Cloud computing applications seem to exist everywhere and nowhere. Consequently, what is now being called the fourth industrial revolution has often been promoted as clean and even green — no dirty smokestacks, no chemical run-off, no adverse environmental impact.
But this is a myth — a myth that conceals the true cost and consequences of these innovations. Cloud distributed services are not disembodied technological spirits residing in the air; they have a very real, physical presence — specifically server farms that consume large amounts of electrical power and need to be cooled down with water.
OpenAI’s ChatGPT may have been developed in San Francisco, but it resides in Des Moine, where the server cluster supporting the large language model quenches its thirst on the Midwest’s rather ample supply of clean, fresh water. In response to a lawsuit filed by local residents, it was revealed that the training of this powerful generative AI system consumed 6% of the district’s water. This number may not seem like a lot, but it tells us two things: First, AI is thirsty and its need for water is only going to increase. How much and what this means is the second item, as there are no federal regulatory standards for reporting and evaluating the environmental impact of AI.
Fortunately, we are beginning to see signs of a possible solution. Last month, Senator Ed Markey introduced a bill to bring some transparency to the actual environmental costs of AI. Whether the bill actually passes into law, however, remains to be seen.
