Impact on the environment
Every word, image, generated video you see does not come from nowhere. Behind it exists a massive physical engine that consumes energy and water resources at an alarming rate.
The hidden reality of AI
Uruguay
In drought-affected areas, citizen protests have erupted against the construction of enormous data centers by multinational companies that planned to extract millions of liters from their drinking water reserves.
Oregon, USA
In 2021 alone, a single Google data center in this region consumed 355 million gallons of water from local supplies to keep its servers cool.
Ireland
The rise of AI has pushed the country's power grid to the limit. The data centers there consume as much electricity as the entire nation, forcing the use of highly polluting diesel generators to prevent collapses.
Virginia, USA.
Thousands of diesel generators the size of train cars operate as backup, emitting toxic fumes and creating serious air quality issues for local communities.
Cerrillos, Chile
Google planned to build a gigantic data center in the Cerrillos commune, which would use millions of liters of water daily for cooling, threatening the central aquifer of a capital that is already suffering from a mega-drought lasting over a decade.
Goodyear, Arizona, USA.
In the middle of the desert, data centers not only devour water but have also destroyed the peace of the community. The immense industrial air conditioning turbines needed to cool the servers operate 24/7.
Water meter
Freshwater evaporated in the last 24 hours to cool AI servers.
100
Liters
Emissions Counter
Tons emitted into the atmosphere by global AI usage in the last 24 hours.
0
Tons of CO₂
The Physical Journey of Your Prompt
Your device
You draft and send the prompt.
networks
The information is sent to the data center.
Data center
Numerous GPUs work to process the information.
Your device
You receive the result.
The anatomy of consumption
Before anyone wrote the first ‘Hello’, models like GPT-4 had to read a large part of the internet. This process required thousands of GPUs (graphics processors) running at 100% for months.
Training a single large model emits about 280 tons of CO₂, equivalent to the entire lifespan of 5 average gasoline cars, including their manufacturing.
It is the invisible and constant consumption. Every time you request an image or a summary, servers have to perform millions of new calculations.
While training is a huge one-time expense, inference (active use) already exceeds that energy consumption every few weeks. It is a constant energy hemorrhage.
Calculate the footprint of a prompt.
water
CO2 emissions
0
ml
0
g
