AI-Driven Data Center Growth: Energy Demand and Grid Impacts

Introduction

The rapid expansion of artificial intelligence, cloud computing, and large-scale digital services is transforming global computing infrastructure and reshaping energy systems. Nowhere is this shift more visible than in the accelerating construction of data centers. These facilities are becoming one of the most energy-intensive categories of large industrial and commercial infrastructure, influencing electricity generation planning, transmission expansion, carbon reduction strategies, and regulatory policy.

The United States currently hosts the largest number of data centers in the world, accounting for nearly 40 percent of global facilities and up to 5 percent of national electricity demand. With more than 3,000 operational data centers and hundreds more planned, growth in energy consumption from this sector is unprecedented. The next decade will determine whether the rise of AI computing contributes to increased carbon emissions or becomes a catalyst for rapid grid modernization, large renewable energy deployments, and advanced system-level optimizations.

This article provides a technical assessment of the siting, energy, and emissions impacts of emerging data centers, along with analytical insights and policy considerations. This article was adapted from the following publication: Springer, Cecilia and Ali Hasanbeigi (2024). Data Centers in the AI Era: Energy and Emissions Impacts in the U.S. and Key States. Global Efficiency Intelligence, LLC.

1.0 Types of Data Centers 

1.1 Hyperscale Facilities

Hyperscale facilities support tens of thousands of servers and are designed for extreme scalability and advanced automation. They are operated almost exclusively by major technology companies. They also feature highly optimized cooling systems and advanced electrical infrastructure. Their energy consumption is large enough to influence local grid development, renewable procurement markets, and long-term capacity expansion planning.

1.2 Colocation Facilities

Colocation centers lease space, power, and cooling to multiple customers who install their own computing hardware. Although individual tenants vary widely in scale, the aggregated demand at a single site can rival that of hyperscale facilities. Colocation operators are major buyers of renewable electricity and have well-established energy reporting and efficiency programs.

1.3 Enterprise Data Centers

Enterprise data centers are owned and operated by individual companies for their own computing requirements. They are typically smaller than hyperscale or colocation facilities but still contribute meaningful aggregate demand at the regional level. Enterprise centers vary widely in technical sophistication and efficiency.

1.4 Edge Data Centers

Edge centers are small, distributed facilities located close to user bases and telecommunications networks to reduce latency. They support services like content delivery networks, IoT platforms, autonomous vehicle communication, gaming networks, and localized processing.

2.0 Energy Use Characteristics of Modern Data Centers

Data centers rely on a complex mix of IT hardware, cooling infrastructure, mechanical systems, and electrical systems. 

2.1 IT Load

IT equipment typically accounts for approximately 60 percent of total facility energy use. The primary contributors are:

• Servers for compute operations

• Storage systems

• Networking equipment

• Ancillary communications and IT infrastructure

The growth of AI and large language models significantly increases the computational intensity of servers, increasing power density and elevating cooling requirements.

2.2 Cooling Systems

Cooling consumes roughly 30 percent of total energy use. Cooling system types include:

• Computer Room Air Conditioners (CRAC)

• Computer Room Air Handlers (CRAH)

• Chillers

• Evaporative cooling towers

• Pumps and fans

• Emerging liquid cooling systems

Cooling is critical for maintaining acceptable thermal conditions to prevent equipment degradation and reduce operational risk. Facilities in hot climates or urban areas with limited access to free cooling typically exhibit higher cooling loads.

2.3 Power Infrastructure

Electrical infrastructure contributes up to 10 percent of total energy use, primarily through:

• Uninterruptible Power Supplies (UPS)

• Power Distribution Units (PDUs)

• Transformers

• Switchgear

• Conversion and distribution losses

2.4 Backup Generation

Most facilities deploy diesel or natural gas generators capable of supplying between 600 kW and 3,500 kW per generator. Although used infrequently, these units impact local air quality and emissions inventories.

2.5 Power Usage Effectiveness (PUE)

PUE is the primary metric for evaluating data center efficiency. It is the ratio between the total facility energy and IT equipment energy. Average PUE across the industry has improved from roughly 2.5 in 2007 to about 1.6 today. Cutting-edge hyperscale facilities report PUE values close to 1.1. Highly efficient cooling systems and favorable climates support lower PUE values.

3.0 Geographic Patterns in Data Center Siting

Data center developers evaluate three foundational siting criteria:

1. Availability of land: Large campuses require significant acreage and plans for future expansion.

2. Availability of reliable, low-cost electricity: Data centers operate with continuous high demand and require resilient grid interconnections.

3. Proximity to fiber networks and end users: Latency requirements and network resiliency influence siting near major population centers and fiber backbones.

Additional siting factors include natural disaster risk, permitting environments, local zoning rules, tax incentives, access to renewable energy, and the maturity of regional supply chains.

4.0 Projected Electricity Consumption Through 2035

Future data center electricity consumption is highly sensitive to growth in AI computing. To account for uncertainty, two growth scenarios are considered:

4.1 Growth Scenarios

• Conservative Growth: 5 percent annual IT load growth through 2030, then 3 percent through 2035.

• High Growth: 10 percent growth through 2030, then 7.5 percent through 2035.

4.2 Efficiency Scenarios

Two PUE trajectories illustrate the effect of operational efficiency:

• Business As Usual Efficiency: gradual decline from 1.58 to 1.4 by 2035

• Advanced Efficiency: decline to 1.2 by 2035

4.3 U.S. Electricity Use Projection

Under combined growth and efficiency scenarios:

• By 2030, U.S. data center electricity demand could reach 345 to 490 TWh per year.

• By 2035, demand could reach up to 655 TWh if growth is high and efficiency improves slowly.

• Achieving widespread PUE values near 1.2 could reduce consumption by approximately 14 percent relative to business as usual efficiency.

5.0 Projected Emissions Trajectories Through 2035

Future emissions depend on three interacting factors:

1. Growth in computing demand

2. Progress in efficiency and PUE improvement

3. Share of electricity supplied by renewable energy

5.1 Renewable Electricity Procurement Scenarios

Two scenarios illustrate renewable adoption:

• Baseline Renewable Procurement: renewables supply 35 percent of data center electricity by 2030 and 45 percent by 2035.

• Ambitious Renewable Procurement: renewables supply 60 percent by 2030 and 80 percent by 2035.

Many hyperscale operators already procure renewable electricity aggressively, but smaller operators lag behind.

5.2 U.S. Emissions Projection

Under baseline renewable adoption:

• Emissions peak around 63 to 83 Mt CO2 per year in 2030, depending on growth.

Under ambitious renewable adoption:

• Emissions flatten or decline even as electricity demand increases.

• Renewable procurement has a larger impact on emissions reduction than efficiency improvements alone.

6. Technological Innovation Trends Affecting Future Energy Use

Several emerging technologies will play a central role in shaping the long-term energy profile of data centers. These innovations influence cooling performance, electrical efficiency, operational flexibility, and overall facility emissions.

6.1 Liquid Cooling

AI and high-performance computing workloads generate far higher heat loads per rack than traditional cloud servers. As power densities rise, conventional air cooling becomes insufficient. Liquid cooling technologies, including direct-to-chip cooling, immersion cooling, and hybrid liquid-air systems, remove heat more efficiently and enable higher rack densities. These systems significantly reduce cooling energy consumption and improve thermal stability, especially in regions with limited free cooling potential.

6.2 AI-Optimized Workload Scheduling

Advances in software and cloud orchestration now allow computing loads to be scheduled according to periods of lower grid carbon intensity. These systems shift non-critical workloads to cleaner electricity hours without modifying PUE. This strategy can substantially reduce operational emissions and aligns data center operations with renewable availability and grid conditions.

6.3 High-Voltage Direct Current Distribution

High-voltage DC distribution is gaining traction as a method to reduce electrical conversion losses. Traditional AC systems require multiple stages of rectification and inversion that introduce efficiency penalties. DC distribution minimizes these steps, improving overall electrical efficiency, enhancing resiliency, and simplifying integration with battery storage and renewable systems.

6.4 Grid-Interactive Energy Storage

Lithium-ion UPS systems are evolving into full grid-interactive assets. Modern UPS systems can charge and discharge in coordination with the grid to support:

• Peak shaving

• Frequency regulation

• Onsite energy arbitrage

• Microgrid islanding and ride-through

This capability reduces reliance on diesel generators, enhances resilience, and allows data centers to participate in ancillary service markets while supporting grid stability.

6.5 Nuclear Energy and Small Modular Reactors

Data center operators are increasingly exploring long-term nuclear power purchase agreements. Small modular reactors are particularly attractive for high-density hyperscale campuses due to their steady baseload output, compact footprint, and zero-carbon generation. Several major operators have expressed interest in SMR integration as a future pathway for dedicated clean power supply.

7. Policy Recommendations for Managing Energy and Emissions

Managing the rapid growth of electricity demand from data centers requires a coordinated policy framework that integrates grid planning, renewable procurement, load management, transparency, and efficiency standards. The following interventions provide a structured approach for regulators and utilities.

7.1 Renewable Energy Procurement Expansion

Policymakers can strengthen clean electricity supply for data centers by supporting:

• Long-term power purchase agreements

• Hourly or 24/7 clean energy matching programs

• Incentives for colocated or directly connected renewable generation

• Utility green tariff programs tailored to large loads

These mechanisms increase renewable penetration and reduce emissions from rapid load growth.

7.2 Improved Load Management

Data centers can support grid stability by actively participating in load flexibility programs. Effective measures include improved forecasting, demand response participation, and shifting discretionary compute loads away from peak system demand periods.

7.3 Grid Infrastructure Optimization

Future-ready grids require investments in:

• Transmission expansion

• Streamlined and efficient interconnection processes

• Wider deployment of grid-interactive UPS and storage systems

• Integration of long-duration energy storage

These upgrades increase renewable hosting capacity and prevent over-reliance on fossil-fuel generation.

6.4 Stronger Data Center Energy Management

Energy management practices can be strengthened through benchmarking systems, incentives for achieving low PUE, and requirements for onsite or near-site energy storage to reduce peak impacts and enhance resilience.

6.5 Improved Public Data and Reporting

Transparent and standardized reporting on energy use, emissions, and renewable procurement helps utilities and regulators plan more accurately. Better data also supports improved long-term forecasting and risk mitigation.

7.0 Conclusion

The coming decade will play a defining role in determining how data center growth influences the energy and emissions trajectory. Demand for AI and cloud computing continues to rise at an accelerated rate, and data centers are becoming key drivers of electricity planning, renewable energy expansion, and grid modernization.

Electricity demand from data centers may double or even triple by 2035. Emissions outcomes, however, depend heavily on renewable energy procurement, efficiency improvements, technological innovation, and regulatory frameworks. With aggressive renewable adoption and advanced operational strategies, data center emissions can stabilize or decline even as computing requirements increase.

Collaboration among policymakers, utilities, regulators, engineers, and technology companies is essential. The choices made today will determine whether data centers act as a major source of emissions growth or become drivers of clean energy transformation and system optimization. The direction chosen will shape the future of digital infrastructure and the broader electricity system.

Reference:

This article was adapted from the following publication: Springer, Cecilia and Ali Hasanbeigi (2024). Data Centers in the AI Era: Energy and Emissions Impacts in the U.S. and Key States. Global Efficiency Intelligence, LLC.