Can AI Data Centres Ever Compensate for Their Environmental Damage?

­For quite some time, much of the excitement in the industry has been around data centres. These massive facilities have been planned all around the world, and most of the talk has been how to provide the power required through ageing energy grids. The main problem is that data centre operators want to deploy the latest Nvidia processors to train and run inference for AI applications, and to run the same number of processors in the same space as before. However, the latest Nvidia processors consume a lot more power than previous generations and that is placing a strain on power grids around the world. This has left data centre operators looking for ways to provide that power on their own, mainly through generators, which are very unsustainable as they normally run on gas or diesel.

Recently, a new Amazon data centre made the news as it was expected to be the biggest polluting facility of any kind in the world. According to ‘The Independent’, it could actually double the previous emissions record if it runs to full capacity. The facility is based in Pecos County, Texas and it will have its own natural gas power plant to provide the energy it requires. The plant is permitted to release 33 million tons of carbon dioxide per year, according to ‘The New York Times’. That is despite the fact that Amazon promises to be Net Zero by 2040. The same source claims that the current largest polluting facility in the world currently is Alabama’s James H. Miller Jr. Power Plant, a coal facility which emits about 16 million tons of carbon dioxide annually.

Those types of figures are incredible, and there are tens of thousands of data centres being built around the world, which may not use as much power as the Amazon one, but are a substantial size, require a lot of power and create tons of emissions each year. The most worrying thing of it all is that a few analysts are starting to question the need for so many facilities, and if they are actually required. There is currently no killer application for AI, and those analysts are asking if that computing power is actually needed. Other countries and cities have started to place moratoriums on new data centres until the electrical grids are ready, or there is a more compelling business case. There has even been a growing number of protests against the location of data centres due to light and noise pollution from current facilities. Large operators, such as Amazon are determined to plow on and install as much capacity as possible, no matter the cost in dollars, or to the environment, but is it time to pause and try to figure out a more sustainable roll-out, one that matches short and medium term demand?