AI, the demand for data centre power and the move into nuclear
One of the significant shifts coming from artificial intelligence (AI) is associated with the incredible demand for data centre power. When analysing rack power density, or the power draw of a single, fully populated rack measured in kilowatts (KW), a typical centre without AI demand has a rack power density of 6-12 kilowatts from central processing unit (CPU) based servers.
However, a data centre supporting AI in the training phase – the process that enables AI models to make accurate inferences – has a rack power density of 26-80 kilowatts, up 4 to 7.5 times, from high-powered graphics processing unit (GPU) based servers.
A data centre supporting AI in the inference phase – the process that enables AI models to produce predictions or conclusions – has a rack power density of 12-40 kilowatts, up 2 to 3.5 times, from a combination of CPU and GPU-based servers.
In terms of U.S. data centre electricity demand, Goldman Sachs is forecasting 15 per cent compound annual growth to 2030, driving data centres to account for eight per cent of total electricity demand by 2030, up from the current three per cent.
And in this vein, Amazon Web Services (NASDAQ:AMZN), Microsoft (NASDAQ:MSFT) and now Google are proposing securing energy for some of their data centres, which will be powered by nuclear energy. The big three will be heralding their “clean growth” credentials, whilst delivering on AI.
Amazon Web Services recently purchased Talen Energy’s 1,200 acre data centre campus, which adjoins its 2.5 gigawatt nuclear power plant, in Pennsylvania.
Subject to regulatory approval, Microsoft has signed a 20-year power supply deal in September with Constellation Energy to supply 835 megawatts of clean energy from the Three Mile Island Unit 1 nuclear plant, also in Pennsylvania.
And earlier this week, Google announced plans to purchase energy from Kairos Power’s small modular reactors (SMRs). Currently, there are only three small modular reactors operational – in Russia, China and India, however, there are over 50 under development, and most of them have capacity of 100 to 300 megawatts. Google expects the first Kairos Power U.S.-based reactor to be on online by 2030, adding 500 megawatts of power by 2035.
After hitting an 18-year high at US$106/lb. in February 2024, the uranium price fell 26 per cent to US$78/lb., however, it has traded up in recent weeks as Russia, which accounts for 13 per cent of global uranium concentrate, 26 per cent of conversion (known as uranium hexafluoride) and 38 per cent of enrichment, threatens an embargo on exports to the West.
Chief Executive Officer of Montgomery Investment Management, David Buckland has over 30 years of industry experience. David is a deeply knowledgeable and highly experienced financial services executive. Prior to joining Montgomery in 2012, David was CEO and Executive Director of Hunter Hall for 11 years, as well as a Director at JP Morgan in Sydney and London for eight years.
This post was contributed by a representative of Montgomery Investment Management Pty Limited (AFSL No. 354564). The principal purpose of this post is to provide factual information and not provide financial product advice. Additionally, the information provided is not intended to provide any recommendation or opinion about any financial product. Any commentary and statements of opinion however may contain general advice only that is prepared without taking into account your personal objectives, financial circumstances or needs. Because of this, before acting on any of the information provided, you should always consider its appropriateness in light of your personal objectives, financial circumstances and needs and should consider seeking independent advice from a financial advisor if necessary before making any decisions. This post specifically excludes personal advice.
Enrico
:
I think its a specialised industry requiring its own specialised energy supply solution. Small modular reactors (SMRs) excel in providing reliable, carbon-free baseload power, especially in regions where renewables struggle due to limited sunlight, wind and space (not Australia). They are particularly useful for replacing fossil fuels for localised industries that require consistent high energy levels, such as AI data bases, while solar and wind can be deployed in various scales making them ideal for remote or highly distributed energy needs such as in Australia. You could make a case for solar and wind to combine with SMRs providing electricity during peak demand periods in densely populated or highly industrialised areas (industry mainly operates in the day, wind and storage is more consistent) reducing the need for a high reliance on expensive nuclear power. If you are over reliant on SRM’s you are also highly reliant on a supply of specialty fuel such as enriched uranium. In appropriate regions, solar and wind are a significantly cheaper source of electricity on a per-kilowatt-hour basis, even when considering storage systems to mitigate intermittency. The relatively recent shift in AI data base requirements and their ever-increasing power needs seems to indicate that grids that will have sizable data bases feeding off them have no choice but to introduce or increase nuclear power capacity given current technology. Ideally domestic users will not be required to share all the addition cost of providing nuclear energy that was driven by a particular industry’s needs. Akin to the article, it should be a commercial decision if a company wants to acquire its own SMR for its own use.
Roger Montgomery
:
You have a lot of people who are in that camp and agree with you.