A new report on Artificial Intelligence (AI) from Greenpeace, produced by Öko-Institut in Germany, says that, by 2030, carbon emissions from AI datacentres will be six times the 2023 level, and notes that ‘leading players in the field of AI are increasingly turning to nuclear energy to meet the growing electricity demands of their data centres.’
It says that Google explicitly defines nuclear as clean energy and has signed agreements to purchase electricity from small modular reactors (SMRs). Amazon similarly justifies its investments in nuclear and has signed three new agreements to support nuclear projects, including SMRs. Meta has, it says, gone a step further, expressing interest in developing its own nuclear power projects specifically to power its AI data centres. And in addition to its support for a revamp of old closed Three Mile Island plant, Microsoft has been promoting ‘advanced nuclear’ as a key part of its carbon-free electricity strategy. Greenpeace notes that Apple is the only company that has not publicly announced any plans to invest in or to use nuclear power.
Why nuclear? Superficially it might seem attractive. As the Greenpeace report says ‘large data centres that provide their services globally are usually working at high operational load around the clock’. And as non-flexible power plants, ‘nuclear plants have precisely the generation profile that they need for their current operation’. However, nuclear comes with big environmental disadvantages, safety issues, security risks and high economic costs. For example, far from getting cheaper with clever new tech, as some hope, Greenpeace says ‘nuclear power is expected to cost more than 120 $/MWh, at least twice as much as the energy price of 45 - 70 $/MWh that is expected for the combination of solar energy with battery storage in the USA according to the International Energy Agency. Instead of relying on nuclear energy, it would therefore make much more economic sense to invest in large battery systems located close to data centres and store renewable energy in these systems so that it is available around the clock’.
So, while it does admit that, given that renewable are variable, they are not well suited to direct support of AI and data centres, Greenpeace suggests that, with requisite energy storage, they can be a better bet for meeting demand, although that might require AI to ‘respond flexibly to the renewable electricity availability’. They say that would be better than the current situation, when fossil gas or coal plants are used to provide energy when there is no renewable electricity available in the power grid. Greenpeace says ‘at present, the additional consumption caused by the expansion of AI data centres is… essentially generated by running climate-damaging fossil fuel power plants over longer periods of time’.
AI may help improve energy and resource use efficiency and sustainability, and Greenpeace says it is not against AI/data centres as such- as long as they deal with their direct local eco-impacts (including water use), and are not used for environmentally damaging purposes (like aiding fossil fuel extraction – or for justifying nuclear). But it wants an ‘energy-efficient AI infrastructure powered 100% by renewable energy’. It says ‘AI developers must take responsibility for their supply chains. They must contribute to the expansion of renewable energy in line with their growth and ensure that local communities do not suffer negative consequences (e.g., lack of drinking water, higher electricity prices) .
However, while the Greenpeace report calls for the commensurate renewable energy capacity to be added to national grids at the same time as new datacentres are built, signs are that developers may not always be able to take advantage of local renewable resources. For example, according to the Guardian, the developer of Elsham Tech Park Ltd in Lincolnshire, where a huge new £10bn data centre is planned, ‘has ruled out on-site renewables as impractical’, arguing that ‘if the system ran on biomass energy it would require the daily delivery of 100 large lorry loads of wood chips. Wind energy would require 10,000 20-metre wind turbines, while an area five times the size of the Glastonbury festival site would be needed if it were to be powered by photovoltaic panels’.
That assessment may be wrong, but according to the developer, it was not a problem since the proposed site was in ‘an ideal location for AI datacentres, due to the significant investment already made in developing the UK’s most advanced clean energy cluster. It has access to a third of the UK’s offshore wind energy as well as access to 66% of the UK’s licensed carbon capture and storage.’ OK, but would AI be the best use of this valuable imported green energy? Or indeed locally generated green power. Even assuming that you think this AI project is really needed…especially given that it is projected to release five times the carbon dioxide of Birmingham airport, including from take-offs and landings, assuming it uses the current grid energy mix. And, also assuming you don’t want to see AI projects like this sometimes being used to justify new nuclear….
That is a real risk, even if SMRs are some way off- probably not, at best, until the early or mid-2030s at any significant scale. But that’s not the only problem. Greenpeace is also worried that ‘within just five years, AI is expected to dominate overall computing demand’, and, unless renewables are mandated, even if some new nuclear gets started up, ‘the additional electricity required will prolong the operation of fossil fuel power plants, putting climate targets at risk’.
A key point is that nuclear is not only costly, but also slow to deploy, while deflecting funding from renewables- which are cheaper and faster to install, and, arguably, in general, a better use of scarce resources. It’s the same old energy choice battle, in a new (AI) context, with the UK government clearly now sold on nuclear- as witness the £14.2 bn extra it has just allocated to the proposed Sizewell C reactor, ahead of the spending review and then £2.5 bn for SMRs- see my last post. A bit surprising given the UK’s financial situation, and the strength of the case against nuclear - see my next post. But AI? It got £2 bn more.
Comments
Post a Comment