Google recently shared some numbers about how much water and electricity its AI queries use. They said a single text question to their AI only uses about five drops of water and a quarter of a watt-hour of energy. But a new article from MIT’s Technology Review questions if these numbers tell the whole story. Experts say these figures might make AI look more eco-friendly and cheaper than it really is.

The article points out that Google’s report only looks at simple text queries. It doesn’t include more complex requests like images or videos. These often take a lot more energy. So, the numbers Google gave—just for one basic question—don’t show the total energy used by all their AI services. Plus, Google only shared the median value. That means half of the queries use less energy, but the other half need more. We don’t know how much more, or how the energy use varies across different kinds of questions.

Another problem is that Google didn’t share the total number of queries their AI handles each day. They mention that Gemini, their AI system, has 450 million active users each month, but that’s only a fraction of their total AI activity. For comparison, OpenAI reports 2.5 billion queries every day with ChatGPT. So, it’s hard to estimate the real environmental impact of Google’s AI tools, especially since they’re used for many things like web summaries, email drafting, and more.

This lack of clear data makes it tough for companies to understand what their future costs might look like. As AI use grows, so do energy and water demands. CIOs and IT leaders have to guess how much these costs could rise by 2026, taking into account new features and different AI tools. They may even think about bringing AI services back in-house, which would mean dealing directly with the energy and water needs of their own data centers. That’s a big change because it involves planning for power supplies, cooling, and infrastructure.

Matt Kimball, an analyst at Moor Insights & Strategy, says CIOs need to talk with their facilities teams now. Power isn’t just a line item on a budget; it’s a crucial resource. Understanding how much power and cooling is available can help manage costs and efficiency. Kimball also suggests that companies should look into updating their storage systems. Moving from older spinning disks to new, high-density flash drives can cut energy use and boost performance.

When it comes to GPUs used for AI, Kimball recommends choosing models that meet the company’s needs without overspending. For example, the RTX6000 PRO GPUs use about 40% less power than the high-end B300 models but still deliver enough performance for many applications. This kind of smart choice can make a difference for both costs and environmental impact.

Simon Ninan, at Hitachi Vantara, adds that as AI grows, traditional cooling methods won’t cut it anymore. Data centers are shifting from air cooling to liquid cooling systems, which are more effective for AI workloads. However, these solutions require big investments to meet environmental standards and reduce energy consumption.

In the end, the push for more AI use means companies have to rethink how they plan for energy and water needs. The numbers shared by Google are a start, but they don’t tell the full story. Businesses that want to stay ahead will need to dig deeper, ask tough questions, and maybe even change how they build and run their data centers. The goal is to balance AI growth with sustainability and cost efficiency.

Inspired by

• https://www.computerworld.com/article/4047859/googles-estimate-of-ai-resource-consumption-leaves-out-too-much.html

Sources

• gmpg.org
• technologyreview.com
• google.com
• pcworld.com
• technologyreview.com