How data centres will cut carbon emissions, not increase them

Yes, you read that headline the right way. Despite the torrents of nonsense you will hear from lots of places (and a lot of energy nonsense these days seems to be concentrated in the Daily Telegraph) data centres and that fellow monster AI is unlikely to bust the electricity grid. Indeed, it is, on balance, more likely to reduce total energy demand.

Nevertheless, lurid tales of the data centre/AI monster devouring power have been spreading, and, they have even been linked to companies such as Meta, Amazon and Amazon issuing press releases about how they are going to source power from another monster, new nuclear power plants. However, as in horror stories in general, these monsters are, at least in almost all cases, entirely mythical.

As a point of fact, despite the seemingly rapidly rising number of data centres in the USA electricity demand in the first half of 2024 was the same as it was two years previously. The UK’s electricity demand has been on a downward path for some time. The International Energy Agency has recently published an analysis saying that data centres will be a small proportion of future growth energy demand (see HERE). Now, there is a very strong case for saying that electricity demand will increase to a greater or lesser extent. However, it rests on increased electrification of transport (starting with Electric Vehicles) and also heat pumps. But I don’t suppose many Daily Telegraph readers are very interested in that.

In the USA lots of data centres are being built in Texas. In the UK London is the biggest hub, housing over 40 per cent of the data centres sited in the UK. Not many people seem to realise that they will not consume gigantic amount of power. Even fewer people know that the alleged super-monstrous AI (which is said to ravenously consume this data centre energy) may, in fact, actually reduce energy demand in various indirect ways.

In fact AI and data centres that feed AI are likely to generate economic growth in service industries - business applications - that have a low energy intensity. GDP is likely therefore to become less energy intensive, replacing output that would otherwise be more energy intensive. Moreover the energy usage of these less energy- intensive applications will be itself reduced through better energy management organised through AI. Goldman-Sachs (see here) estimates that AI will see ‘the economy benefiting, from enhancing office productivity and sales efforts, to the design of buildings and manufactured parts, to improving patient diagnosis in healthcare settings,’ etc This means that as the efficient use of energy in the economy increases faster than the demand for energy services, then energy usage will decline, not increase. AI processing centres will themselves reduce the power they use as they take advantage of new tools that are becoming available to reduce power consumption. See HERE.

One way in which energy efficiency is going to be increased is directly as a result of the application of AI itself to improve the way that energy is managed, especially in buildings. Andrew Warren, the Chair of the British Energy Efficient Federation (BEEF), says in the October issue of the Energy In Buildings Newsletter ‘According to management consultants McKinsey, AI has the potential to deliver energy savings of up to 20% in buildings and 15% in transport systems. Additionally, AI-driven solutions can help businesses reduce CO₂ emissions by up to 10% and cut energy costs by 10-20% – particularly so in the field of industrial energy management.’

The overall effect of AI-driven increases in data centre consumption is visualised using the chart below. The demand for energy servicesA (green line) represents increasing energy demand under conditions of earlier industrial periods, whilst the demand for energy servicesB (orange line) represents the demand for energy services under a AI influenced services dominated economy. As can be seen also the green line represents the trend towards energy efficiency improvements that occurs as the economy refines its equipment, buildings, plant etc to use energy more efficiently. The orange line represents the increase for energy services in a AI influenced services economy. The key point is that it grows at a slower rate than overall energy efficiency improvements. This leads to a decline in energy consumption overall.

Of course this chart assumes that the energy economy is structured as before in its energy supply technologies. Of course energy will be increasingly supplied by electricity technologies. This chart ignores this effect, but is constructed so that we can focus on the effects of AI and data centres.

On top of all of this there are opportunities for data centres to sell otherwise waste heat to neighbouring businesses and housing estates in district heating systems. Indeed, it was reported last year that ‘The UK Government has awarded £36 million ($44.5m) to a district heating system in West London which will share data center waste heat with up to 10,000 new homes’. This development is taking place in Ealing.

We should make this the norm rather than the exception with data centres. We should take a leaf from German practice. Under the German Energy Efficiency Act (approved in 2023) data centres are required to re-use an increasing amount of waste heat - at least 20 per cent for those data centres which come into operation from 2028. It should be noted that low carbon waste heat from electricity is increasingly renewable in source and this will substitute for natural gas consumption. As a result, carbon emmisions will be lowered. Hence not only can data centres reduce energy consumption overall by the indirect effects described above, it can also reduce them directly by substituting the low carbon waste heat for natural gas consumption that is normally used for heating

The reaction in some circles to the (it seems) mythical monster of the threatening data centres and AI becomes all the more stranger when it comes to nuclear power’s role in the story. Mythical nuclear power stations (that is ones that are unlikely to be built) are hypothesised as the answer to this (non-) problem. US companies such as Google, Amazon and Meta have been sounding their support for new nuclear power. In fact their support seems to consist of usually of vague announcements of future support. You would think of course, that if more electricity is needed, then we need to issue more contracts for renewable energy. This is especially as nuclear power plant seem very difficult to organise. However, such thoughts do not feature strongly in the minds of many Daily Telegraph readers.

The only definite support new nuclear power is actually getting in the wake of the imaginary AI monster seems to come from Amazon who has pledged $500 million towards a 300 MW small modular reactor (maybe the sort of offering that can please Trump supporters). Quite how a smaller type of nuclear reactor is going to deliver much of an increase in power supplies compared to the booming construction of wind and solar plant does not seem to compute with the Trumpists either. But it falls in with the ever-mythical notion that cheap nuclear power is (again) just around the corner story that we have heard for so many decades. What’s changed now? Really, not very much in fact - but the mixture of monster AI power demands and nuclear power coming to the rescue is really like something out of a 1950s science fiction story. And that’s all it is likely to remain.