In a recent Google search, I found that we collectively run 417 billion Google searches a month worldwide. (Sometimes the Gemini search results are returned as well, which is the AI search result, at the top of the page.)
Whilst the 417bn number may be either shocking, or understandable, depending on your point of view, I thought I’d do some rough calculations.
At 0.0003 kWh per search, this equates to 125,100,000 kwh per month, which is like running 125,100 houses, the size of an average European City. (This kWh figure is from Google themselves, so take this as a conservative estimate.)
Of course, these figures are an average, or an educated approximation, but the fact still stands that we are very disconnected to the energy usage and understanding of the physical infrastructure that powers our digital lives. There are neither prompts, nor information to make us consider the carbon emissions of our app usage or phone calls for example – and yet all of it goes through our Critical National Infrastructure.
When we buy food in super-markets, there are calorie counts and nutritional information on the labels, so that we have the details needed to make the choices that are ‘right’ for us.
At the recent DataCloud Energy and ESG Europe 2025 event in Brussels, Mark Acton said:
“Surely, we should have the information to allow us to make informed choices about the environmental and energy consumption impact of the digital services that we choose to consume. The added benefit is that this would also be helpful for the general public to understand that the power demand for data centres is actually driven by consumption of these digital services.”
Now with the advent of generative AI searches such as Gemini, Copilot or ChatGPT, we are finding that these searches are consuming up to 20 times the amount of energy of a normal Google search. Additionally, when Google adds Gemini results to your search we are now running these two searches at the same time – even if we don’t want to run these searches, they are being run for us anyway. At the moment, the results just appear in various places, as a kind of product enhancement. At the same time, we are seeing products for sale online that are being promoted using ChatGPT, which adds to this unnecessary energy consumption demand.
There are a plethora of predictions that will ‘predict’ how many GWs of energy AI will need in the coming years, not all of it will come to fruition, but at the moment, the only inhibitors to this are very practical things like power infrastructure availability, or supply chain delays. There are no rules or regulations to stop any one of us from building an LLM (Large Language Model) data centre, given enough funding. These are the big data centres that we’re hearing about such as the 320MW DC in Hertfordshire.
We’ve seen, and welcome, the EU AI Act last year, which has real fines for high-risk AI applications if they don’t comply with required practices. Apart from that, I’ve seen nothing, but would be happy to be corrected or informed on any good work going on here.
The lack of regulation and the disconnect between our daily digital consumption, and the impact to the planet, means we’re effectively allowing a Wild West situation when it comes to developing AI applications.
I’m sorry to dampen the mood in a booming sector, but we really need to think about what we’re using AI for, instead of blindly allowing the carbon emissions for frivolous uses of AI.
This is a recent example: Loona take a look and see if you think this is the right use of our planet’s resources.
It’s all very well moaning about it, but what can we do? Here are the two areas that might help.
1) Embrace incoming regulation around Ethical and Responsible AI. Get involved where you can and have your voice heard.
2) I’d love to see an innovative software company come up with some new metrics around human productivity vs carbon emissions, or their company’s net impact on the planet. This would be a great challenge to the big cloud service providers.
#datacenter #CNI #frivolousAI #ethicalAI #responsibleAI