DATA DRAIN
AI Data Centers and Their Extensive Waste
While data centers have been around for decades, supporting search engines like Google, AI data centers are relatively new.
The prompts that go into large language models like ChatGPT or Gemini require a much more intensive workload than your typical search. This is because AI data centers have to support training and inference that's typically not included in normal data centers. They require a lot of different moving parts working at the same time to allow all the computers to work together and accelerate the speed at which AI can train, absorb workloads, and make that “thinking” screen when you type in a prompt shorter.
Why Is AI More Intensive?
This coordinated activity creates fluctuations in energy demand. Due to the large amounts of intricate computing occurring constantly within the servers that process data, the threat of overheating is increased. Just like when you have too many tabs open on your computer, AI servers begin to overheat as well.
As a result, the overheating puts a strain on both the power and cooling systems. In fact, it has been found that, compared to the average search in a traditional search engine, AI responses consume up to ten times more energy.
However, in order to power these components, large generators (which consume fossil fuels) are incorporated in the design of a data center, releasing large concentrations of harmful exhaust into the surrounding ecosystem and communities.
Even more so, the cooling structures within the data centers produce a constant low-frequency humming, disturbing communities and animals nearby.
Even more so, the cooling systems of data centres are also strained heavily by the amount of energy AI requires. Most data centers use liquid cooling to maintain their systems, a process that requires freshwater in either cold coils or plates to extract heat, or immersing the GPUs and other components in the liquids to dissipate the heat.
As a result of this process, the water often evaporates into the atmosphere, a substantial loss of freshwater considering that:
training only one model of a large language model like Claude can consume up to 700,000 liters of water
A Gemini data center was reported to use around 980 million gallons of water per year, equivalent to over 4 million households
In 2024, AI data centers throughout the United States consumed 14 billion gallons of water
However, there are ways to reduce this risk. It is possible for data centers to use non-potable and recycled wastewater instead of fresh drinking water, making the water consumption less impactful. However, these changes often cost more money and are only done when companies have the right pressure and incentives.
You can:
Attend public meetings and show your town council that there are voices that oppose data centers in your area
Demand public hearings to force local leaders to hold open forums rather than allowing data center projects to pass through quietly
Sign anti-data center petitions in your area
Partner with border advocacy initiatives and advocate for moratoriums
By protesting and signing petitions, you can reduce the harm data centers can pose to the environment and humans.
Comments