data center sector investment | Up to $3T (2026-2030) | n/a | About 100 GW of new capacity, real estate, fit-out and debt |
Each firm’s latest public figure as of October 2026. Gartner data center systems from the 3Q26 Market Databook dataset, current U.S. dollars. Earlier figures from each firm’s prior public release.
Five curves for the same build-out
In 2026 the forecasts sit close together. IDC puts AI infrastructure at $497 billion. Gartner puts data center systems at $878.5 billion. Goldman puts capex at the five largest US hyperscalers at about $800 billion and global AI investment at $1 trillion. Dell’Oro said in February that global data center capex would approach $1 trillion.
By 2030 the curves split. Gartner’s growth slows sharply, from 71.5% in 2026 to 8.7% in 2030 in constant currency. IDC’s grows about 30% a year from 2025 to 2030. Dell’Oro’s broader figure keeps climbing to more than $3 trillion, and Citi’s AI capex line reaches $3.8 trillion.
Goldman’s year-by-year view is the steepest in the near term. It expects capex at the five largest US hyperscalers to reach $1.2 trillion in 2027 and $1.4 trillion in 2028. Those five companies alone would spend more in 2028 than IDC expects the whole world to spend on AI servers and storage in 2030.
Multi-year totals run from $2.9 trillion to $8.9 trillion
In April 2025, McKinsey estimated that data centers would need $6.7 trillion in capex worldwide by 2030. Of that, $5.2 trillion would go to AI data centers, within a scenario range of $3.7 trillion to $7.9 trillion. McKinsey’s total includes chips, buildings and power.
Gartner’s latest forecast puts equipment spending alone above that. Data center systems total $8.07 trillion from 2026 to 2030 in the 3Q26 dataset. Gartner’s number includes general-purpose servers and leaves out buildings and power. Even so, the comparison shows how far expectations have moved in 18 months.
The Goldman Sachs Global Institute’s baseline puts cumulative AI capex at about $7.6 trillion from 2026 to 2031 across compute, data centers and power. Goldman says this is a scenario built on stated assumptions, not a forecast. It models annual AI capex rising from $765 billion in 2026 to $1.6 trillion in 2031.
Citi’s $8.9 trillion for 2026 to 2030 is the largest cumulative figure. Morgan Stanley estimates about $2.9 trillion of global data center capex from 2025 to 2028. JLL expects the sector to need up to $3 trillion over five years to add roughly 100 GW, nearly doubling global capacity from 103 GW to 200 GW by 2030.
Forrester does not publish a data center total, but its global tech forecast points the same way. It expects AI-specialized computers, including AI servers, to account for more than 80% of computer equipment spending by 2030, up from 43% in 2024.
Power is the constraint every forecast names
The IEA’s July update projects global data center electricity use roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, about 3% of world demand. Consumption at AI-focused data centers grew 50% in 2025 and triples by 2030 in the IEA’s projection. Citi expects new AI-related power capacity of 19 GW in 2026, 29 GW in 2027 and 45 GW in 2028. JLL expects AI to account for half of global data center capacity by 2030, up from about a quarter in 2025.
The grid is where the dollar forecasts and the physical forecasts part ways. BloombergNEF calculates a 63 GW gap by 2033 between what AI chip shipments imply the US should build and what its own forecast says will get built, because of energy constraints. That gap is the clearest signal of which outer-year spending numbers are most at risk.
The return question got bigger
Higher spending forecasts raised the bar for returns too. Bain’s 2026 Global Technology Report says funding AI’s compute demand would require $6 trillion in annual revenue by 2031. Existing consumer and enterprise AI applications could bring in $1.2 trillion to $1.8 trillion of that, leaving $4.2 trillion to come from new revenue. A year earlier, Bain’s estimate was $2 trillion by 2030 with an $800 billion shortfall. “AI infrastructure is being built well ahead of the demand curve,” said David Crawford, chairman of Bain’s global technology practice.
Goldman frames the same question at the company level. In a September 25 note, strategists led by Ryan Hammond estimated the five largest US hyperscalers need roughly $300 billion in annual AI revenue to break even on about $800 billion of 2026 spending. That is the steady-state revenue needed to cover depreciation and running costs, not a one-year payback. Goldman estimates AI cloud revenue is running only about $70 billion above the pre-AI trend, leaving a gap of roughly $230 billion a year. It also expects the four largest cloud providers to issue about $250 billion in bonds in 2026, roughly a third of capex, up from $108 billion in 2025.
Who funds the build-out
Morgan Stanley’s financing map shows how much of the build-out sits outside hyperscaler cash flow. Of $2.9 trillion in data center capex through 2028, it expects $1.4 trillion to come from hyperscaler cash flows and $800 billion from private credit. The remaining $700 billion comes from private equity, venture and sovereign capital, corporate debt and securitized credit. Citi offers the counterpoint, saying hyperscalers are generating cash returns of roughly 30% on invested capital from their infrastructure spending.
The forecasters name the same constraints. Dell’Oro says growth depends on “the sustainability of investment, power availability, and supply chain conditions.” It adds that enterprise investment “remains constrained by uncertain AI returns.” Gartner warns that memory prices have more than doubled over the past 12 months and expects the pressure to last through 2027.
Each firm’s forecast has risen with hyperscaler guidance. If that guidance flattens, the outer-year numbers, which moved most this year, are also the ones most likely to come down.
How to read these forecasts
- Match the scope before comparing numbers. Check whether a figure counts AI-only systems, all data center equipment, or equipment plus buildings and power. The same build-out can measure $1.2 trillion or $3.8 trillion in 2030.
- Weight near-term figures more heavily. 2026 numbers moved least this year because orders and guidance anchor them. Forecasts for 2029 and 2030 changed by 42% to 76% within a year.
- Track hyperscaler guidance first. Dell’Oro and Goldman both tie their revisions to hyperscaler capex plans, and the top four US hyperscalers could make up half of global capex.
- Separate spending from returns. None of these forecasts measures whether the spending pays off. Bain’s $4.2 trillion revenue gap is a different question from how much gets spent.
- Watch power before price. BloombergNEF’s 63 GW US gap and the IEA’s doubling of data center electricity use set the physical ceiling on outer-year dollar forecasts.
- Expect the next edition to move again. Goldman’s data shows consensus estimates fell short of actual capex growth in 2024 and 2025.
Frequently asked questions
How much will the world spend on AI infrastructure in 2026? It depends on scope. IDC forecasts $497 billion for AI server and storage systems. Gartner forecasts $878.5 billion for all data center systems and $1.48 trillion for its broader AI infrastructure category. Goldman Sachs estimates about $800 billion in capex at the five largest US hyperscalers.
How much will be spent on data centers by 2030? Dell’Oro forecasts more than $3 trillion in worldwide data center capex in 2030. Citi projects $3.8 trillion in global AI-related capex in 2030. Gartner forecasts $2.16 trillion in data center systems spending. IDC forecasts $1.21 trillion in AI infrastructure.
Why do AI infrastructure forecasts differ so much? They count different things. IDC counts AI server and storage systems. Gartner counts all data center equipment. Dell’Oro adds physical infrastructure and every class of buyer.
How much are the hyperscalers spending in 2026? The 2026 ranges for Amazon, Alphabet, Microsoft, Meta and Oracle add up to $750 billion to $830 billion. Goldman’s calendar-year estimate is about $800 billion, nearly double the roughly $413 billion they spent in 2025.
How much revenue does AI need to justify the spending? Bain estimates $6 trillion in annual revenue by 2031, with $4.2 trillion still to be found beyond existing AI applications. Goldman estimates the five largest US hyperscalers need about $300 billion in annual AI revenue to break even on 2026 spending.
How much power will data centers use by 2030? The IEA projects 950 TWh worldwide in 2030, about double 2025. BloombergNEF forecasts 118 GW of US data center capacity in 2030.
Which forecast was revised the most in 2026? Dell’Oro’s 2030 data center capex outlook, which rose from $1.7 trillion in February to more than $3 trillion in August.
How I built this analysis
All figures come from each firm’s public press releases, earnings releases, research pages and news coverage of their notes, linked below, plus the Gartner 3Q26 Market Databook and its dataset. I calculated the percentage revisions, Gartner’s cumulative totals, the sum of the hyperscaler capex ranges, BloombergNEF’s December 2025 value from its stated revision, and the 2Q26 Gartner data center systems figure. That figure is the 3Q26 value of $2,157.6 billion less Gartner’s published revision of $794.9 billion. Where a firm published a scenario rather than a forecast, the post says so. Charts are my own work.
My work at Dassault Systèmes includes analyst briefings with IDC, which had no role in this post.
More analysis from this blog
- Gartner’s $9.76T IT spending forecast: servers will outspend all IT services by 2030
- Gartner’s $5.95 trillion AI forecast puts the chatbot era on a 2027 deadline
- Roundup of agentic AI forecasts and market estimates, 2026
This post is my personal analysis and does not represent my employer.
Sources
Forecast and research firms
- Dell’Oro Group, “AI Buildout Maintains Momentum as Data Center Capex Surpasses $3 Trillion by 2030”, 18 August 2026.
- Dell’Oro Group, “AI Boom Drives Data Center Capex to $1.7 Trillion by 2030”, 11 February 2026.
- Gartner, Gartner Market Databook, 3Q26, G00862058, 25 September 2026, including the accompanying dataset.
- Gartner press release, Gartner Forecasts Worldwide AI Spending to Grow 49.5% in 2026, 16 September 2026. See also CIO coverage, 23 September 2026.
- IDC, AI Infrastructure Spending Holds Near $90 Billion in Q1 2026, 21 July 2026.
- IDC, AI Infrastructure Spending Caps Historic Year at $90 Billion in Q4 2025, 16 April 2026.
- IDC press release, Artificial Intelligence Infrastructure Spending to Reach $758Bn USD Mark by 2029, 28 October 2025.
- Goldman Sachs, Why AI Companies May Invest More than $500 Billion in 2026, 18 December 2025.
- Goldman Sachs, Global AI Investment Is Forecast to Exceed $1 Trillion in 2026, 7 August 2026.
- Goldman Sachs Global Institute, Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out, 1 May 2026.
- Yahoo Finance, Goldman Sachs Says a $7.6 Trillion AI Spending Boom Is Coming, 26 September 2026.
- McKinsey & Company, The cost of compute: A $7 trillion race to scale data centers, 28 April 2025.
- Bain & Company, $2 trillion in new revenue needed to fund AI scaling trend, 23 September 2025.
- Bain & Company, Global AI market could hit $6 trillion annually by 2031, 7th Global Technology Report, 29 September 2026.
- TechTimes, Goldman Sachs Quantifies AI’s $230B Annual Gap, on the Goldman Sachs note of 25 September 2026, 28 September 2026.
- Citi, Citi’s 2026 Global TMT Conference Preview, 1 September 2026.
- Reuters, Citigroup raises AI capex and revenue forecasts, 10 March 2026.
- Morgan Stanley, AI Market Trends 2026, 9 March 2026, and Thoughts on the Market, 23 July 2026.
- JLL, Global data center sector to nearly double to 200GW, 6 January 2026.
- IEA, Key Questions on Energy and AI, executive summary, July 2026.
- BloombergNEF, Six things to know about BNEF’s new US data center capacity outlook, 29 July 2026.
- Forrester, “Forrester Global Tech Forecast, 2025 To 2030”, 2 February 2026.
Company results and coverage
- NVIDIA, Financial results for second quarter fiscal 2027, 26 August 2026.
- Oracle, Q1 fiscal 2027 results, 10 September 2026.
- Digital Applied, AI capex scorecard, earnings week July 2026, 31 July 2026, and Yahoo Finance on Alphabet capex guidance, 23 July 2026.
