AI's Dirty Secret Is Far Worse Than We Thought
AI's Dirty Secret Is Far Worse Than We Thought
We can't keep this up.
It’s no secret that AI has a bit of an energy problem and, in turn, an emissions problem. Training, maintaining, and running enormous AI models such as ChatGPT, Grok, or Stable Diffusion consumes the same amount of energy as entire cities. But The Guardian recently uncovered the sheer extent of the planet-destroying emissions these AIs are producing. Their investigation revealed that the in-house data centres (a crucial piece of AI infrastructure) of AI leaders Google, Microsoft, Meta, and Apple produce 7.62 times more emissions than these companies officially reported! How have they managed to cover this up? Well, let me explain.
As I said, AI consumes a lot of energy. Modern advanced AI requires terabytes upon terabytes of organised data in order to “train” the AI model. AI companies both store this data in and conduct AI training in data centres. These are essentially giant warehouses filled with enormous servers of memory and huge GPU-based supercomputer clusters. These supercomputers work at maximum capacity to process this data, analysing the trends found within it and, in the process, creating an AI. This takes a huge amount of energy! The training of ChatGPT-4 alone used around 7,200 MWh of energy, or enough to power a small US city for a year. When you query an AI like ChatGPT, you effectively send a request to this data centre; the AI combs through all of the data and, using statistics based on the trends it found, cobbles together a response. This also takes a tonne of energy. In fact, Google’s AI search uses ten times the energy of a regular search.
Data centres have been around for decades, but using them for AI like this is far more energy-intensive than it used to be. Not only that, but an abundance of new data centres are being built around the world to power the AI boom. As such, banks like Goldman Sachs are predicting that data centre power demand will grow 160% by 2030.
All the major non-startup AI companies, such as Apple, Amazon, Google, Meta, and Microsoft, have had their technology-based emissions scrutinised for years, so they have pledged to invest and develop in-house renewable energy, as well as buy-in renewable energy, to reach net-zero emissions as soon as possible. However, their energy consumption is now ballooning as a result of the AI and in-house data centres that power them, which has thrown a wrench in these eco-friendly plans.
This is where The Guardian comes in. You see, it turns out these companies are using “creative accounting” to make it appear as if their AI projects produce close to no emissions. But this is far from the truth.
These data centres use so much energy that they often need on-site carbon-belching turbines to keep them running, as the grid or any local renewable energy simply can’t supply enough energy. So, rather than paying for a proper low-carbon or net-zero energy solution, they instead turn to renewable energy certificates. These are basically carbon credits that show a company purchased renewable energy-generated electricity elsewhere to match a portion of its electricity consumption on-site. This allows them to “offset” their data centre emissions and claim they produce far fewer emissions than they actually do.
It turns out these AI companies have gone hog wild and purchased a tonne of these renewable energy certificates, allowing them to claim their data centres are far better for the environment than they actually are.
The Guardian did some digging and found that the actual emissions of these data centres were, on average, 7.62 times higher than claimed! But there were some outliers. Google’s data centre emissions were only 2.7 times higher than reported, and Amazon’s were only 5.6 times higher. However, Microsoft’s was 21 times higher, Apple’s was 402 times higher, and Meta’s was an astonishing 3,100 times higher!
Why does this matter? I mean, after all, the overall effect is the same; that’s why they are allowed to offset these emissions by buying renewable energy credits, right?
Well, that’s not entirely true.
You see, we can’t reach net-zero by burning fossil fuels in one location and buying renewable energy in another and claiming it has zero impact. This still increases the amount of carbon in the atmosphere. It also hinders consumers’ ability to choose climate-friendly products. If capitalism is to have any hope in hell of stopping climate change, industries’ or businesses’ actual emissions need to be crystal clear. Without clarity, governmental action on climate change could be hindered due to a lack of information about the actual carbon intensity of an industry or business. It’s also just straight-up lying to people. These AI companies are already some of the least trusted entities on the planet; this just adds to that concept.
The AI industry needs to be more transparent and ethical. From copyright infringement to hiding their emissions and inflating a market balloon, they aren’t exactly acting as the good guys. For your sake and mine, I hope we can hold this untamed industry to account.
Thanks for reading! Content like this doesn’t happen without your support. So, if you want to see more like this, don’t forget to Subscribe and help get the word out by hitting the share button below.
Sources: The Guardian, Will Lockett, Planet Earth & Beyond, TRG Data Centres