Power Hungry Machines

In a nondescript data centre campus in West Des Moines, Iowa, row upon row of NVIDIA H100 GPUs hum at a constant pitch, each processor drawing 700 watts of power whilst generating enough heat to warm a small home. Multiply that single GPU by the 16,000 units Meta used to train its Llama 3.1 model, and you begin to glimpse the staggering energy appetite of modern artificial intelligence. But the electricity meters spinning in Iowa tell only half the story. Beneath the raised floors and between the server racks, a hidden resource is being consumed at an equally alarming rate: freshwater, evaporating by the millions of litres to keep these silicon brains from melting. The artificial intelligence revolution has arrived with breathtaking speed, transforming how we write emails, generate images, and interact with technology. Yet this transformation carries an environmental cost that has largely remained invisible to the billions of users typing prompts into ChatGPT, Gemini, or Midjourney. Th