Data centers are the backbone of the digital world, supporting everything from cloud services to real-time data analytics, streaming, and telecommunications. These facilities house servers, storage systems, and networking equipment that process and store vast amounts of data every second. However, this intense computational demand comes with an equally significant energy requirement, making efficient and reliable power-sourcing critical data center operations.
The growth of power demand by AI-driven data centers is rapidly outpacing traditional energy requirements due to the intensive computational power needed for large-scale machine learning models and AI processing. The energy consumption of the data centers has surged to the point that they often consume as much electricity as small cities.
This demand puts a strain on the electric grid, challenging its capacity to deliver a stable supply while also prioritizing renewable energy transitions. The traditional grid infrastructure, designed for a more predictable and moderate load, now faces a dual challenge: not only meeting the heightened baseline demand from data centers but also balancing intermittent renewable energy sources like wind and solar.
Without significant investment in grid upgrades, including enhancements in capacity, reliability, and grid-edge technologies like battery storage, the grid struggles to support the escalating load from AI centers. Not to mention the timeline associated with the grid required investment does not correlate with the projected growing demand by data centers.
Nuclear and other consistent, low-carbon power sources are being explored as potential solutions to bridge this gap, providing the reliable and continuous energy needed to fuel the next generation of data-driven innovation. However, full utilization of small modular reactors (SMRs) and microreactors (after completion of the development and testing) is not expected before 2030.
Goldman Sachs’ Generational Growth report released in April 2024 predicts that the power use from data centers will increase 1.8x-3.4x compared to 2023 levels. The following graph (regenerated from the same report) shows the predicted global power demand by data centers until 2030.
Figure 1 – Global Data Center Electricity Consumption (source)
Data centers rely on various energy sources to maintain 24/7 operations and ensure uninterrupted service. Here are the primary power sources used by data centers worldwide:
1. Grid Power: Most data centers draw their primary power supply from the national or local electricity grid, which can vary significantly in terms of reliability and sustainability depending on the region. Grid power provides a steady source of electricity, but data centers often add redundancy systems to protect against grid outages.
2. Onsite Diesel Generators: Diesel generators are commonly used as backup power sources, particularly for critical applications that demand high uptime. In the event of a grid failure, diesel generators kick in almost immediately to supply power, ensuring continued operation. However, they are carbon-intensive and contribute to greenhouse gas emissions, so they are generally used as a last resort.
3. Renewable Energy (Solar, Wind, and Hydropower): Renewable energy sources like solar, wind, and hydropower are increasingly popular among data centers as companies strive for sustainable and eco-friendly operations. Solar panels are sometimes installed on or near data center buildings, and partnerships with wind and hydro farms are common. While these sources can offset grid usage, they often require backup solutions due to variability in energy production.
4.Battery Energy Storage Systems (BESS): Battery storage systems are becoming essential components of modern data centers. These systems provide short-term power in the event of an outage, acting as a bridge until backup generators or other systems are fully operational. Lithium-ion batteries are widely used due to their efficiency, although new technologies like flow batteries are emerging as viable options for long-duration storage.
5. Fuel Cells: Fuel cells, often powered by hydrogen, offer a cleaner alternative to diesel generators and can provide a continuous, scalable power supply. They produce power through chemical reactions rather than combustion, resulting in fewer emissions. Companies like Microsoft and Google have tested fuel cell systems to power specific data center operations, finding them promising for long-term sustainable energy use.
6. Natural Gas Generators (Gas Turbines and Gas Engines): Although less common than diesel, natural gas generators are used in some data centers as a more environmentally friendly alternative. They have lower carbon emissions and are more efficient than traditional diesel generators, though they still rely on fossil fuels.
7. Cogeneration (Combined Heat and Power): Cogeneration systems generate electricity and use the excess heat produced in the process, improving overall energy efficiency. Data centers can utilize the exhaust heat in special-type chillers to generate the required cooling load. With the advancement in liquid cooling in data centers, this configuration can be considered the most efficient.
These power sources offer unique benefits and trade-offs, from environmental impact to reliability and cost. Data centers often mix these sources, combining primary and backup power systems to create a resilient energy framework.
Typical Data Center Power Supply and Auxiliary Arrangement is shown in the following figure. Most data centers use redundant sources of power to ensure uptime. The power is distributed through a primary switch gear to supply power to UPS and battery backup which are connected to the PDUs & PSUs. In addition to the power distribution to support the power demand by IT equipment, a considerable part of the power consumed by the cooling system
Figure 2 – Typical Data Center Power Distribution
Major Consumers of Power in Data Centers
1. Servers and Storage Devices (45-55%): Servers and storage systems consume the largest share of power, usually around 45-55% of a data center’s total energy. These systems are always running, processing, and storing data, with more advanced servers for AI and machine learning tasks using even more power.
2. Cooling Systems (30-40%): Cooling systems are essential to prevent equipment from overheating, consuming 30-40% of total power. Cooling involves various methods, including air conditioning, liquid cooling, and increasingly, innovative solutions like immersion cooling. Energy-efficient cooling systems are becoming a priority due to their substantial energy draw.
3. Uninterruptible Power Supplies (UPS) (5-10%): UPS systems provide backup power during fluctuations or outages, maintaining a consistent power supply until generators or other backups engage. Typically, UPS systems consume about 5-10% of total power, mainly in standby mode, as they are essential for maintaining service continuity and stability.
4. Networking Equipment (5-10%): Routers, switches, and other networking gear are crucial for data transmission within the data center and beyond, accounting for about 5-10% of total energy use. This percentage can increase in data centers with high data traffic volumes or intensive communication requirements.
5. Power Distribution Units (PDUs) (1-3%): PDUs distribute power to all data center components. Though they consume less power individually (around 1-3%), their role in maintaining steady power flow and monitoring usage across devices is vital to overall efficiency.
6. Lighting and Security Systems (1-3%): Lighting and security, including surveillance cameras, access controls, and monitoring systems, together consume about 1-3% of total energy. While they represent a smaller share of the energy budget, they remain critical for the physical security and operational visibility of the facility.
6. Automation and Monitoring Systems (1-2%): These systems include software and hardware that monitors and manages data center operations, such as environmental conditions and energy use. Although they use only 1-2% of the power, they are essential for optimizing energy efficiency and enabling proactive maintenance.
The following graph provides an overview of power consumption by major consumers in data centers. The graph is generated based on the reported results by Kazi et al. at IEEE Access, Volume 9., 2021.
Figure 3- Major power consumers in data centers (source)
Using microgrids and behind-the-meter power generation offers significant benefits for data centers, primarily by enhancing reliability, efficiency, and sustainability. Microgrids enable data centers to generate and manage their own power locally, which reduces dependency on the central grid and increases resilience against outages. This setup is especially valuable for mission-critical facilities that require uninterrupted power, as it allows data centers to operate independently or in parallel with the main grid. Behind-the-meter power generation, such as on-site solar arrays, fuel cells, and even microreactors, provides real-time energy that avoids transmission losses and reduces demand charges. This localized power generation also supports data centers’ sustainability goals, as it often incorporates renewable sources and reduces carbon emissions. Additionally, by managing energy consumption more flexibly, data centers can optimize energy costs, participate in demand response programs, and contribute to grid stability, creating both economic and environmental advantages.
Data center microgrids use a variety of fuel sources to maintain reliable and sustainable power. Common sources include natural gas and diesel for backup generators, providing quick response power during outages. Renewable sources like solar and wind are increasingly integrated, often paired with battery storage to manage their intermittent nature. Fuel cells, powered by hydrogen or natural gas, offer a clean and efficient option for steady, on-site power generation with low emissions. Emerging technologies, such as small modular nuclear reactors (SMRs), are also being considered to provide constant, carbon-free power for microgrids, making them an attractive future option for data center sustainability.
Selecting a fuel source to power a data center is a decision influenced by multiple parameters including:
- Energy Assessment
- Sustainability and Emission Reduction
- Availability
- Long-Term Costs and Savings
- Scalability
- Reliability
Considering that each data center is unique in terms of the fuel source and power generation technology the existing gap in the generation capacity of the grid and the rising power demand of the data center, selecting the correct power generation technology becomes critical in supporting the continuous development of data centers.
As presented in Table. 1, Solar, wind, and Nuclear are the fuel sources with zero direct emissions; however, there are some limitations with each technology:
- Solar and wind are not location-agnostic and can be generated in areas suitable for this type of generation. In addition, their availability is limited to the time that solar or wind power is available unless they are coupled with energy storage systems to increase their availability when the energy source (solar/wind) is not available.
- Nuclear technology development in the scale required by data centers is a long process and the earliest operation of this technology is expected by 2030.
The following chart provides an overview of the typical lead time for the main driver used for each technology. The shown timeline may vary based on the specific requirements and is presented to provide a figure in terms of the availability of each equipment type (excluding the site construction).
Based on the limitations of the zero-emission technologies and gird availability, the practical solution to support the immediate and growing power demand of the data centers is a hybrid microgrid including wind, solar and natural gas as the energy source for power generation. This configuration results in the lowest achievable emissions while generating reliable and available power to support this demand.
The microgrid solution shall be further optimized by utilizing an Energy Management System to ensure the best utilization of each energy source and maximum emission reduction. In addition, to further increase the overall efficiency of the microgrid, the exhaust heat from the natural gas power generation can be utilized in a chiller and provide the required cooling for data centers. This configuration is estimated to have an overall efficiency above 75%.
Figure 5 – Hybrid Microgrid for Data Centers
Conclusion
The hybrid microgrid microgrids offer several significant benefits to data centers in terms of reliability and cost:
- Grid Independence: Microgrids allow data centers to operate independently from the main power grid, ensuring continuous operation during grid outages or failures. This reduces vulnerability to wide-scale blackouts or disruptions in the utility power supply.
- Continuous Power Supply: Hybrid microgrids provide a steady and reliable power source, crucial for data centers that require uninterrupted operation. The consistent fuel supply through pipelines enhances overall reliability compared to other backup power options.
- Operational Continuity: By integrating combined cooling, heating, and power (CCHP) systems, renewables, and energy storage, microgrids ensure operational continuity even during extended grid outages as well as increased efficiency.
- Flexibility in Power Management: Microgrids offer the ability to dynamically route power and turn data halls on and off based on demand (and grid conditions). This flexibility allows data centers to optimize their power usage and respond quickly to changing needs or grid instabilities.
- Peak Shaving Capability: Microgrids can handle spikes in power demand, reducing reliance on the traditional grid during peak times. This not only enhances resilience but also helps in managing costs and grid stability.
- Improved Cold Weather Performance: Natural gas generators in microgrids are easier to start and operate in cold temperatures compared to diesel generators, enhancing reliability in harsh weather conditions.
- Reduced Dependency on Diesel Backup: Hybrid microgrids can reduce or eliminate the need for diesel generators, which are prone to failures and have limited fuel storage capacity.
- Integration of Multiple Energy Sources: The proposed hybrid microgrid configuration allows integration of various energy sources including renewables and energy storage, creating a more diverse and resilient power system.
- Islanding Capability: In case of grid disturbances, microgrids can “island” or disconnect from the main grid, continuing to power the data center independently.
- Advanced Energy Management: Sophisticated control systems in microgrids allow for real-time optimization of energy use, enhancing overall system resilience and efficiency.
By providing these capabilities, hybrid microgrids significantly enhance the ability of data centers to maintain operations during various disruptions, from short-term power quality issues to long-term grid outages, thereby greatly improving their overall resilience.
While the initial investment in hybrid microgrids may be slightly higher than traditional power solutions, the long-term benefits in terms of reliability, operational efficiency, and potential cost savings make them an attractive option for data centers. Additionally, as data centers strive for greener operations, hybrid microgrids offer a cleaner alternative to other power generation technologies, potentially leading to cost savings through improved environmental performance.
