Everyone Suddenly Cares About Water Use

Apparently we’re all deeply concerned about water waste and carbon emissions now. Excellent.

Two planned AI datacentres in England are predicted to produce more carbon emissions than ExxonMobil’s entire UK operations. Analysis by Foxglove estimates that Wapseys Wood in Buckinghamshire and Quest Park in Bedfordshire could emit 4.577 million tonnes of carbon a year when fully operational. ExxonMobil emitted 3.9 million tonnes in 2023.

The two sites would require 1.3GW of electricity between them and, because of delays connecting to the National Grid, are seeking to build their own gas-fired power stations!

They are not alone. There are currently 315 datacentres waiting to connect to the UK electricity grid, representing 73GW of demand. That is almost twice the UK’s peak winter electricity demand.

Then there is water.

England could face a public water supply shortfall of five billion litres every day by 2055, according to the Environment Agency. Another billion litres a day could be required by energy production, food production and emerging technologies.

Water UK says government forecasts for future water demand explicitly exclude datacentres, despite plans to treble UK datacentre capacity by 2030.

Current UK datacentres use around 6.6 million litres of water every day. Trebling capacity could increase that to 19.8 million litres. During hot weather, when everyone else also needs more water, demand could become much higher. Water UK estimates peak demand could reach 42 million litres per day. Under a worst-case scenario involving particularly inefficient new facilities, it could reach 77 million litres.

AI has a significant environmental footprint, but can we now have the slightly more uncomfortable conversation?

Suddenly everyone cares about water

Viral posts about AI water consumption have become remarkably popular. And apparently we are supposed to gasp at every milliliter of water consumed by Claude while completly ignoring the cow in the room.

Bryant Research estimates ChatGPT uses around 18.2 billion litres of water.

Dairy production uses approximately 4,555 billion litres. Nearly 250 times as much. And that is dairy alone.

Animal agriculture accounts for more than a quarter of humanity’s water use. Producing one kilogram of beef requires around 15,400 litres. Beef production globally uses approximately 2.49 trillion litres every day. Trillion. Because numbers that large become meaningless, consider this.

One billion seconds is about 31.7 years.

One trillion seconds is about 31,709 years.

That is the difference in scale we are discussing.

Apparently deleting old emails is now suggested as one way ordinary people can help reduce pressure on datacentre servers. Meanwhile we continue breeding billions of animals into existence, growing enormous quantities of crops to feed them, providing drinking water, cleaning facilities and producing vast quantities of milk and flesh.

This isn’t an argument for ignoring AI

AI absolutely deserves scrutiny.

Datacentres should disclose their water use. Building them in already water-stressed regions should be halted. Recycled and non-potable water should be used wherever possible. Closed-loop systems and more efficient cooling should be encouraged. And perhaps the government should know what enormous pieces of supposedly “critical national infrastructure” are actually being used for before allowing them to bypass local planning scrutiny.

Some uses will be valuable.

AI can support scientific research, improve productivity and potentially help replace animal experiments. Datacentres also process essential health, financial and defence information. Environmental cost therefore has to be considered alongside necessity, benefit and alternatives. That is where comparisons with animal agriculture become particularly awkward.

What exactly are we getting in return?

Animal agriculture is not merely thirsty.

Its environmental destruction stretches across climate change, deforestation, land use, biodiversity loss and water pollution. One peer-reviewed paper attributed 53% of global average temperature rise between 1750 and 2020 to animal agriculture, compared with approximately 19% to fossil fuels. And this is happening while Britain increasingly experiences worsening drought and extreme heat.

Then add zoonotic disease risks, antibiotic resistance, worker trauma and the health consequences associated with consuming animal products. And we still haven’t mentioned the direct victims. The individuals imprisoned, mutilated, separated from their families and killed because somebody wants a particular flavour or texture.

AI’s environmental impact has to be weighed against what AI provides and whether the same benefits could be achieved more efficiently. Apply exactly the same test to animal agriculture.

We can produce food without breeding and killing animals. Plant-based alternatives already exist. Bryant Research reports that replacing animal flesh with plant-based alternatives could reduce associated water use by up to 99%.

So we have one technology with serious environmental problems but potential social benefits, if regulated quickly, and significant opportunities to become more efficient. Then we have an industry consuming staggering quantities of water and land, driving environmental destruction and killing billions of sentient individuals for products we do not need.

Guess which one keeps becoming the viral environmental emergency.

Nuance is often the first casualty of a moral panic. We saw much the same thing with ultra-processed foods. Find something new and frightening, remove all context, then compete to produce the most terrifying statistic.

Scrutinise AI.

Regulate datacentres.

Demand efficiency.

Question whether burning gas and draining water supplies to generate another chatbot is remotely sensible. But if your concern is genuinely water scarcity or climate change, rather than simply joining the latest online panic, you cannot keep pretending animal agriculture is invisible.

At this point the temptation to recycle every viral anti-AI environmental post and simply change the nouns is becoming almost unbearable.