The data center boom threatens climate goals. How to fight back
Amazon, Google and Microsoft aren’t the only companies that should worry about the massive increase in electricity required for artificial intelligence. Read More
Data centers will account for up to 12 percent of all U.S. electricity consumption by 2028, or 132 gigawatts annually, triple what they consumed in 2023, according to a new analysis by Lawrence Berkeley National Laboratory.
That’s roughly equivalent to the annual power demand of countries such as Norway or Sweden.
Energy efficiency improvements and the addition of more solar and wind power to the U.S. grid helped tame emissions associated with that growing load over the past decade, but the voracious appetite of servers and networking gear needed to power artificial intelligence algorithms and services are upsetting that balance, the analysis published Dec. 20 found.
Measures can be taken by corporate sustainability teams when looking to lessen the impact of their companies’ digital strategies that starts with creating a checklist for making responsible data center selections when considering software development.
“The current and possible near-future surge in energy demand highlights the need for future research to understand the early-stage, rapidly changing AI segment of the data center industry and identify new efficiency strategies to minimize the resource impacts of this growing and increasingly significant sector in our economy,” Lawrence Berkeley researchers said.
Tough balancing act for Amazon, Google and Microsoft
The expansion will make it tougher for the biggest cloud computing companies — Amazon Web Services, Google and Microsoft — to hit emissions reduction and water conservation targets set prior to their AI expansion strategies. This will be a dominant theme for these companies, and their corporate customers, in 2025 and beyond.
The three spent a combined $180 billion on data center expansion in 2024, according to technology research firm Dell’Oro Group. To mitigate the emissions associated with that boom, they’re racing to find carbon-free solar, wind and geothermal electricity to power the sites, negotiating new sorts of utility tariffs and inking unique deals to bring nuclear power back onto the U.S. grid.
Still, it will be tougher for Amazon, Google and Microsoft to deliver on short-term goals for the 2025 and 2030 time frame. Amazon’s emissions are up 34.5 percent since 2019, Google recorded a 48 percent increase since the same year, and Microsoft’s footprint is up 29.1 percent since 2020.
Beware the data center ripple effect
The hyperscalers’ struggles should be top of mind as companies look at data centers to run their artificial intelligence and digitization services, but most aren’t prioritizing sustainability metrics, according to a summer 2024 survey of 1,200 leaders responsible for IT investments.
More than three-quarters (76 percent) of the respondents reported that AI has progressed beyond limited adoption at their company, but only one-third (33 percent) listed sustainability metrics such as energy consumption among their top three priorities for deployments.
“Every company that is participating in the AI revolution is contributing to the problem and having trouble figuring out how to curtail” the impact, said Simon Ninan, senior vice president of business strategy at IT infrastructure consulting firm Hitachi Vantara.
One challenge is that there’s frequently no centralized buying authority for AI applications; investment decisions are often made by business division heads, Ninan said. “It’s easy to think of this as someone else’s problem, he said.
Collaborate with IT decision makers
Sustainability teams can mitigate the potential impact of their company’s digitization strategy by teaming with counterparts in information technology to create a checklist for responsible data center infrastructure and software development decisions, according to Ninan and other experts.
Information about co-location providers and data center developers represents the fastest growing part of the index compiled by GRESB, which provides ESG performance data for real estate and infrastructure investors.
A large majority of investors (90 percent) believe metrics about water and energy consumption are material, especially as they’re considering the average of $1 billion in capital it takes to build a new data center complex, said Chris Pyke, chief innovation officer for GRESB.
Nearly 100 percent of the companies reporting to GRESB in 2024 provide information about energy and emissions; north of 90 percent track data for air pollution, water, waste and biodiversity and habitat.
Considerations for data center selection
Companies can future-proof their data centers by thinking differently about where they’re located and how they’re designed, said Jennie Karnes, vice president of data center solutions for commercial real estate services and investment firm CBRE.
“The data center of tomorrow is going to be built very differently than today,” Karnes said during a December briefing about data center trends.
Here are factors climate-conscious companies should consider as part of their data center selection process as demand explodes, according to Ninan, Karnes and other CBRE experts.
Geographic location
Companies should study the decarbonization plans of local utilities and understand the impact today and in the future for the electric grid serving that site, said Mike Lash, first vice president for the CBRE data center solutions team. While it’s likely corporations will spread their AI development across multiple locations, the power needed for training complex algorithms is best served by a well-connected, centralized location.
Ohio, for example, has become a hotbed for expansion — AWS alone is spending $10 billion in the state — largely because American Electric Power has made substantial investments in transmission infrastructure, which has been a sticking point for other regions, Lash said.
Geography also matters from a resilience perspective. As the climate warms, that’s a consideration for cooling systems — which can demand a lot of both power and water. Picking a location that’s at lower risk of flooding and natural disasters is also essential.
Regional capacity addition plans
Antiquated permitting processes have contributed to a seven-year backlog in getting new projects connected to the U.S. electric grid. States and regional transmissions organizations with shorter backlogs stand to benefit.
A related factor is the land available for projects and existing interconnection infrastructure, which could favor cities and regions seeking to capitalize on decommissioned coal plants, existing renewable capacity or industrial infrastructure investments. That’s important for on-site renewable energy investments; solar could require a site of 3,000 to 6,000 acres, which can be a limiting factor.
Among states seeking to attract projects:
- Indiana, which has announced big deals with Meta and Google, even though the state’s renewable energy aspirations are relatively small.
- Michigan, which is offering big tax incentives through a bill approved by its legislature that’s pending approval by Gov. Gretchen Whitmer. There’s concern that the measure could derail Michigan’s plan to rely entirely on renewables by 2040.
- Pennsylvania, which offers tax breaks for data center equipment.
Climate strategies of potential partners
The most efficient AI strategies will rely on a mix of resources hosted with cloud computing companies and equipment managed on site or in a data center co-location facility. “Hybrid is the answer and more companies are coming to terms with that,” Hitachi Vantra’s Ninan said.
For that reason, companies investing in AI should review the emissions reduction roadmaps and other environmental sustainability plans — particularly water consumption — of potential data center partners.
The world’s biggest co-location company, Equinix, for example, has pledged a 50 percent reduction in its Scope 1 (direct operations) and Scope 2 (purchased energy) emissions by 2030. Digital Realty (No. 2) has committed to reducing the emissions from Scope 1 (direct operations) and Scope 2 (purchased energy) by 68 percent per square foot by 2030. Equinix has pledged a 50 percent reduction in the same time frame.
Both companies are roughly halfway toward those goals. Their goals for water conservation are less detailed, although Digital Realty ironically uses AI to identify opportunities to prevent leaks.
So far, none of the Big 3 cloud computing companies have adjusted their emissions reduction commitments. They’re stepping up purchases of carbon-free energy, including nuclear and geothermal generation capacity; and investing in data center design changes and processors intended to counter climbing energy use, according to spokespeople.
Microsoft, for example, signed a 20-year purchase agreement to bring a shuttered nuclear reactor at Three Mile Island back online. That deal brings its portfolio of renewable energy projects to 34 gigawatts in 24 countries. Amazon had contracted for about the same capacity, as of February 2024, and Google is also a voracious buyer of “carbon free” energy.
AWS is adopting a broad range of design component changes and liquid cooling approaches designed to deliver a 46 percent reduction in mechanical energy consumption. All three companies are allied closely with AI chip leader Nvidia, which pitches its graphics processing units as a way to improve data center efficiency due to much faster processing times.
Google and Microsoft are also becoming more involved with development at the ground floor. In September, Microsoft disclosed an AI infrastructure partnership with $100 billion in investment potential in collaboration with investors BlackRock, Global Infrastructure Partners and MGX.
Google, meanwhile, in December announced a $20 billion partnership with Intersect Power, a clean energy project development, and private equity firm TPG Rise Climate to “synchronize clean power generation and data center growth in a novel way,” as Google parent company Alphabet President Ruth Porat put it in Google’s Dec. 10 blog post.
The venture promises to speed the time it takes to get renewables onto the electric grid, with the first project expected to be partially operational in 2026.
“Deep, collaborative partnerships combined with creative problem-solving are the only way that we can meet the explosion of AI growth, as well as society’s accelerating electricity demand,” said Intersect Power founder and CEO Sheldon Kimber. “Our approach makes it possible to do all of this and unlock meaningful opportunities for rural economic development along the way.”
[Master the essential skills to tackle your company’s unique sustainability challenges at GreenBiz 25, happening Feb. 10-12 in Phoenix, Arizona.]
Subscribe to Trellis Briefing
