SUSTAINABILITY & AI

Is ChatGPT the next energy guzzler

How much energy does your AI query consume

There is an extreme heat wave in India right now and, as I sit in my air-conditioned room, shielded from this extreme heat wave, I stumbled upon a startling statistic on Twitter: A single query on GPT consumes about 15 times the energy of the same query on Google.

To put this into perspective, here’s a simple bar chart comparing the energy consumption of using Google search, GPT, and running a light bulb for an hour.

Bar chart comparing energy consumtpion of different sources drawn using ChatGPT-4

Imagine turning this statistic into a comparison of CO2 emissions for typical daily activities — the results are even more shocking.

Comparison of CO2 emission of various individual activities — Chart drawn by GPT-4

Essentially, A heavy AI user, who relies on AI for creating content or analyzing data, could unknowingly lead to emissions that could rival taking a flight!

As someone who drives an electric car and relies heavily on GPT daily for all sorts of tasks — this becomes quite a revelation.

Energy Breakdown: What Happens When You Raise a GPT Query?

Here is a simple breakup of what happens when you raise a query:

1. Computation: The computational model gets activated to generate a response parsing the query via the hundreds of parameters. This consumes the majority of energy.
2. Data Retrieval: Next, Energy is also used to retrieve any necessary data from storage that might support the query processing. This also includes the smaller portion of the energy that goes into the networking required to receive the query from the user and send back the response
3. Maintenance and Cooling: Finally, a portion of energy is used for maintenance of the data center and cooling of the hardware.

Breakdown of energy consuption of a ChatGPT query — Chart created by GPT-4

this analysis doesn’t even include the energy used during AI model training, which is significantly higher.

AI and Energy Demand: A New Challenge on the Block

While we can’t deny the AI wave sweeping across our lives, in a world that is already struggling to meet the COP goals and contain global warming within sustainable limits — this is a new challenge on the block.

World Economic Forum estimates that AI demand for computational power is doubling every 100 days. The energy required to run AI tasks is already accelerating with an annual growth rate between 26% and 36%. This means by 2028, AI could be using more power than the entire country of Iceland used in 2021.

There are emerging solutions like improving short-term efficiencies, implementing power usage caps during model training, and exploring long-term options like quantum computing.

Also, there is hope that harnessing the power of AI will drive energy consumption in other areas.

Conclusion

As we and tech companies weave AI more deeply into our daily lives into the most mundane tasks, from simple searches to complex creative endeavors, the need to consider its environmental impact becomes crucial.

Let’s ensure our technological advances don’t set back our environmental efforts. The pressing question remains: Will these efforts be enough, and are they happening quickly enough?