Creating Resilience
Data Centers, AI and Sustainability: Navigating the Carbon Paradox
April 2, 2025 3 Minute Read
Data centers, often viewed as energy and water-intensive resource consumers, have come under scrutiny from a sustainability perspective. However, they also provide environmental benefit by supporting remote work, which reduces emissions from employee commuting. Additionally, data centers enable content delivery without the need for physical media, resulting in reduced emissions associated with transportation, production and disposal.
AI is projected to consume between 85 and 134 TWh of electricity annually by 20271, which is comparable to the total energy consumption of Singapore over approximately 1.5 to 3 years. This represents a significant increase, given that data centers collectively consumed 460TWH in 2022. This anticipated rise could exacerbate the prevailing negative perception of data centers as energy-intensive operations. However, AI also offers potential for net energy savings, a paradox that characterizes the ongoing debate about the relationship between data centers, AI and sustainability.
Understanding how to realize these opportunities is essential, and we outline three strategies for achieving net energy savings.
-
Facility Electricity Consumption Optimization
AI can be used to optimize energy consumption in data centers and other asset types such as manufacturing facilities, corporate offices and residential properties. By analyzing energy-usage patterns, forecasted weather and expected occupancy levels, AI can create consumption strategies that optimize energy use during periods of affordable, clean-energy availability, while also reducing the on/off cycling of equipment on-site.
Unlike consumer AI systems such as large language models (LLMs), optimization-focused AIs require relatively few data points—around 10,000 compared to billions for LLMs.2 This significantly reduces the computational power needed for each operation while still providing substantial benefits.
In data centers or manufacturing facilities, operators can enhance optimization by scheduling production or processing activities to coincide with periods of low-carbon energy availability.3 This can have a significant impact, particularly if data is shared with the grid operator.
Additionally, how the plant reacts to the optimized energy consumption can inform the specifications for replacement equipment. This rightsizing improves efficiency and reduces the embedded carbon and water associated with manufacturing new equipment, particularly if the replacements require fewer resources to produce.
-
Smart Grids
AI models can quickly and accurately predict grid behavior by analyzing past trends and current data, such as consumption, time of day and weather conditions. This capability could be further leveraged to better maintain grid balance while prioritizing the most cost-effective sustainable forms of energy.4
When integrated with consumer-level optimization products, AI enables a rapid response mechanism for deploying virtual power plants in buildings through consumption reductions or activating generator sets. Combined with spot pricing for electricity, this strategy can result in significant cost savings. It transforms facilities from mere consumers to “prosumers”—both producers and consumers—enhancing renewable energy integration and strengthening grid resilience.
-
Colocation of Data Centers: Heat Export and Low Carbon Electricity
Data centers used for AI applications often do not require the same low-latency performance as many other data center operations. This flexibility allows them to be situated in areas with high renewable-energy availability or areas that utilize district heating systems. Connecting to district heating systems can repurpose waste heat, thus reducing carbon emissions associated with the compute consumption. For instance, a data center in Sweden that hosts AI applications will emit approximately 40 times less CO2 than a comparable data center in Singapore for the same compute power due to Sweden’s low-carbon electricity availability. Sweden also has widespread availability of district heating, enabling further reductions in carbon associated with the compute power.
Potential Benefit Balance of AI on Energy
By leveraging the benefits from optimizing facility energy consumption, smart grids or colocation of data centers, significant reductions in energy use and carbon emissions can be achieved. The table below illustrates the potential return on investment (ROI) in terms of energy savings per kilowatt-hour (kWh) and reductions in carbon emissions (kgCO2e) by investing in energy optimization AI.
| Energy/ Carbon impact of Energy Efficiency AI | Expected Reductions Yielded from Energy Efficiency AI | ROI of resource investment | Assumptions | |
| Energy | 1kWh | 10 – 140 kWh | 1,000%-14,000% |
Between 1%-2% of total AI consumption used for Energy Efficiency AI Between 1%-5% of global air conditioning energy reduced by the AI optimization (based on a study outlining 8%-19% of building consumption reduction) |
| Carbon | 1kgCO2 | 115 – 1,700 kGCO2 | 11,500%-170,000% |
Same assumptions as above AI consumption has a carbon factor of 0.04kgCO2e/kWh (Sweden Carbon Intensity) Saved energy carbon intensity 0.481 kgCO2e/kWh (global average electricity carbon factor) |
| Energy/ Carbon impact of Energy Efficiency AI | Expected Reductions Yielded from Energy Efficiency AI | ROI of resource investment | Assumptions | |
| Energy | 1kWh | 10 – 140 kWh | 1,000%-14,000% |
Between 1%-2% of total AI consumption used for Energy Efficiency AI Between 1%-5% of global air conditioning energy reduced by the AI optimization (based on a study outlining 8%-19% of building consumption reduction) |
| Carbon | 1kgCO2 | 115 – 1,700 kGCO2 | 11,500%-170,000% |
Same assumptions as above AI consumption has a carbon factor of 0.04kgCO2e/kWh (Sweden Carbon Intensity) Saved energy carbon intensity 0.481 kgCO2e/kWh (global average electricity carbon factor) |
The increased resource demand associated with AI may contribute to the perception that AI and data centers are at odds with sustainability goals. However, it is important to recognize the significant energy and carbon emissions reduction benefits that this technology can provide. By weighing the inputs against the benefits, a more nuanced, holistic view of the advantages of leveraging technology emerges. This shift could help reframe the perception of AI and data centers to acknowledge their potential contributions.
Like any resource-intensive technology, AI can have substantial environmental impacts. Therefore, it’s crucial to optimize resource use and minimize consumption from the outset. By doing so, we can harness the full potential of AI to accelerate sustainability.
2 https://www.phaidra.ai/blog/the-4th-industrial-revolution-building-a-hyper-efficient-sustainable-future
3 https://www.weforum.org/agenda/2024/07/generative-ai-energy-emissions/
4 https://www.iea.org/commentaries/why-ai-and-energy-are-the-new-power-couple
Related Insights
-
As demand for data storage and processing continues to rise, there is a growing need to design, build and operate efficient data center infrastructure.
-
CBRE explores top data center markets in North America, reporting on supply, demand, and pricing trends for 2024.
-
A worldwide power shortage is limiting the global data center market’s growth and fueling price increases. AI is projected to drive future demand.
