Achieving net-zero emissions by 2050 appears increasingly elusive. Both Google and Microsoft have acknowledged that their emissions are on the rise due to activities related to artificial intelligence (AI) in data centers. Additionally, a July report indicated that Irish data centers account for 21% of the nation’s electricity consumption. What steps can data centers take to effectively reverse this trend?
Mark Acton, owner and chief analyst at Acton Consulting, emphasizes the urgent need for change within the industry. He believes that excessive emphasis on metrics like power usage effectiveness (PUE) is misguided, especially as modern facilities have already reached a point of diminishing returns in these areas. “Data centers don’t create the energy demand or consume power,” Acton explains. “They simply act as an overhead to the power delivered to the IT systems they support. We need to focus on optimizing the IT load.”
Typically, data centers operate with low utilization rates, excessive redundancy, and software duplication, none of which is optimized for energy efficiency or environmental impact. Acton notes that this demand is fundamentally driven by our choices in digital services and platforms. For real change to occur, consumers must adjust their behaviors and preferences, although access to information is currently lacking.
The concept of an “energy passport” for consumers could provide insights into energy consumption and promote new charging models, or even a tax on digital services, according to Acton. Currently, many digital services seem “free” for users, obscuring the total costs, including their emissions. Imagine if internet usage were charged by the hour or if there were small fees for images and uploads. “This could limit demand,” Acton argues. “There are far more effective strategies we could adopt. An energy-labeling scheme for digital services could help consumers make informed choices, perhaps even issuing warnings like, ‘Consuming this service may harm the planet.’ That could result in more significant energy and environmental benefits than merely adjusting regulations or tinkering with PUE.”
Furthermore, software must be created with energy efficiency in mind to tackle the challenges of extensive redundancy and the need for standby systems. A data center can be highly efficient, but if multiple similar centers exist, those efficiencies may cancel each other out. Acton argues, “To label that as ‘efficient’ is poor accounting.” However, large corporations often view such issues as minor expenses or consider them to be someone else’s responsibility.
Max Schulze, director of the green think tank Leitmotiv Digital, points out a long-standing misconception that digital services are burden-free. We now face the challenge of creating business models that allow profitability without harming the environment, a feat that few companies have managed to achieve. “Even self-repairable phones still require manufacturing,” Schulze notes.
It’s essential to scrutinize which technologies persistently consume excessive resources and to rethink carbon taxes and emissions trading, ensuring these measures don’t just enable organizations to sidestep genuine emissions reductions. Digital innovations must also make economic sense, especially in light of generative AI advancements. “We cannot subsidize these innovations in the hope that they will eventually resolve our problems,” Schulze cautions, “It’s unlikely and may simply result in wasted energy.”
Regulatory measures should enhance transparency regarding the environmental costs of digital services, thereby empowering consumers to make informed decisions. As Schulze highlights, consumers currently lack the ability to evaluate AI models based on their environmental impact, as such metrics are not available. Tools primarily compete on features alone.
We need a societal philosophical conversation about digitization. Many people do not genuinely need continuously delivered 30-second videos on their devices, Schulze suggests. This requires investors to incentivize different approaches, pushing for a rethinking of market structures. “Investors would be drawn to a diverse and mature digital market that prioritizes sustainability,” he asserts.
Matt Watts, chief technology evangelist at storage provider NetApp, concurs that there needs to be a more robust commitment to integrating emissions reduction strategies across all facets of business operations. “Servers use about 15% of data center power. While implementing more efficient storage systems can help, it results in only marginal improvements,” he explains.
Improving monitoring and measuring strategies is crucial for understanding the scope of energy waste. Watts highlights that around 41% of data remains unused, and analyst estimates suggest it could be even higher. Tackling data waste represents an accessible opportunity for substantial emissions reductions across numerous organizations. Education will be vital here, as change often encounters resistance—different departments frequently fail to take ownership of the problem. “Those generating data see it as the IT team’s responsibility, while the IT team considers themselves ‘just custodians’,” he says. This longstanding issue highlights the lack of a clear ownership structure in data management.
Matt Rees, chief technology and operating officer at Neos Networks, points out that while IT naturally consumes a considerable amount of power, it is also becoming smaller and more efficient. Furthermore, interconnected industries often mean that emission reductions in one area can lead to increases in another, complicating clarity on the best approaches to take. “Ultimately, we operate in a business designed to provide more bandwidth in response to demand,” Rees explains.
There’s a pressing need for organizations to achieve more with less, which might involve reconsidering the necessity for round-the-clock services. Other strategies should also be considered, such as integrating renewable energy sources and improving energy efficiency through the replacement of outdated equipment, effective cooling solutions, virtualization, and automation.
Fred Lherault, field CTO for EMEA at Pure Storage, echoes this sentiment, noting that AI’s power consumption is closely tied to GPU requirements, which often lack designs focused on power efficiency. While investing in new technology might initially escalate emissions, it is crucial to focus on timely access to necessary data over time. Lherault mentions that some manufacturers report average GPU utilization rates of only 35%, indicating that organizations sometimes acquire far more GPU capacity than needed, often due to delays in data or preparatory work in data science.
“Organizations must learn to achieve their goals with fewer resources, especially considering supply chain constraints,” Lherault emphasizes.
John Booth, managing director and principal consultant at Carbon3IT, highlights that despite years of established best practices, many organizations continue to make fundamental mistakes that lead to significant energy waste. “Moreover, many operators shift responsibilities for servicing and cooling management to external parties. These third-party providers often lack the specialized knowledge necessary for precise control in data centers,” Booth states.
Some organizations take service reports at face value, which typically follow conventional practices that do not focus on broader improvement strategies, including procurement policies that prioritize one-for-one replacements rather than innovative solutions. Many don’t stay updated on current standards or options for environmental monitoring. “In some cases, they could forgo chillers entirely, significantly reducing procurement costs and emissions,” Booth adds. He points out that the design communities for data centers often continue to apply outdated practices from several decades ago.
While best practices and new standards have evolved, there’s a question regarding how many organizations have modernized their operations, policies, and practices accordingly. Additionally, fostering development for more ethical and green financing is one of the numerous strategies available for progress, Booth notes.
Interestingly, around 15-20% of the UK’s energy supply comes from Europe via interconnectors. Booth suggests that funds spent on energy imports cannot be reinvested into developing a more efficient, renewables-driven energy grid in the UK. “We’re essentially providing billions of pounds for energy we could produce ourselves. While this energy might appear inexpensive and reduces the need for UK infrastructure, it represents a short-term strategy,” he concludes.