Increasingly, headlines are warning that AI will drain the power grid. But the reality of data center energy use is far less sensational — and far more efficient.
Surging demand for computing power is real, but data centers across the country are implementing efficiency measures to keep pace. From AI models that fine-tune cooling systems to “follow-the-sun” load shifting and AI programs that maximize energy usage in chips, Google’s data center team, in particular, is taking the lead.
Engineering fellow and Google VP Partha Ranganathan has spent years pioneering cutting-edge energy strategies and systems architecture for data centers. For him, the mission isn’t just making data centers more efficient. It’s setting the record straight on their energy and water usage.
“There are legitimate concerns that we need to be very responsible about, thinking about energy efficiency for AI. But honestly, some of the claims are sensationalized a little too much in my mind,” Ranganathan said in an interview for season five of the Where the Internet Lives podcast. “We are still in the very early stages of how we are thinking about energy efficiency for AI systems, and we’ve been making some tremendous advances.”
These advances – including hardware, software, and strategy improvements – center around one goal: accommodating AI workloads while minimizing resource consumption.
On the hardware front, Google has deployed custom accelerators like Tensor Processing Units (TPUs) and Video Coding Units (VCUs) tailored for specific workloads to improve efficiency. To manage the heat density of these new high-performance servers, and, where appropriate, it is installing liquid cooling systems, which are significantly more energy efficient than traditional air cooling.
Google is also leveraging AI to design its own chips, slashing the design lifecycle from years to months in an effort to streamline the development of efficient hardware. “Every aspect of chip design can benefit from using AI to accelerate and amplify the process,” Ranganathan said.
This “virtuous cycle,” as Ranganathan described it, of AI supporting AI, lends itself to energy-efficiency gains in the software itself. Behind the scenes, that means using AI to optimize data center operations like fine-tuned cooling. It also means boosting the energy efficiency of AI search summaries by a factor of ten in just one year.
Ranganathan’s approach to efficiency extends to how an entire “warehouse-scale” computer is managed and conceptualized. He argues that innovation comes from designing the system holistically—from the building itself to the silicon, software, and AI models. He explained that Google has also embraced a methodology where compute workloads are moved geographically throughout the day to match the availability of carbon-free energy. If the sun is shining or the wind is blowing in a specific location, workloads shift to use that clean energy.
Ranganathan emphasized that these efficiency advances lead to more advanced AI, something that benefits everyone.
“ Compute is the oxygen that powers the AI revolution. Data centers are basically where all the magic happens,” he said. “ I cannot think of a time in my career where we have had the amazing opportunities we do today to be in the foundation of enabling a new revolution – the intelligence revolution – with all the associated benefits that we have.”
For the full conversation with Partha Ranganathan, listen to his interview on season 5 of Where the Internet Lives.
This is partner content, brought to you by Google. It borrows from an interview that appeared on Where the Internet Lives, a podcast produced in partnership by Latitude Studios and Google.
Where the Internet Lives is an award-winning podcast about the unseen world of data centers. Follow on Apple, Spotify, or wherever you get your podcasts.
