It’s well-known that artificial intelligence (AI) is resource-intensive, and recent data is shedding more light on this issue. According to a report by Epoch AI, training Elon Musk’s latest AI, Grok 4, cost nearly half a billion dollars and used as much energy as a small town consumes in a year.
Amazon co-founder MacKenzie Scott has donated over $19 billion to charity in just five years
Diamond batteries powered by nuclear waste promise 28,000 years of clean energy
Staggering Financial and Environmental Costs
The Epoch AI report provides detailed figures on the cost of creating a cutting-edge large language model. Specifically, training Grok 4 required $490 million just for computing power, which is nine times more than what was spent on Meta’s Llama 3 model.
Environmentally, the training process used 310 million kWh of electricity (equivalent to the annual consumption of a town with 4,000 residents), required 750 million liters of water for cooling (equivalent to 300 Olympic-sized swimming pools), and emitted 140,000 tonnes of CO2 (comparable to the emissions of an airliner over three years, which may not seem significant at first glance).
These figures do not even account for the costs of salaries, research and development, or the energy needed to operate the service on a daily basis.
The Blame on the “Colossus” Supercomputer
This excessive resource consumption can be attributed to the massive infrastructure used by xAI, Elon Musk’s company. Grok 4 was trained on the “Colossus” supercomputer, reportedly the world’s largest dedicated to AI. It consists of over 200,000 of Nvidia’s latest and most powerful chips.
What Do We Make of This?
These figures reveal the tangible reality behind the “cloud” of artificial intelligence. Far from being an ethereal technology, cutting-edge AI is a heavy industry with significant environmental and financial impacts, and it is anything but dematerialized.
This race for power raises serious questions about the sustainability of this development model. Only a few tech giants, backed by billions in funding, have the resources to build such supercomputers. This creates an almost insurmountable barrier to entry and concentrates immense power in the hands of a few companies. The question is no longer just whether AI is “intelligent,” but also whether its cost to the planet is acceptable. And for you, is the ecological cost of artificial intelligence a concern?