Mainstream compute uses way more electricity than AI – So why aren’t we talking about it?
In this article
- Mainstream compute, not AI, will remain the dominant energy draw for years to come
- AI deserves attention, but shouldn’t overshadow the need to address traditional IT’s energy footprint
If you only followed the headlines, you’d think AI is about to single-handedly crash the power grid with its ravenous energy appetite. However, scrutinizing the numbers actually reveals a surprising reality: the lion’s share of the current and near-term future data center energy use isn’t from AI at all. It’s from everything else; the so-called “mainstream IT” found in corporate IT estates, colocation data centers, and cloud servers. It turns out that, while AI may be growing quickly, it’s mainstream compute that is predicted to be IT’s biggest data center energy hog.
Without a doubt, AI’s predicted energy footprint is significant and must be addressed, but it should be considered alongside the rest of IT’s energy use. The risk of focusing solely on AI could mean missing a major opportunity for the industry to contain IT’s ballooning energy demands overall.
It turns out that, while AI may be growing quickly, it’s mainstream compute that is predicted to be IT’s biggest data center energy hog.
Beyond the AI Hype: The Real Energy HogsMainstream compute’s dominant share of IT’s energy needs is borne out in numerous analyses of current and future-state IT. Well-regarded global research bodies estimate that AI was responsible for only 10%-15% of worldwide data center energy consumption in 2024, placing ubiquitous general-purpose IT at a whopping 85%-90%.1 The Uptime Institute’s research also finds AI-centric computing to only be responsible for a relatively small slice of overall data center energy consumption. Its modelling—depicted in the following infographic—indicates that AI will only account for 20-30% of data center energy use by the end of the decade, and estimates that the energy demands of mainstream compute servers running everyday applications will grow almost twice as much as AI computing during that time.2 Some researchers also point out that the 2030 AI energy figures could be overstated; highlighting that not all giant data center builds will necessarily come to fruition.3 These numbers drive home a provocative point: the world’s “digital workhorses” (mainstream IT systems) will collectively consume much more electricity than the splashy new AI systems in the foreseeable future. Yet, talk of IT sustainability nowadays nearly always defaults straight to AI.
To be clear, none of this is to dismiss the fact that AI’s energy consumption is growing at a faster rate than conventional servers, with fluctuating loads from AI training adding increased outage risks and challenges to power grids. We must keep in mind that although AI is a big deal, conventional IT isn’t shrinking, and we cannot afford to ignore it if we want to meaningfully reduce IT’s emissions.
We must keep in mind that although AI is a big deal, conventional IT isn’t shrinking, and we cannot afford to ignore it if we want to meaningfully reduce IT’s emissions.
The biggest energy wins are already within reachIf we are to successfully manage IT’s energy demands, we need to redirect some of the attention currently focused on AI’s projected energy needs toward the efficiency of traditional IT. We already know how to improve efficiency in traditional data centers: IT sustainability experts have proven tactics across data, software, infrastructure, and facilities already in place that could yield industry-wide improvements if adopted at scale. Enterprises can cut energy use by rationalizing data storage strategies, modernizing or retiring inefficient applications, upgrading aging hardware to more efficient systems, and improving workload placement to raise utilization rates. On the physical side, better airflow management, wider adoption of direct liquid cooling, as well as heat recovery and redistribution technologies, are readily available across both AI and mainstream IT and are increasingly accessible across geographies.
Bringing in mainstream IT into the AI energy conversation is not about choosing one over the other. We can walk and chew gum at the same time. Taking a whole-of-IT view, the industry can accommodate the AI revolution while ensuring that the entire digital ecosystem becomes more efficient. Thankfully, most efficiency techniques are applicable across both AI and mainstream compute.
IT efficiency is more than just reducing energy consumption, it’s a far-reaching business strategy that can mean the difference between success and failure. Organizations that master it secure better cost control, swifter deployment velocity, and the ability to effectively scale digital services. Indeed, the next era of IT sustainability leadership will belong to those who broaden their focus to address the big picture; optimizing all their computing, not just what dominates the headlines.
Find Out More: