Quick answer: The IEA's April 2026 update projects global data center electricity consumption roughly doubling from 485 TWh in 2025 to about 950 TWh by 2030, with AI-focused facilities tripling over the same period. In the U.S., Goldman Sachs now flags a structural power shortfall of 9.3 GW in 2026, widening to 45 GW by 2028 — a gap large enough that every major hyperscaler has now signed at least one nuclear or SMR power deal to secure supply.
Key Takeaways
- Global data center electricity consumption reached roughly 485 TWh in 2025 and is projected to approach 950 TWh by 2030, per the IEA's April 2026 report.
- AI-optimized, accelerator-heavy facilities are growing electricity demand roughly three times faster than conventional server infrastructure.
- U.S. data centers could account for 9%–17% of national electricity consumption by 2030, up from roughly 4%–5% today, with some Virginia grid zones already exceeding 39%.
- Power availability, not land or permitting, is now the leading cause of construction delay across major markets.
- Hyperscalers have collectively committed to nearly 10 GW of nuclear and small modular reactor capacity since 2023 to bypass grid interconnection queues.
- Energy-rich secondary markets are emerging as the next generation of data center hubs as primary markets hit grid capacity limits.
The rapid expansion of artificial intelligence is reshaping the global data center industry less through software breakthroughs than through raw electricity demand. Large-scale model training and inference are pushing hyperscale facilities into a new era of power intensity, where the binding constraint on growth is no longer server availability — it is grid capacity.
Traditional cloud computing never demanded this much energy density. AI infrastructure routinely doubles or triples power consumption per rack, and that shift is now visible in transmission planning, utility rate cases, and corporate energy strategy across every major market.
Across multiple industry forecasts, the consensus has firmed up: global data center electricity consumption is on track to roughly double by 2030, with the United States absorbing a disproportionate share of that growth given its concentration of hyperscale operators.
In this briefing
Global Power Demand Dashboard — 2026
950 TWh
Global data center demand by 2030
IEA, April 202645 GW
Projected U.S. power shortfall by 2028
Goldman Sachs Research~10 GW
Nuclear & SMR capacity committed by hyperscalers
Industry deal trackers, 20261.6%
North American colocation vacancy rate
CBRE, 20251. Why AI Consumes So Much More Power Than Traditional Computing
Investors hear about AI-driven electricity demand constantly, but many still underestimate the scale of the gap between conventional cloud computing and modern AI workloads.
AI training and inference run inside specialized facilities packed with servers, storage, networking gear, cooling infrastructure, and backup power systems. According to the IEA, servers and accelerators account for roughly 60% of electricity consumption inside a modern data center, with cooling, networking, and storage making up the rest.
What sets AI apart is its dependence on accelerator chips — GPUs and custom silicon — that draw far more power per rack than general-purpose CPUs ever did. The IEA's modeling finds that electricity consumption from these accelerated servers is growing at roughly 30% a year, versus about 9% annual growth for conventional servers, making accelerated computing the dominant driver of future data center power demand.
➡️ Read Also: Investing in Data Center Commercial Real Estate in 2026: AI, Power Demand, and the Next CRE Boom
AI Servers vs. Traditional Servers: Annual Growth Rate
Figure 1. Electricity demand from AI-focused accelerated servers is growing more than three times faster than traditional servers. Source: IEA, Energy and AI, 2026.
2. Hyperscale AI Growth and the Rise of Power-Intensive Infrastructure
The primary driver behind this shift is the continued expansion of hyperscale data centers operated by Microsoft, Google, Amazon, and Meta. These facilities increasingly support GPU- and accelerator-heavy workloads built specifically for AI training and inference, rather than general cloud services.
In many current builds, power demand per rack has more than doubled compared with facilities designed just three or four years ago, with next-generation AI clusters reaching power densities that would have been unworkable under older cooling and electrical designs.
As a result, real estate availability has stopped being the primary bottleneck for new data center supply. Electricity access and grid interconnection capacity have taken its place.
3. CBRE Perspective: Power Becomes the Primary Constraint
CBRE's research shows the data center market operating under historically tight supply conditions, with growth increasingly capped by energy availability rather than demand.
Key trends CBRE has tracked:
- North American data center absorption reached approximately 2,497.6 MW in 2025, a 38% year-over-year increase.
- Primary market supply reached 8,155 MW in H1 2025, yet vacancy still fell to a record low of 1.6%.
- Around 74% of capacity currently under construction is already pre-leased, mostly by hyperscalers and AI operators.
- Power shortages are now the leading cause of construction delays across major markets.
CBRE also points to a structural shift in pricing: leases are increasingly quoted against power capacity (kW/MW) rather than square footage, and rents in high-demand markets have risen sharply — in some cases between 20% and 54% within short periods. In Asia Pacific specifically, CBRE forecasts a potential 15–25 GW supply shortfall by 2028, driven by energy constraints and a shortage of AI-ready infrastructure.
4. JLL Analysis: Speed-to-Power Becomes the New Location Strategy
JLL's research highlights a structural change in how developers evaluate sites. Proximity to population centers and fiber connectivity used to dominate location decisions. Today, the deciding factor is what the industry calls "speed to power" — how quickly a site can actually be energized.
Key findings from JLL's latest market coverage:
- Global data center capacity could reach roughly 200 GW by 2030.
- Nearly 100 GW of additional capacity is expected to come online between 2026 and 2030.
- AI workloads are projected to account for roughly 50% of total data center usage by the end of the decade.
- New developments are scaling rapidly, moving from 10–50 MW campuses to 100–500+ MW campuses.
- North American colocation vacancy sits at an all-time low of 2.3% (below 1% in Northern Virginia specifically), with rents rising roughly 11% annually for five consecutive years.
Less than 10% of existing U.S. data center inventory is currently considered AI-ready, particularly with respect to high-density power distribution and liquid cooling. JLL also notes that grid connection timelines in primary markets now regularly exceed four years, pushing operators toward on-site generation, battery storage, and gas-bridging arrangements just to hit their deployment schedules.
➡️ Read the related post: How Canada's Renewable Energy Advantage Is Attracting AI Data Center Investment
5. McKinsey Perspective: A Multi-Trillion-Dollar Infrastructure Shift
McKinsey frames this as one of the largest infrastructure investment cycles in modern history, driven almost entirely by digital and AI infrastructure expansion.
Its key projections:
- Global data center demand growing at roughly 22% CAGR to around 220 GW by 2030.
- Total global capital expenditure potentially reaching $6.7–7 trillion.
- U.S. data center electricity demand rising from roughly 147 TWh in 2023 to about 606 TWh by 2030.
- Data centers potentially accounting for 11%–12% of total U.S. electricity consumption by 2030.
Under this trajectory, data centers could represent 30%–40% of all new U.S. electricity demand growth this decade, with AI workloads as the dominant driver and inference becoming an increasingly large share alongside model training.
U.S. Data Center Electricity Demand Growth
Figure 2. U.S. data center electricity consumption could increase more than fourfold by 2030. Source: McKinsey & Company.
6. The Global Data Center Power Boom in Numbers
Several major research organizations are converging on the same broad trajectory for data center electricity demand, even where their precise figures differ.
Current benchmarks and forecasts:
- Global data center electricity consumption reached approximately 415 TWh in 2024 and roughly 485 TWh in 2025, a 17% year-over-year increase.
- The IEA's April 2026 update projects consumption reaching about 950 TWh by 2030.
- Longer-range estimates put demand above 1,200 TWh by 2035.
- Data centers currently account for roughly 1.5%–2% of global electricity consumption, a share expected to approach 3% by 2030.
Those percentages look modest in isolation, but data center demand is growing more than four times faster than electricity demand from most other sectors combined — which is exactly why utilities, regulators, and infrastructure investors are having to rethink long-term grid planning right now rather than later in the decade.
➡️ Read the related article: AI Data Centers Are Increasing Pressure on Water Resources Across US Cities
Global Data Center Electricity Consumption Forecast (2024–2035)
Figure 3. Global data center electricity demand is on track to nearly triple between 2024 and 2035, driven by AI workloads and accelerated computing. Source: IEA, April 2026.
7. Energy System Impacts: IEA, LBNL, and EPRI Findings
Independent energy agencies confirm the scale of the transformation underway, even where their modeling assumptions differ.
International Energy Agency (IEA)
- Global data centers consumed roughly 485 TWh in 2025, up 17% year-over-year.
- Consumption is projected to reach roughly 950 TWh by 2030 in the IEA's base case.
- AI-focused data center electricity use is expected to triple over the same period.
- Near-term bottlenecks — chip supply, grid interconnection delays, and permitting backlogs — are reducing the likelihood of the IEA's more aggressive upside scenarios.
Lawrence Berkeley National Laboratory (LBNL)
- U.S. data centers consumed roughly 176 TWh in 2023 (4.4% of national electricity use).
- Consumption could reach 325–580 TWh by 2028.
- That would equal 6.7%–12% of total U.S. electricity demand.
Electric Power Research Institute (EPRI)
- U.S. data centers currently consume roughly 4%–5% of national electricity.
- That share could rise to 9%–17% by 2030.
- Some states face extreme concentration — Virginia's data center load could reach 39%–57% of local electricity demand.
Together, these findings point to a consistent reality: global percentages remain moderate, but local and regional grid impacts are already extreme in the markets carrying the heaviest data center concentration.
8. Regional Growth: Where Electricity Demand Is Rising Fastest
The United States and China are expected to dominate global data center electricity growth for the remainder of the decade.
United States
- Data center electricity demand could increase by roughly 240 TWh, growth of about 130% by 2030.
- Per-capita data center electricity consumption could exceed 1,200 kWh annually.
China
- Electricity demand could increase by roughly 175 TWh, growth of around 170% by 2030.
Europe
- Data center consumption is expected to grow by more than 45 TWh, an increase of roughly 70%. City-level concentration is already extreme — data centers account for an estimated 33%–42% of electricity use in Amsterdam, London, and Frankfurt, and nearly 80% in Dublin.
Southeast Asia
Emerging markets such as Singapore and southern Malaysia are becoming important regional hubs on the back of strong cloud and AI demand. Electricity demand from data centers across parts of Southeast Asia is expected to more than double by 2030.
These regional patterns suggest future data center development will keep shifting toward markets with abundant power resources and supportive grid policy, rather than toward markets simply close to population centers.
Regional Data Center Electricity Demand Growth by 2030
Figure 4. China and the United States are expected to drive most global data center electricity demand growth through 2030. Source: IEA regional modeling.
Why This Matters for Investors
The AI infrastructure boom is not benefiting technology companies alone. Several adjacent sectors stand to gain directly from this buildout:
- Data center REITs
- Electric utilities
- Transmission infrastructure developers
- Renewable energy operators
- Nuclear and SMR developers
- Industrial real estate owners
- Cooling technology providers
- Battery storage companies
As AI adoption expands, capital is increasingly flowing into the physical infrastructure required to support digital workloads — not just into the software layer sitting on top of it.
9. The Hyperscaler Pivot to Nuclear: Energy Procurement Becomes Strategic
Facing multi-year interconnection queues — the U.S. queue alone now holds more than 2,600 GW of projects waiting for grid connection, with average waits around five years and high withdrawal rates — major technology companies have stopped waiting on utilities and started underwriting new generation capacity directly.
By mid-2026, every major hyperscaler had signed at least one nuclear or small modular reactor (SMR) agreement:
- Microsoft signed a $16 billion, 20-year power purchase agreement to restart Three Mile Island Unit 1 — now rebranded the Crane Clean Energy Center (835 MW) — with commercial operation targeted for the second half of 2027 after a FERC transmission waiver cleared the last major hurdle.
- Google committed to up to 500 MW across six to seven small modular reactor units with Kairos Power, alongside a separate 1,800 MW agreement with Elementl Power, with first reactors targeted around 2030.
- Amazon expanded its offtake agreement with Talen Energy to nearly 2 GW from the Susquehanna nuclear plant through 2042, while separately leading a $700 million investment round in X-energy to develop up to twelve Xe-100 SMR units, and joining a broader $50 billion initiative with Korea Hydro & Nuclear Power and Doosan Enerbility to accelerate SMR supply chains.
- Meta holds the largest combined commitment, up to 6.6 GW across TerraPower, Oklo, Vistra, and Constellation, though on a longer delivery timeline stretching into the early 2030s.
Across all announced deals, industry trackers now put combined hyperscaler nuclear and SMR commitments at roughly 10 GW — a figure that was close to zero just three years ago. Microsoft has also introduced what it calls a "community-first" infrastructure approach, pledging to absorb grid-upgrade costs directly so nearby ratepayers aren't left subsidizing hyperscale demand.
The structural shift here matters more than any single deal: hyperscalers are no longer just large energy consumers. They are becoming energy system participants, underwriting first-of-a-kind reactor designs and taking on development risk that utilities and regulators previously carried alone.
Hyperscaler Nuclear & SMR Capacity Commitments (2023–2026)
Figure 5. Meta currently leads announced hyperscaler nuclear and SMR commitments, though on a longer delivery timeline than Microsoft's Three Mile Island restart. Figures reflect publicly disclosed deals as of mid-2026 and may understate total committed capacity.
10. Understanding Where Data Center Electricity Actually Goes
AI gets most of the attention, but electricity consumption inside a data center is spread across several distinct systems.
A typical hyperscale facility allocates power roughly as follows:
| Component | Share of Electricity Use |
|---|---|
| Servers & AI Accelerators | ~60% |
| Cooling Systems | 7%–30% |
| Storage Systems | ~5% |
| Networking Equipment | Up to 5% |
| Other Infrastructure | Remaining Share |
Cooling's share is climbing fastest, because higher-density AI clusters generate far more heat than conventional server halls. That is accelerating investment in liquid cooling and next-generation thermal management, since air alone can no longer keep dense GPU racks within operating range.
Data Center Electricity Consumption Breakdown
Figure 6. Servers and AI accelerators account for the majority of electricity consumed inside hyperscale data centers. Source: IEA.
11. Structural Challenges and System Risks
Despite record investment, several constraints remain unresolved:
1. Grid Capacity Limits
Interconnection delays exceeding four years are now common in major markets, and the U.S. queue alone holds more than 2,600 GW of projects awaiting connection.
2. Energy Supply Constraints
Renewables alone cannot meet near-term demand spikes at hyperscale speed, which is part of why nuclear and gas-bridging solutions have re-entered the conversation.
3. Regional Overconcentration
Markets like Northern Virginia are facing extreme load pressure, with local grid operators openly flagging capacity constraints.
4. Cost Inflation
Rising electricity, transformer, and component costs are compressing margins even as revenue from AI services scales.
5. Sustainability Pressure
Balancing AI-driven growth against corporate carbon-reduction targets remains genuinely difficult, particularly where new capacity leans on gas-fired backup generation in the interim.
12. Alternative Scenarios: Why Forecasts Still Vary
Although most forecasts point toward rapid growth, the IEA identifies three variables that could meaningfully change the outcome:
Lift-Off Scenario
Faster AI adoption, stronger investment, and rapid deployment of new AI applications push electricity demand beyond current base-case expectations.
High-Efficiency Scenario
Advances in chip design, software optimization, and cooling technology meaningfully reduce energy consumption per unit of computation.
Headwinds Scenario
Grid bottlenecks, supply chain constraints, permitting delays, and slower-than-expected AI adoption cap demand growth well below current projections.
Even under the more conservative of these scenarios, electricity demand from AI infrastructure stays on a firm upward trajectory through the rest of the decade — the range of outcomes is about how fast, not whether.
Future Data Center Bottlenecks
Figure 7. Industry priorities have shifted from real estate availability toward electricity access and grid capacity.
13. Future Outlook: A Redefined Energy–Technology Nexus
The intersection of artificial intelligence and energy infrastructure is one of the most significant structural shifts in modern industrial history. Efficiency gains in chips, cooling, and software will offset some of this demand, but they are very unlikely to fully counterbalance the pace of AI-driven growth.
The longer-term trajectory points toward:
- Continued acceleration of data center construction, concentrated wherever power can be secured fastest
- Expansion into energy-rich secondary markets as primary hubs hit grid limits
- Greater integration of private, on-site power generation
- Rising importance of nuclear and hybrid energy systems as a baseload solution
- Increasing competition for electricity itself as a strategic, allocatable resource
Advisory: What Each Stakeholder Should Do Next
The throughline across every forecast in this briefing is the same: electricity, not capital or land, is now the resource that determines who gets to build and how fast. Here is what that means in practice for the groups most exposed to it.
For Developers & Site Selectors
- Screen sites on confirmed interconnection timelines first, not land cost or fiber proximity — a four-plus-year grid queue can sink an otherwise strong site.
- Build liquid cooling and higher power-per-rack assumptions into base-case designs rather than retrofitting later; cooling's share of total electricity use is only rising.
- Treat nuclear-adjacent or power-certain sites as a distinct, premium category — current market data shows a 15%–25% lease-rate premium for power-certain locations.
For Investors & Capital Allocators
- Weight power availability and interconnection status above headline capacity numbers when comparing competing projects — announced MW and energized MW are not the same asset.
- Track adjacent beneficiaries of this cycle directly: utilities, transmission developers, nuclear/SMR developers, and cooling technology providers all carry exposure without the concentration risk of pure GPU plays.
- Model nuclear-backed projects against delay scenarios explicitly — SMR delivery timelines commonly run three to five years past initial targets, and a site that only works if nuclear arrives on schedule is not a bankable base case.
For Utilities & Grid Planners
- Prioritize transmission and transformer procurement well ahead of signed capacity commitments, given multi-year equipment lead times already reported across major markets.
- Explore direct commercial structures — large-load tariffs, behind-the-meter generation, flexible curtailment agreements — that let hyperscale demand connect faster without shifting cost onto residential ratepayers.
For Policymakers & Regulators
- Streamline interconnection queue reform in parallel with data center permitting, since power and construction approvals are currently moving on mismatched timelines.
- Track cumulative regional power and water commitments across projects rather than reviewing sites individually — city-level concentration (Dublin, Amsterdam, Northern Virginia) shows how quickly aggregate strain outpaces any single project's footprint.
Frequently Asked Questions
How much electricity will data centers use by 2030?
The IEA's April 2026 base case projects global data center electricity consumption reaching approximately 950 TWh by 2030, roughly double the 485 TWh consumed in 2025.
Why are hyperscalers signing nuclear power deals?
Grid interconnection queues now average around five years in many U.S. markets, and AI facilities need continuous, high-capacity baseload power that intermittent renewables alone can't reliably guarantee. Nuclear and SMR agreements let hyperscalers secure long-term supply outside the standard utility queue.
What share of U.S. electricity will data centers use by 2030?
Estimates range from about 9%–12% (EPRI, McKinsey) to as high as 17% (EPRI upper case), up from roughly 4%–5% today. Some regional grids, like Virginia's, could see data centers approach 39%–57% of local demand.
Is power availability really a bigger constraint than land for new data centers?
Yes, according to CBRE, JLL, and the IEA. Power shortages are now the leading cause of construction delays, and less than 10% of existing U.S. data center inventory is considered AI-ready in terms of power density and cooling.
Core Insights Review contributors publish research-based analysis and editorial insights on commercial real estate, PropTech, smart infrastructure, sustainable construction, industrial real estate, and emerging technologies shaping the future of the built environment.
Check for more information: Core Insights Review
