Open AI Wants to Build Data Centers That Would Consume More Electricity Per Year Than the Whole of the U.K.
Over the past few months, the newswires have been hot with stories about the large-scale data centers that will be required to meet the needs of the forthcoming revolution in Artificial Intelligence (AI). How much electricity will these new data centers consume and what does that mean for the electricity demand forecasts underpinning the plans for Net Zero?
Recent Date Centre Announcements
To give a flavor of the scale of data center developments that are coming, it is helpful to look at recent announcements from large tech companies. Back in March, it was announced that Amazon had bought a 960MW data center that is powered by an adjacent nuclear power station. In April, Mark Zuckerberg CEO of Meta that owns Facebook and Instagram said energy requirements may hold back the build-out of AI data centers. He also talked about building data centers that would consume 1GW of power.
Last month, Oracle chairman Larry Ellison announced that Oracle was designing a data center that would consume more than 1GW that would be powered by three small modular nuclear reactors (SMRs). Then Microsoft also got in on the act when it announced it had done a deal with U.S. utility Constellation to restart the 835MW Three Mile Island (TMI) Unit 1 nuclear power plant to power its data centers. Anxious not to be left out, Sundar Pichai, CEO of Google said they too were working on 1GW data centers and saw money being invested in SMRs.
Scale of AI Energy Demand
When companies bandy about such large numbers it is sometimes difficult to visualize just how big they are. For context, consider that a 1GW data center would consume 8.76TWh of electricity each year. Seven of Altman’s enormous 5GW data centers would consume 306.6TWh. According to DUKES data (Table 5.6) the UK generated 292.6TWh in 2023. The plans for ChatGPT alone would consume more electricity in a year than the U.K., the sixth largest economy in the world, managed to generate. Now consider what the total demand is going to be when you add in the requirements the likes of Amazon, Oracle, Microsoft, Meta, Google, and X.
Net Zero Electricity Plans
Clearly, AI energy demand is going to be huge and if we want to compete in this new industry, we are going to need cheap, reliable, and abundant energy. However, the plans for U.K. electricity use in 2050 are tiny by comparison. The Royal Society assumed 570TWh of annual demand in its report on long-term storage. In their latest FES report, the NG ESO assumed a total 615-719TWh of demand across Industrial & Commercial, Residential, and Transport sectors in their pathways that achieve Net Zero by 2050.
The RS report relies solely on wind and solar renewables plus hydrogen storage, whereas the FES report supplements renewables with BECCS, gas with carbon capture, together with some nuclear and hydrogen. It is easy to see how the AI revolution could consume an extremely high proportion, or even all the electricity that is being planned for 2050. It is difficult to see how there is going to be enough electricity to go around.
