Managing the Environmental and Climate Impact of AI  

AI for sustainability

AI will fast become a key element to address environmental concerns. We see this already in:

Energy efficiency: AI-powered smart grids can support the development of more efficient systems, making energy available to customers as and when needed. There is further scope for AI to improve energy-intensive processes in factories.

Waste management systems: AI can contribute to increasing recycling rates by identifying materials for sorting—reducing pressures on methane-emitting landfills.

Public transport systems: AI allows mass transit systems to be operated more efficiently, reducing congestion times and associated vehicle emissions.

Sustainable agriculture: AI tools already allow monitoring of weather and soil conditions, thereby improving water usage and pest management efficiency.

Climate modeling: AI climate modeling tools will support climate solutions for mitigation and adaptation action, including through the UNFCCC’s AI4Climate Action Initiative. With the incidence of extreme weather events increasing, climate modeling will also provide more accuracy on time and localized information.

Data center power needs are surging

Moody’s now expects global data center capacity to double over the next five years—driven in large part by AI. Amid concerns about rapidly growing usage, Big Tech risks being targeted as the “new oil and gas.” Its impact on often-fragile power grids, the emissions intensity of the power it uses, and its vast demand on increasingly challenged freshwater resources have raised alarm bells as new data centers are added.

An AI query consumes up to ten times the energy of a standard Google search. This means data centers draw on vast amounts of power, threatening the stability of often outdated and fragile power grids. The International Energy Agency (IEA) estimates that data centers already accounted for 1%–1.5% of global electricity consumption in 2022—with vast regional differences.

In the US, data centers already used 4% of supplied energy in 2022—a figure expected to grow to 6% by 2026. In Europe, data center power consumption is expected to almost triple by 2030 and will require a big increase in electricity supply, according to McKinsey. Goldman Sachs estimates power demand in Europe could grow by 40% over the next ten years. Data centers are therefore expected to account for around 5% of total European energy consumption over the next six years compared to around 2% today. In the Republic of Ireland alone, data centers account for 21% of the country’s electricity final consumption.

As these increasing needs will not be covered fast enough by low-carbon power sources where and when they are needed (including due to insufficient grid infrastructure upgrades), in the short-term greenhouse gas-emitting gas generation is growing in most markets. Gas combined with carbon capture and storage (CCS) could allow the scaling up of energy supply rapidly with a reduced carbon footprint. ExxonMobil plans to operate power plants fueled by natural gas and equipped with carbon capture technology to supply data centers with energy, and Chevron is considering similar moves. Supply would be ensured by directly plugging into the data center—avoiding power utility and associated grid connection issues. CCS deployment at scale, however, remains largely unproven.

There are not many options: AI data centers will increasingly have to be built next to energy production sites where they can possibly access low-cost and low-emission power continuously. Big Tech have become large investors into wind and solar energy generation, but in the US, they are increasingly turning to baseload from nuclear power where they are willing to offer long-term off take contracts, or power purchase agreements (PPAs). This is illustrated by Amazon supporting the development of small modular reactors (SMR)—up to 300MW—in Washington State and in Virginia; Google signing PPAs from other SMRs yet to be built; and Microsoft signing a 20-year PPA from a reopened unit at Three Mile Island plant (shuttered in 2019).

Is the integrity of Big Tech’s net zero targets ensured?

Access to reliable emissions data from the many AI-enabled tools remains problematic. Morgan Stanley forecasts that global emissions from data centers could almost triple by the end of the decade. Data centers—because of the buildout in generative AI—could contribute 5.1% of global emissions by the end of the decade versus 1.9% in 2024. Google and Microsoft indeed blamed AI for their increases in market-based emissions. To meet their greenhouse gas protocol-aligned net zero targets, Big Tech companies have resorted to the use of both renewable energy certificates (RECs) and carbon offsets—two increasingly controversial measures.

Renewable Energy Certificates 

RECs are purchased by companies to demonstrate that they match a share of their energy consumption by buying renewable energy electricity elsewhere. This allows for a calculation of “market-based” emissions that do not depend on time and location of generated renewable power. This is different from “location-based emissions,” which account for actual emissions generated where operations are located. While the Greenhouse Gas Protocol allows for the use of RECs, their integrity is being increasingly questioned as both timing and location matter in terms of real global emissions. Indeed, using certificates issued in one area—while operations take place in another—would technically allow buyers to understate their reliance on fossil fuel-based power where their operations take place. Large technology groups are already by far the biggest corporate buyers of RECs, yet the sum of their location-based emissions was estimated to be 275% higher than the sum of official figures (including RECs). Furthermore, the relationship between REC trading and investments flowing toward renewables remains potentially tenuous, as their low prices provide little additional incentives. 

Carbon Offsets

Large carbon offset purchases have been announced by Big Tech companies, led by Meta purchasing close to 4 million carbon removal credits from forest restoration projects in Latin America and Microsoft acquiring 500,000 carbon credits from a direct air capture project (DAC) in Texas. The use of carbon offsets in net zero strategies, however, remains a controversial approach amid doubts on their emissions reduction claims.

Water: An additional critical issue

Data centers are water-intensive. The mining and manufacturing needed to produce AI hardware are already highly water-intensive and polluting processes, but data centers housing powerful servers generate a lot of heat with enormous water needs to cool operations. The water consumption of data centers globally is expected to keep going up. In the US state of Virginia—host to the world’s largest concentration of data centers—water consumption in the “data center alley” jumped by two-thirds since 2019. The issue becomes particularly critical when data centers are located in particularly water-stressed areas, where they face mounting opposition over who gets priority access to the scarce resource. One reason for the high water consumption is limited water reuse in cooling; circular water solutions therefore become essential for the sustainability of data centers.

What to watch

Certification of data centers according to their environmental and climate footprint would increase trust in the operations of data centers. Technological progress with more energy-efficient computing elements and grid upgrades would further reduce the tensions around aging power infrastructure.

Certification

The AI Action Summit to be hosted by France (February 10–11, 2025) is to focus on the environmental impact of AI—with the possibility of ranking AI companies according to their carbon and environmental footprint. A new label developed by the Swiss Datacenter Efficiency Association (SDEA) considers all critical aspects of data centers and IT infrastructure performance into a single certification. It includes power usage effectiveness, heat recycling, server utilization, and carbon emissions. It integrates these elements into a unified framework that covers both infrastructure efficiency and environmental sustainability.

Technological progress 

Chipmaker Nvidia recently unveiled a new chip that would be 25 times more energy efficient than previous models.

Grid upgrades

More effective government interventions are needed to upgrade crumbling grids to ensure that scarce low carbon energy can be made available when and where it is needed.

Materials presented by Edelman's public & government affairs experts. For additional information, reach out to Nikolaus.Schultze@Edelman.com