The Impact of Artificial Intelligence on Power Consumption in Data Centers
I know artificial intelligence (AI) has not been widely accepted and many want it ban altogether, but that’s not going to happen. AI is here to stay, and we really do need it in today’s data intensive world. It’s estimated that over 400 million terabytes of data are produced every day globally, and there’s no way to extract useful information from mega-data without AI.
It’s a good bet that smart grid technology is contributing terabytes to that total daily, and it’s being stored and processed along with the rest of it. That requires data centers and sophisticate computers, which affect our power grid. What got me thinking about this was our new Energy Secretary, Chris Wright, coming here to New Mexico this week.
He’s on a nationwide tour of the national laboratories and his first stop was at New Mexico’s Los Alamos and Sandia National Labs. Wright shared his excitement with the work going on at Sandia and Los Alamos. These labs are the leaders in the Nation’s AI research and development efforts. He mentioned that Los Alamos had announced a partnership with OpenAI, which will be a key part for future AI development.
AI Driven Power Consumption
The Secretary also spoke about the Nation’s growing need for more electricity due to the increased demand that AI and data centers are driving. It got me pondering some other press releases I had run into recently. They highlighted recent reports and studies from research groups and government agencies investigating the power consumption of datacenters and AI tools.
There was one from IEA (International Energy Agency) noting that the average data-center power requirement is 5-10 megawatts. And hyperscale data centers can consume 100 megawatts or more with numbers growing constantly. In 2024, data centers accounted for around 1% of the global electricity consumption according to IEA. If you look at the U.S., China, and the European Union, they each use about 2-4% of their grid’s capacity for data centers. This demand growth is expected to grow to around 10% or more by 2030.
It's interesting that we talk in terms of terabytes and megawatts, which are great for giving the scope of an issue. But from the perspective of quantities describing things we use every day, they’re not too meaningful. That’s what caught my attention with a news item from Goldman Sachs. It addressed the amount of power consumed by AI-enhanced web inquiries. The Goldman Sachs commentary said that the average AI-enhanced internet search consumes up to 10 times more energy than a non-AI inquiry does.
Doing Routine Stuff With AI
Google’s “AI Overviews” uses an enhanced generative-AI feature that smartens up searches. Google says it’s an evolving technology that is improving continuously, but it can make mistakes. I did notice a few questionable topics, but for the most part, the responses were focused on my specifics. Overall, the AI-enhanced assisted queries saved me a lot of time, but I hadn’t thought about the power those simple searches used until I found the Goldman Sachs observations.
It could be argued that with more efficient AI-enhanced queries, it requires less searches to get the information we need. On the average, my traditional searches produced dozens of pages of responses, each requiring some amount of reviewing. More often than not, I would end up rewording the query and doing several more searches.
By reducing this process, and finding the material with only one or two inquiries, it would be a bargain even with the ten times energy cost added. Even with old-school searches costing one-tenth of an AI search, it may actually save energy in the long run. With that in mind, as I wrote this editorial, I concentrated on utilizing Google’s “AI Overviews” feature for my web searches.
I found my time spent in research was more productive than it had been without AI. In most cases the lead items on the first response page usually gave me the information I needed without more searching. Granted, this isn’t hard empirical data, but it’s a tangible examples of how AI improves one searcher’s efficiency.
It’s estimated that the internet’s database was about 147 zettabytes at the end of 2024 and growing, so we need help. Someone looking for specialized information is much like a utility searching its multiple databases for specific information and saving time is saving money. I have a much better appreciation of how AI’s cost/benefit ratio applies to ferreting out meaningful intelligence. I hope it helps you too!
