Data Centers with Jonathan Koomey - #27

In this episode I talk with Dr. Jonathan Koomey, the founder of Koomey Analytics and a former researcher & scientist at Lawrence Berkeley National Laboratory. The focus of this episode is on data centers.We first start with Jon’s interest in the history of science which he studied at Harvard for his undergraduate. We go over Jon’s PhD thesis on why cost-effective energy efficiency measures may not be adopted in new office buildings. Then we get to data centers. Jonathan discusses the difficulties in obtaining the energy usage data for data centers, the tradeoff between innovation & understanding energy use, limitations of AI in research, a phenomenological model named Koomey’s law that Jon developed, GPUs, software to develop more energy efficient AI models, energy efficient vs inefficient data centers, how companies decide where to build data centers & much more. I hope you enjoy! Koomey Analytics Website: www.koomey.com PAPERS:Jon’s PhD Thesis: https://www.researchgate.net/publication/268339858Estimating Bitcoin Electricity Use: A Beginner’s Guide: https://coincenter.org/estimating-bitcoin-electricity-use-a-beginners-guideImplications of Historical Trends in the ElectricalEfficiency of Computing: https://www.researchgate.net/publication/224128141Characteristics of low-carbon data centres:  https://www.nature.com/articles/nclimate1786Separating fact from fiction in data center electricity forecasts: A guide for regulators: https://gridlab.org/portfolio-item/data-center-load-forecast-reportElectricity Demand Growth & Data Centers: A Guide for the Perplexed: https://bipartisanpolicy.org/report/electricity-demand-growth-and-data-centers/ ARTICLES:Data center energy use: truth versus myth: https://www2.lbl.gov/Science-Articles/Archive/data-center-energy-myth.htmlKoomey’s Law: https://en.wikipedia.org/wiki/Koomey%27s_law2024 United States Data Center Energy Usage Report: https://escholarship.org/uc/item/32d6m0d1To better understand AI’s growing energy use, analysts need a data revolution: https://www.cell.com/joule/fulltext/S2542-4351(24)00347-7 BOOKS:The Design of Everyday Things by Don NormanReinventing the Bazaar by John McMillanLinked: The New Science of Networks by Albert-László Barabási TEXTBOOK:Turning Numbers into Knowledge: Mastering the Art of Problem Solving by Jonathan Koomey MUSIC INTRO/OUTRO:Music: massobeats - https://youtu.be/4uJKRDkwwXo VIDEO:China & AI Supply Chains: DeepSeek, Huawei, Transformers, Robotics, & More with TP Huang: https://youtu.be/NAXOP5gt_5I CONNECT: LinkedIn:https://www.linkedin.com/in/adrian-dolinay-frm-96a289106/ GitHub: https://github.com/ad17171717 X:https://twitter.com/DolinayG Odysee: https://odysee.com/@adriandolinay:0  PODCAST: Apple Podcasts:https://podcasts.apple.com/podcast/id1765996824Audible: https://www.audible.com/pd/B0DC73S9SNiHeart Radio: https://iheart.com/podcast/202676097/ Spotify: https://open.spotify.com/show/60dPNJbDPaPw7ru8g5btxV |-Video Chapters-|0:00 – Intro 1:39 – Jon’s interest in science and energy 2:37 – Energy issues in 2026 7:24 – Jon studying the History of Science at Harvard 10:19 – Jon’s research at UC Berkeley 11:38 – Jon’s PhD Thesis on energy efficiency in new office buildings 19:51 – Rated power vs. actual power draw for hardware 26:32 – Why is energy usage not well measured by tech manufacturers? 28:05 – Critical thinking & the limitations of AI 34:05 – Koomey’s Law 38:59 – CUDA the software behind Nvidia’s GPUs 40:53 – AI software development in China 43:50 – Efficient vs inefficient data centers 49:47 – The shift from On Prem to Cloud 51:39 – How do companies decide where to build data centers? 57:24 – Data centers in the AI age 1:02:19 – Will the US & China dominate in data center compute? 1:05:35 – Does the type of AI algorithm affect energy efficiency of datacenters? 1:08:54 – The issues with assumptions about AI 1:17:38 – Jon’s ideal data center policies 1:24:59 – Book recommendations 1:29:51 – Conclusion