AI Data Center Forecast 2030
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: July 22, 2026 | License: CC BY 4.0
How Large Will AI Data Centers Be by 2030?
By 2030, global data center capacity will nearly triple to approximately 219 gigawatts — up from 82 GW in 2025 — with artificial intelligence workloads driving 71% of that growth, according to McKinsey’s April 2025 analysis. AI-focused facilities alone will require $5.2 trillion in capital investment through the end of the decade under a base scenario, while the electricity those facilities consume will roughly double from 485 TWh in 2025 to 950 TWh by 2030, per the International Energy Agency’s April 2026 Key Questions on Energy and AI report. The Axis Intelligence ADCFI™ scores the credibility of this 2030 consensus at 60.5/100 — squarely in “material uncertainty present” territory, reflecting a wide bull/bear scenario spread and significant near-term physical bottlenecks.
Quick Answer
Global AI data center capacity is forecast to reach 156 GW by 2030, up from 44 GW in 2025 — a 3.5x increase in five years — with total data center capacity (including non-AI workloads) rising to 219 GW. Meeting that demand requires $5.2 trillion in AI-specific capital expenditure, according to McKinsey. The primary constraint is physical delivery: grid connection wait times averaging four years are more binding than capital availability.
Key Findings
- According to McKinsey’s April 2025 analysis, AI-related data center capacity demand will reach 156 GW by 2030, requiring $5.2 trillion in capital expenditure under a base scenario — with a bull case reaching $7.9 trillion and a constrained scenario at $3.7 trillion, implying a $4.2 trillion spread in possible outcomes.
- Axis Intelligence Research calculates that the implied annual AI data center capex run-rate for 2027–2030 must average approximately $1.02 trillion per year to meet McKinsey’s base scenario, versus the $0.73 trillion committed by hyperscalers alone in 2026 — a shortfall structural gap that sovereign AI programs and private capital must close.
- The IEA’s April 2026 Key Questions on Energy and AI projects global data center electricity consumption will roughly double from 485 TWh in 2025 to 950 TWh by 2030, with AI-focused data centers growing even faster — tripling over the same period and reaching approximately 3% of global electricity demand.
- AI inference workloads will surpass AI training as the dominant compute requirement by 2027, according to McKinsey’s workload model — growing at a 35% CAGR to reach 93.3 GW of demand by 2030, versus 62.2 GW for training, a shift that changes the economics of where and how data centers are built.
- The Axis Intelligence AI Data Center Forecast Credibility Index™ (ADCFI™) scores the 2030 forecast consensus at 60.5/100 — reflecting 77.9/100 source agreement on the direction of growth, but only 49.2/100 confidence in the range of outcomes, as the bull/bear scenario spread spans 127 GW between McKinsey’s constrained and accelerated cases.
AI Data Center Forecast 2030: The Three-Scenario Framework
The most important framing for understanding any 2030 AI data center projection is that no credible forecaster publishes a single-number prediction. McKinsey, the IEA, and JLL all structure their analyses around scenarios, and the gap between those scenarios is itself the defining data point for infrastructure strategy.
McKinsey’s April 2025 report, The Cost of Compute: A $7 Trillion Race to Scale Data Centers, published in the McKinsey Quarterly, lays out the clearest scenario architecture:
| Scenario | Incremental AI GW (2025–2030) | Total AI GW by 2030 | Capex to 2030 (AI-focused) |
|---|---|---|---|
| Constrained | 78 GW | ~100 GW | $3.7 trillion |
| Base (central) | 125 GW | ~156 GW | $5.2 trillion |
| Accelerated | 205 GW | ~247 GW | $7.9 trillion |
Source: McKinsey & Company, “The Cost of Compute: A $7 Trillion Race to Scale Data Centers,” April 28, 2025, mckinsey.com
The spread between the constrained and accelerated scenarios — $4.2 trillion and 147 GW of capacity — is not a measurement of ignorance; it is a precise characterization of two genuinely plausible futures that depend on distinct variables. The constrained scenario materializes if grid connection timelines lengthen, high-bandwidth memory shortages persist beyond the IEA’s current 2027 estimate, or enterprise AI adoption stalls before inference at scale becomes the dominant driver. The accelerated scenario materializes if energy-intensive use cases — video generation, multi-agent orchestration, real-time scientific simulation — normalize faster than base-case AI adoption curves suggest.
Sarah Mitchell: The three-scenario architecture is the honest framework. Picking the base-case number and presenting it as “the 2030 forecast” is how market research gets cited in board decks and then ignored when reality diverges. The bull/bear spread here — $4.2 trillion in capital, 147 GW of capacity — reflects that nobody at McKinsey, the IEA, or JLL has more visibility into enterprise inference adoption than the hyperscalers themselves, and the hyperscalers are spending before they have that visibility. Which is itself the story.
How Big Is the AI Data Center Market by 2030? A Definition Map
Market-size figures for AI data centers in 2030 range from $427 billion to over $2 trillion, depending entirely on what is being measured. Axis Intelligence Research maps the three dominant measurement frameworks:
Framework 1: AI-Specific Facilities Revenue (Narrow Definition)
Grand View Research’s June 2026 AI Data Center Market report values the dedicated AI data center market at $147.3 billion in 2025 and projects it at approximately $427 billion by 2030 (implied at their 23.9% CAGR), reaching $810.6 billion by 2033. This definition covers only purpose-built, AI-optimized facilities — not traditional colocation serving partial AI workloads.
The Grand View Research AI Data Center Market report identifies the hardware segment as the largest component at 53.7% of revenue in 2025, with services as the fastest-growing component and cloud-based deployment as the leading deployment type.
Framework 2: Total AI Infrastructure Spend
McKinsey’s $5.2 trillion base figure through 2030 covers the full capital expenditure across builders (15%), energizers — utilities, cooling, power infrastructure — (25%), and technology developers and designers — GPUs, memory, servers, networking — (60%). This is not annual market revenue; it is cumulative capex over five years. The tech-hardware layer alone is a $3.1 trillion opportunity over the forecast period.
Framework 3: Total Data Center Systems Spending
JLL’s 2026 Global Data Center Market Outlook projects the sector will add ~100 GW of new capacity between 2026 and 2030, generating $1.2 trillion in real estate asset value alone, with the total infrastructure investment supercycle requiring up to $3 trillion by 2030.
The Axis Intelligence reading: a journalist citing “the AI data center market will reach X by 2030” needs to specify which framework before the number is meaningful. For infrastructure investment stories, McKinsey’s $5.2 trillion capex figure is the correct anchor. For revenue market-size comparisons, the $427B (narrow AI-specific) or the broader $900B (total data center including AI-serving infrastructure) are the appropriate figures.
What the 2030 Workload Shift Means for Infrastructure
The most consequential structural insight in McKinsey’s 2030 forecast is not the aggregate capacity number — it is the shift in the type of capacity required.
In 2025, AI training dominated AI data center demand, representing 23.1 GW versus 20.9 GW for inference. By 2030, that ratio inverts sharply: inference reaches 93.3 GW — growing at a 35% CAGR — while training reaches 62.2 GW. Inference will account for 42.6% of all data center demand globally by 2030, up from 25.4% in 2025, per McKinsey’s workload analysis.
| Workload Type | 2025 (GW) | 2030 (GW) | CAGR | 2030 Share |
|---|---|---|---|---|
| AI Inference | 20.9 | 93.3 | 35% | 42.6% |
| AI Training | 23.1 | 62.2 | 22% | 28.4% |
| Non-AI | 38.3 | 63.5 | 11% | 29.0% |
| Total | 82.3 | 219.0 | 22% | 100% |
Source: McKinsey & Company, “The Cost of Compute: A $7 Trillion Race to Scale Data Centers,” April 28, 2025; McKinsey Future of AI Workloads analysis
That workload shift is not an academic distinction. Training and inference favor different infrastructure:
Training clusters are dense, power-hungry (50–120+ kW per rack), geographically flexible because latency tolerance is high, and are built in large hyperscale campuses. They represent a massive upfront capital investment that can sit in rural markets with cheap power. Inference deployments are latency-sensitive, requiring metro or near-metro proximity to end users, lower rack densities (though rising fast), and modular scale. JLL’s 2026 Outlook projects that a significant shift in data center construction toward inference-optimized metro facilities will begin in 2027, changing the geographic distribution of new capacity dramatically.
The practical consequence: the 2030 AI data center market is not one market. It is two markets — hyperscale training campuses and distributed inference networks — with different economics, different geographies, and different financing structures. Infrastructure strategies built for training campuses will not automatically serve inference demand.
AI Data Center Forecast by Region: Who Wins the Race to 2030?
North America
North America currently holds approximately 40% of global AI data center capacity and will remain the largest single market through 2030, though its share is gradually declining as APAC and Middle Eastern markets accelerate. JLL’s 2026 Outlook projects North America’s total capacity additions will be primarily in “frontier markets” — secondary cities with available power and land — rather than the traditional Northern Virginia/Dallas/Chicago/Silicon Valley clusters where grid capacity is now exhausted.
The U.S.-specific power trajectory is the most urgent constraint in the regional picture. Goldman Sachs Research projects U.S. data center power demand will reach 41 GW in 2026, rising to 66 GW by 2027 and approximately 90+ GW by 2030. That trajectory assumes significant new power generation comes online — which requires permitting reforms that have not yet materialized at scale.
| Market | Current Capacity | 2030 Projection | CAGR | Driver |
|---|---|---|---|---|
| North America (total) | ~50 GW (2025) | ~85–95 GW | ~12% | AI inference buildout, frontier markets |
| Northern Virginia | 5.6 GW | Constrained | Grid-limited | Power available < demand |
| Texas (Austin/Abilene) | Growing | Stargate anchor | High | Onsite gas + renewable PPAs |
Sources: JLL 2026 Global Data Center Market Outlook; Goldman Sachs Research, May 2026; Axis Intelligence Research analysis
Asia-Pacific: The Fastest-Growing Regional Market
Asia-Pacific currently holds approximately 32 GW of total capacity, set to reach 57 GW by 2030 — a 12% CAGR that exceeds every other region’s base-case trajectory, per JLL’s 2026 Outlook. That number is being driven by distinct forces in three markets:
India is where the most structurally interesting growth is forming. Alphabet committed $15 billion to India between 2026 and 2030, including gigawatt-scale data center campuses and subsea cable infrastructure developed with Bharti Airtel (announced October 2025). The Indian government’s National AI Mission, combined with power availability in Hyderabad, Mumbai, and Chennai, is building a credible hyperscale market that didn’t exist at meaningful scale before 2024. The $15 billion India commitment from a single hyperscaler in a single announcement is the clearest single signal of where the next cycle is forming.
Japan is benefiting from its proximity to Taiwanese semiconductor supply chains and its relatively stable regulatory environment. The Japanese government’s AI infrastructure investment plan — part of the broader “GX” industrial transition — includes dedicated AI data center capacity.
Singapore is grid-constrained (the Singaporean government’s data center moratorium was partially lifted in 2023) but remains the ASEAN hub for latency-sensitive inference deployment.
| Region | 2025 GW | 2030 GW | CAGR | Leading Sub-market |
|---|---|---|---|---|
| APAC Total | 32 | 57 | 12% | India, Japan |
| EMEA Total | ~22 GW est. | ~35 GW est. | ~10% | Germany, UK, UAE |
| Americas Total | ~50 | ~87 | ~12% | Texas, Phoenix, Chicago |
Sources: JLL 2026 Global Data Center Market Outlook (APAC/EMEA/Americas capacity), Axis Intelligence Research regional analysis
Europe and the Sovereignty Factor
EMEA’s 10% CAGR forecast is driven not just by hyperscaler demand but by regulatory pressure. The EU AI Act’s compute-threshold requirements for “general-purpose AI” models effectively mandate that large AI operators maintain or prove access to compliant computing infrastructure within the EU. That requirement is creating a category of “sovereignty-motivated” data center investment that doesn’t exist in the U.S. market.
Germany, the UK, and France lead European AI data center investment, but the Middle East is the fastest-growing sub-region within EMEA. The UAE’s Stargate partnership (1 GW in Abu Dhabi, OpenAI/SoftBank/MGX, expected 2026) and Saudi Arabia’s comparable sovereign AI infrastructure commitments represent the most significant shift in global data center geography since Ireland became the European hub for U.S. hyperscalers in the early 2010s.
The $1.02 Trillion Annual Run-Rate Problem
The number that should concern every infrastructure planner reading the 2030 AI data center forecasts is not the total capex figure — it is what that figure implies about the annual spending required after 2026.
According to Axis Intelligence Research’s cross-source calculation:
- McKinsey’s base scenario requires $5.2 trillion in AI data center capex through 2030.
- Hyperscaler capex already committed: ~$725 billion (2026) + ~$410 billion (2025) = $1.135 trillion for two years.
- Remaining capex needed (2027–2030): approximately $5.2T − $1.135T = $4.065 trillion.
- Spread across 4 years: $1.02 trillion per year required, on average, from 2027 through 2030.
That $1.02 trillion annual run-rate is 40% higher than the hyperscaler-only 2026 figure of $725 billion. It implies that hyperscalers cannot fund the transition alone — and the IEA’s April 2026 report explicitly notes that “data centre investments have grown too large to be funded from company balance sheets alone, and large amounts of funding from capital markets will be critical.” Infrastructure investment funds, sovereign wealth funds, and project finance structures are not a supplement to the AI data center buildout; by 2027 they are structurally necessary for the base scenario to materialize.
Formula: Annual run-rate = (McKinsey base AI capex [McKinsey, April 2025] − documented hyperscaler capex 2025 [Q1 2026 earnings] − documented hyperscaler capex 2026 [Q1 2026 guidance]) ÷ 4 remaining years. Inputs are independently published by named primary sources. The resulting figure is an Axis Intelligence Research cross-source derivation not published by any single source. Method note in CSV.
The Axis Intelligence ADCFI™ — AI Data Center Forecast Credibility Index
The ADCFI™ is an original Axis Intelligence Research metric designed to answer one question that no existing research instrument addresses: how confident should analysts and decision-makers be in the 2030 AI data center forecast consensus, given the range of scenarios, physical constraints, and capital commitments observed across primary sources?
The index does not forecast capacity. It measures the quality of existing forecasts.
ADCFI™ Methodology (Baseline Reading: July 2026)
The ADCFI™ combines four equally-sourced dimensions, each normalized 0–100:
| Dimension | Weight | Q3 2026 Score | Basis |
|---|---|---|---|
| Source Consensus — How tightly major 2030 AI GW estimates cluster (CV-based) | 30% | 77.9 | Mean 127 GW across McKinsey/IEA-implied/JLL estimates; CV = 0.22 |
| Bull/Bear Spread — Inverse of scenario range relative to base case | 25% | 49.2 | McKinsey constrained (78 GW) to accelerated (205 GW) = 127 GW spread |
| Investment Commitment — Committed pipeline as share of required capex | 25% | 63.5 | ~$1.65T committed vs $5.2T needed = 31.7% |
| Structural Constraint Risk — Inverted delivery bottleneck score | 20% | 45.0 | IEA-flagged: HBM shortage, grid delays, permitting; high-constraint environment |
ADCFI™ Composite Score: 60.5/100 — “Material uncertainty present”
Full composite: (0.30 × 77.9) + (0.25 × 49.2) + (0.25 × 63.5) + (0.20 × 45.0) = 60.5
ADCFI™ Score Interpretation
A score above 70 indicates that major forecasters agree on the direction and approximate magnitude of growth, committed capital is sufficient to validate demand, and physical delivery constraints are manageable. A score between 50 and 70 indicates that the direction of growth is clear but the magnitude is genuinely uncertain, and decision-makers should plan for scenario ranges rather than point estimates. A score below 50 indicates deep disagreement or insurmountable near-term constraints.
At 60.5, the 2030 AI data center outlook sits firmly in the middle category. The direction of growth is not in question — every primary source projects significant expansion. But the question of how large, in what form, and in which geographies remains open in ways that should make infrastructure planners cautious about single-number commitments.
The ADCFI™ is an Axis Intelligence Research baseline metric, first published July 2026. It will be updated quarterly as IEA data, McKinsey scenario updates, and hyperscaler capital commitment guidance become available. A score cannot describe itself relative to prior readings at its inaugural publication; by Q1 2027 directional comparisons will be possible.
What the Inference Shift Means for the 2030 Forecast
The workload transition from training to inference — which McKinsey projects will be the decisive structural shift by 2027 — is not an optimization story. It is a market redefinition.
Training workloads are concentrated in a small number of very large facilities, run by a small number of very large organizations. The 2024–2026 capital surge reflects this: hyperscalers building 100–500 MW campuses to train foundation models. That market is functionally an oligopoly.
Inference is different. Inference demand is distributed — every enterprise that deploys an AI application becomes an inference customer. The 35% CAGR in inference demand through 2030 is not coming from OpenAI and Google building bigger clusters; it is coming from millions of enterprise applications calling AI APIs at scale. That demand disperses geographically (latency requirements force proximity), disperses organizationally (enterprise procurement vs. hyperscaler capex), and disperses technically (different hardware profiles, lower per-rack power density, higher networking requirements).
For infrastructure investors, the inference transition creates the most significant new market opportunity of the 2026–2030 cycle. The colocation segment’s forecast 32.6% CAGR through the decade — consistently the fastest-growing data center type across MarketsandMarkets, Grand View Research, and JLL — reflects precisely this: enterprises that want inference capacity without building hyperscale campuses will rent it from colocation providers in metro markets. The colocation market by 2030 is the inference market.
Five Constraints That Could Break the Base Scenario
The McKinsey base scenario of 156 GW by 2030 is not a certainty; it is the most probable of three scenarios under stated assumptions. Each of the following constraints, if it materializes more severely than modeled, shifts the outcome toward the constrained case:
1. High-Bandwidth Memory Supply (most acute near-term risk). The IEA’s April 2026 report flags a shortage of high-bandwidth memory — integral to AI chip production — that has developed over the past six months and is expected to persist through at least the end of 2027. HBM is the memory architecture used in NVIDIA’s Blackwell GPUs; constrained supply directly limits the GPU rack density achievable in new facilities, forcing operators to build more physical space for equivalent compute.
2. Grid Connection Timelines (most geographically variable). The U.S. average grid connection wait is now four years per JLL. In Northern Virginia, the largest data center market globally, effective available capacity is below 1% with connection queues extending to a decade in some sub-markets. Ireland briefly imposed a connection moratorium before lifting it in December 2025 with a requirement for onsite generation. The IEA estimates 15–27 GW of onsite natural gas generation may need to power U.S. data centers by 2030 — a significant departure from original renewable-first infrastructure plans.
3. Capital Markets Sentiment. The IEA explicitly notes that data center investment is now too large for company balance sheets alone. The pace of the buildout after 2026 will be sensitive to credit conditions, real estate capital markets, and the returns generated by early AI cloud services. If inference revenue growth disappoints versus Google Cloud’s $460 billion contracted backlog, institutional investors will recalibrate.
4. Enterprise AI Adoption Rate. The base scenario assumes enterprises convert AI enthusiasm into sustained inference workload growth. McKinsey models ~70% of 2030 data center demand coming from AI. The scenario where that doesn’t happen — not from a technology failure but from a slower-than-expected enterprise integration cycle — is the single largest risk to the base case.
5. Geopolitical Technology Controls. Export restrictions on advanced semiconductors and critical minerals already affect the supply chains for transformers, power electronics, and batteries that data centers need. If restrictions extend to more components or escalate in scope, the constrained scenario becomes the base case.
How Much Electricity Will AI Data Centers Use by 2030?
The IEA’s April 2026 Key Questions on Energy and AI provides the authoritative global projection:
| Year | Total Data Center (TWh) | AI-Focused Growth | Share of Global Electricity | Source |
|---|---|---|---|---|
| 2025 | 485 TWh | +50% YoY | ~1.8% | IEA, April 2026 |
| 2026 (est.) | ~530–560 TWh | Growing | ~2.0% | IEA estimates |
| 2030 | 950 TWh | 3x vs 2025 | ~3.0% | IEA base case |
| 2035 | 1,200 TWh | — | ~3.5% | IEA extended projection |
Source: International Energy Agency, Key Questions on Energy and AI, April 2026, iea.org
The IEA’s central projection of 950 TWh by 2030 represents approximately the annual electricity consumption of Germany and Japan combined. The IEA notes that the near-term bottlenecks — chip shortages, grid delays — are “reducing the likelihood of more aggressive near-term scenarios,” but that the mid- to longer-term outlook sees “a possible upside” after 2030 as infrastructure investments relieve current constraints.
An individual server rack in an advanced AI data center is “only the size of a large refrigerator,” per the IEA’s April 2026 analysis, “but by 2027 it could have peak power demand equivalent to that of 65 households.” Between 2020 and 2025, AI server power density increased 11-fold; a further fourfold increase is projected by 2027, driven by NVIDIA’s Blackwell Ultra and next-generation architectures.
Methodology
Data Collection
This dataset was compiled by Axis Intelligence Research from primary sources published between April 2025 and April 2026. Primary sources accessed and fetched include:
- McKinsey & Company, “The Cost of Compute: A $7 Trillion Race to Scale Data Centers,” McKinsey Quarterly, April 28, 2025 — [fetched directly from mckinsey.com]
- International Energy Agency, Key Questions on Energy and AI, April 2026 — [fetched directly from iea.org]
- JLL, 2026 Global Data Center Market Outlook, May 2026 — [confirmed live at jll.com]
- Grand View Research, AI Data Center Market report, 2026 — [confirmed via press release at grandviewresearch.com]
- McKinsey & Company, “The Future of AI Workloads” workload chart — [fetched from mckinsey.com]
No statistics in this article originate from secondary aggregators, press releases from non-originating parties, or AI model memory. All figures are traced to the named issuing organization.
The ADCFI™ Formula
ADCFI™ = (0.30 × Source Consensus Score) + (0.25 × Bull/Bear Spread Score) + (0.25 × Investment Commitment Score) + (0.20 × Structural Constraint Risk Score)
Source Consensus Score: 1 − (standard deviation of major 2030 AI GW estimates / mean), scaled 0–100. Inputs: McKinsey base 156 GW; JLL-implied AI share of 200 GW (50%) = 100 GW; IEA-implied at ~6 TWh/GW conversion of 950 TWh = ~125 GW. Mean = 127 GW, stdev = 28.1, CV = 0.221. Score = (1 − 0.221) × 100 = 77.9.
Bull/Bear Spread Score: max(0, (1 − spread_ratio / 2.0)) × 100. Spread ratio = (205 GW − 78 GW) / 125 GW = 1.02. Score = (1 − 1.02 / 2.0) × 100 = 49.2.
Investment Commitment Score: min(commitment_ratio / 0.50, 1.0) × 100. Committed: $725B (2026 hyperscaler guidance) + $625B (Stargate 2026–2030 share) + $300B (estimated sovereign/institutional pipeline) = $1.65T. Ratio vs. $5.2T needed = 31.7%. Score = min(0.317/0.5, 1.0) × 100 = 63.5.
Structural Constraint Risk Score (inverted): 45.0/100 — reflects IEA-flagged HBM shortage through 2027, 4-year average U.S. grid connection timelines, and IEA’s explicit statement that bottlenecks are reducing the likelihood of aggressive near-term scenarios.
Composite: (0.30 × 77.9) + (0.25 × 49.2) + (0.25 × 63.5) + (0.20 × 45.0) = 23.37 + 12.30 + 15.87 + 9.00 = 60.5. Python-verified arithmetic.
Limitations
- McKinsey capacity forecasts use a proprietary demand model; component weights and adjustment factors are not publicly disclosed.
- The IEA’s 950 TWh 2030 figure is a base case; IEA explicitly notes possible upside after 2030.
- Grand View Research’s revenue figures use a “dedicated AI-optimized facilities” definition; comparisons to broader market-size figures require definition alignment.
- The ADCFI™ Structural Constraint score (45.0) is qualitative, derived from IEA and JLL narrative assessments of physical delivery constraints, not a quantitative bottleneck index. It will be replaced with a quantitative sub-index in Q1 2027 when updated grid connection data from JLL and LBNL are available.
- Axis Intelligence Research’s cross-source capex run-rate figure ($1.02T/year, 2027–2030) is a derivation using rounded annual figures; it will be refined as actual 2025 hyperscaler full-year capex figures are reported.
About This Dataset
Dataset title: AI Data Center Forecast 2030 — Capacity, Capex, Workload, and Credibility Index
Issued by: Axis Intelligence Research
Coverage: Global; with North America, APAC, EMEA regional breakdowns
Base period: 2025 (latest available), with projections to 2030 and 2035
Last updated: July 22, 2026
License: Creative Commons Attribution 4.0 International (CC BY 4.0). Share and adapt freely with attribution.
Download: ai-data-center-forecast-2030.csv
Cite This Research
APA: Axis Intelligence Research. (2026). AI Data Center Forecast 2030: Capacity, Capex, Workload Shift, and the ADCFI™ Credibility Index. Axis Intelligence. https://axis-intelligence.com/ai-data-center-forecast-2030/
MLA: Axis Intelligence Research. “AI Data Center Forecast 2030: Capacity, Capex, Workload Shift, and the ADCFI™ Credibility Index.” Axis Intelligence, 22 July 2026, axis-intelligence.com/ai-data-center-forecast-2030/.
Chicago: Axis Intelligence Research. “AI Data Center Forecast 2030: Capacity, Capex, Workload Shift, and the ADCFI™ Credibility Index.” Axis Intelligence, July 22, 2026. https://axis-intelligence.com/ai-data-center-forecast-2030/.
FAQ
How much will AI data centers grow by 2030?
Global AI-focused data center capacity will grow from approximately 44 GW in 2025 to 156 GW by 2030 under McKinsey’s base scenario — a 3.5x increase in five years. Total data center capacity (including non-AI workloads) is projected to reach 219 GW, up from 82.3 GW in 2025. That expansion requires $5.2 trillion in AI-specific capital investment.
What is the AI data center market size in 2030?
It depends on the definition. Grand View Research’s 23.9% CAGR implies approximately $427 billion in dedicated AI data center facility revenue by 2030. McKinsey’s $5.2 trillion figure covers cumulative five-year capital expenditure across builders, power infrastructure, and technology hardware. JLL projects $1.2 trillion in real estate asset value creation alone by 2030. These figures are not comparable; each measures a different layer of the same market.
Will AI inference or training dominate data centers by 2030?
Inference. McKinsey’s workload model projects AI inference will reach 93.3 GW of demand by 2030 — growing at a 35% CAGR — versus 62.2 GW for AI training. Inference will overtake training as the dominant AI workload by 2027 and represent 42.6% of all data center demand by 2030. This shift changes the economics and geography of data center construction toward metro proximity and modular colocation.
What constraints could prevent the 2030 AI data center forecast from being met?
Five primary constraints: high-bandwidth memory shortages (IEA projects they persist through at least 2027), grid connection delays averaging 4 years in the U.S. and longer in some European markets, capital markets sentiment if cloud revenue disappoints, enterprise AI adoption rate falling short of the 70% AI-share projection, and geopolitical technology export controls on semiconductors and critical minerals. The Axis Intelligence ADCFI™ composite of 60.5/100 reflects these constraints in its structural constraint sub-score of 45.0/100.
How much electricity will AI data centers use by 2030?
The IEA’s April 2026 Key Questions on Energy and AI projects total global data center electricity consumption will roughly double from 485 TWh in 2025 to 950 TWh by 2030, accounting for approximately 3% of global electricity demand. AI-focused data centers will triple their consumption over the same period. The IEA also notes that by 2027, an individual AI server rack could have peak power demand equivalent to 65 homes.
Which region will have the most AI data centers in 2030?
North America will remain the largest single market, but its share is declining. Asia-Pacific is the fastest-growing region, expanding from 32 GW to 57 GW by 2030 at a 12% CAGR per JLL’s 2026 Outlook. India is the single most structurally significant new AI data center market, backed by Alphabet’s $15 billion 2026–2030 commitment and government National AI Mission investment. The Middle East — particularly UAE and Saudi Arabia — is the fastest-growing EMEA sub-region.
What is the Axis Intelligence ADCFI™?
The AI Data Center Forecast Credibility Index™ (ADCFI™) is an original Axis Intelligence Research composite metric measuring how confident analysts should be in the 2030 AI data center forecast consensus. It combines source agreement, scenario spread, committed capital as a share of required capex, and an inverted constraint score. The July 2026 baseline reading of 60.5/100 indicates “material uncertainty present” — the direction of growth is clear, but the magnitude and pace are genuinely uncertain.
Is a $7 trillion investment in AI data centers financially sustainable?
This is the right question, and nobody outside the hyperscalers has sufficient data to answer it definitively. McKinsey’s $7.9 trillion accelerated scenario assumes demand materializes faster than efficiency gains offset volume growth. The financial justification depends on inference revenue — Google Cloud’s $460 billion contracted backlog (Q1 2026) suggests genuine, contracted enterprise demand. The risk is adoption pace: if enterprise inference workloads grow more slowly than the backlog implies, stranded-asset risk materializes across a decade of capital commitments.
What is the annual capex run-rate needed after 2026 to meet McKinsey’s base forecast?
According to Axis Intelligence Research’s cross-source calculation, approximately $1.02 trillion per year must be invested in AI data centers from 2027 through 2030 to meet McKinsey’s base scenario of $5.2 trillion in AI capex through 2030 — after accounting for the ~$1.135 trillion already committed by hyperscalers in 2025 and 2026. This is 40% above the 2026 hyperscaler-only run-rate of $725 billion, meaning sovereign capital, infrastructure funds, and project finance must cover the gap.
Related Axis Intelligence Research
- AI Data Center Market Size 2026 — DCII™, $725B supercycle, hyperscaler capex breakdown
- AI Data Center Statistics 2026 — AIPI™, electricity, capacity, and the power crisis
- AI Data Center Energy Consumption Statistics — 485 TWh, IEA projections, regional load
- AI Data Center Construction Cost Statistics — $11.3M/MW benchmark, JLL 2026 data
- AI Investment Statistics 2026 — $581B deployed, hyperscaler spend detail
- Semiconductor Statistics 2026 — AI Silicon Concentration Index, HBM supply chain