Imagine this: You ask ChatGPT for a recipe, create a picture with Midjourney or let Claude help you with an idea. Fun and useful, right? But in the background, thousands of servers are slurping up electricity like crazy – enough to provide light and heat for large European cities. Is AI the new climate disaster threatening the planet, or can it actually save us from climate chaos?
First, we’re in the positive corner, looking at all the positive aspects of AI. AI could become the ultimate climate warrior, join WWF and promote a green future. Think of AI optimizing wind turbines and solar panels, predicting extreme weather with pinpoint precision, monitoring deforestation from satellites or designing more efficient batteries. Studies show that AI could cut global emissions by billions of tons – if we use it smartly!
On the savannas of Africa, predators are tracked with AI-powered cameras that monitor the animals’ movements and health. It also identifies poachers and helps local rangers protect the animals.
Something similar is happening in the Himalayas, where scientists are monitoring the Snow Leopard with AI-powered cameras. This is the most endangered cat species. Before the cameras researchers had to walk for weeks and months before they could spot the cats. Now AI-powered cameras provide information on all the movements and health of these beautiful animals.
Now to the more realistic corner and looking at the climate footprint that AI has.
AI consumes:
- Electricity
- Water
- Minerals
- Area
AI gives back:
- Electronic waste
- CO2 emissions
Power:
There have been several posts here in Norway, and internationally, about crazy power consumption. Even our Prime MinisterJonas G Støre has spoken out on several occasions. In April 2024, researchers from NTNU, led by Jonas Nøland, predicted an infinite power demand and pressure on power supplies. They point out that each Navida Blackwell GPU has a power requirement of up to 14.3 KW. When we know how many Navida sold in 2024 and how much they have announced of deliveries in the period 2027-2030, the figures are as follows:
• By 2026, AI will require as much power as the whole of Sweden.
• By 2030, AI will require as much power as the whole of Great Britain. (figures from IAE)
• In 2030, data centers will consume as much power as the country of Japan, about 3 times as much power as Great Britain.
Internationally, all the big technology giants admit that nuclear power must be in place before there is enough electricity for all the AI plans. In the US, Microsoft has reopened the Three Mile Island nuclear power plant that was originally closed in 2019. This plant will now be reopened in 2028 to provide enough power for Microsoft’s AI efforts. Both Google and Amazon confirm the same. They are working on small modular nuclear power plants (SMRs) to get clean and stable power for AI.
But it takes time to get nuclear power in place (even in the US) and in the meantime these players are making do with what they get in terms of energy. In the US, they are postponing shutting down coal-fired power plants to feed AI with energy. President Trump has also announced that the coal industry will be the guarantee of getting enough electricity for AI. If you don’t know how harmful coal-fired power plants are, you can read about it here.
In the US, Oracle and OpenAI are collaborating to build several of the world’s largest AI data centres. These are called Stargate, but it is uncertain whether they are related to our new proud AI centre in Narvik. The new data centres for Oracle and OpenAI will use gas power plants. It is not as polluting as coal power, but by no means clean, or renewable energy.
With coal and gas power, AI is in no way sustainable. We hope SMR or other renewable sources will come soon, so that this can become sustainable in the long term.
Water:
AI requires a lot of water in production, and for cooling.
• Researchers from the University of California released an alarming report on water consumption as early as 2023. Already in 2027 (only 1 year to go) water consumption is estimated to be 4-6 times the annual water consumption of Denmark. It is especially generative AI that requires a lot of water.
• There are many different predictions about what water consumption will be in 2030, the most aggressive comes from Fabrizio Palermo, CEO of ACEA – Europe’s second largest water utility. He says that water consumption could reach the entire US water consumption – every year.
• Let’s hope he’s wrong, that it might only be half the US’s annual consumption.
• But there’s no doubt that AI is thirsty!
We know that closed systems for cooling are now being introduced, meaning that the water is used several times and the total consumption is to be reduced. We also know that work is being done on liquid cooling that uses little water. The big Tech giants have promised to be climate neutral and water neutral by 2030. Let’s hope that’s true, even though most have already admitted that they won’t be able to do it – Because of AI.
Minerals:
Navida’s servers are full of minerals from slave labor and conflict zones. The minerals we are focusing on in this article are:
• Cobalt
• Lithium
• Coltan
• Gallium
• Copper
• Tungsten
• Germanium
Several of these minerals (Cobalt and Tungsten) are mined in conflict zones and are a disaster for both the environment and human rights. Those who want to learn more about Cobalt can read Siddhart Kara’s award-winning book “Cobalt Red – how the blood of Congo is powering our lives“, or see the feature I have also written about modern slavery and the environment
By 2030, cumulative consumption of conflict minerals from AI will have risen to between 5 tons and 20 tons
Copper:
Copper is one of the worst minerals in terms of mining and production of toxic waste. Those who want to see details can read about the new copper mine in Finnmark. It will fill 17 trucks per hour with toxic waste. This waste will be dumped directly into the Repperfjorden, destroying all life in the fjord. All over the world there are such mines that together generate large mountains of toxic waste, and will destroy forests, wildlife, rivers and soil. One of the biggest drivers of demand for copper is AI data centres. According to Carbon Credits.com, AI data centres will consume 400,000 tons of copper on average, each year over the next 10 years. Cumulatively, this will correspond to 4.3 million tons in 2035. If we look at the mine in Finnmark, this will result in toxic waste equivalent to 50-180 times that volume. Luckily this toxic waste is dumped in Myanmar, Kongo, Amazonas, Indonesia etc. And also, now in Finnmark, in Northern Norway.
Electronic waste:
Due to their enormous power consumption, AI servers quickly become electronic waste. Researchers from several universities have released similar reports, but the one from these people gives a good picture of the madness:
Peng Wang, Institute of Urban Environment, Chinese Academy of Sciences https://orcid.org/0000-0001-7170-1494
Ling-Yu Zhang, Institute of Urban Environment, Chinese Academy of Sciences
Asaf Tzachor, Reichman University https://orcid.org/0000-0002-4032-4996
Eric Masanet , University of California, Santa Barbara
Wei-Qiang Chen, Institute of Urban Environment, Chinese Academy of Sciences https://orcid.org/0000-0002-7686-2331
The official report can be read here:
In the report, they operate with 3 different scenarios for how much electronic waste Generative AI will generate until 2030
• Conservative forecast gives a cumulative weight of 9 million tons
• Moderate forecast gives a cumulative weight of 11.2 million tons
• Optimistic forecast gives cumulative weight 16.1 million tonnes
The weight of the world’s largest aircraft carrier USS Gerald R Ford weighs approximately 100,000 tons when fully loaded.
The weight of electronic waste from Generative Waste in 2030 will then be 90 to 160 times the weight of the world’s largest aircraft carrier.
This is alarming, but we hope to see documentation of how much of this is recycled and traded in an environmentally sound manner.
CO2 emissions
Globally, AI data centers now contribute 2.5-3.7% of greenhouse gas emissions, more than the airline industry! Last year, Google and Microsoft reported increased greenhouse gas emissions by 48% and 30% respectively since 2020, despite their zero-emission targets by 2030. Both companies blame AI.
As it stands now, it is a long time before AI can make a positive contribution to the environment, by 2030 we see these results:
- Electricity consumption on par with the entire UK
- Water consumption will be between 6 * Denmark and 1 time the US – every year.
- 5-30 tons of minerals from conflict zones and slave labour
- CO2 emissions that exceed the airline industry
- Electronic waste that is equivalent to 90-160 times the weight of the world’s largest aircraft carrier.
It is difficult to see that this can be sustainable. At the same time, this race is led by American Tech companies, which is problematic in these geopolitical times. We need to know who is driving the development, and who is securing values that promote our Norwegian, Nordic and European values.
Instead, we should promote Norwegian solutions that can be more sustainable. SINTEF Digital is actively working with Green AI. Ole-Christoffer Granmo’s Tsetlin Machine is also an exciting alternative. The latter promises to be both sustainable and explainable. Digital sovereignty comes as a bonus.
With the consumption of electricity, water and minerals, it is difficult to see that AI contributes positively to the environment. When it also produces large amounts of CO2 and electronic waste, the math becomes even worse. But let the big Tech companies answer for themselves, and hear what plans they have to put sustainable solutions in place.