GenAI could accelerate global warming and climate change

Soaring daily temperatures, wildfires in Canada, inexplicably hot ocean currents; the worries around climate change have intensified. The COP28 Summit currently underway in Dubai is looking at all the ways in which human activity is creating uncertainty in the climate as we know it, impacting our lives in a multitude of ways.

There is another topic that rarely makes it to conversations on climate change – technology. Since the last year, the advent of Generative Artificial Intelligence (GenAI) has created a wave of frenzied excitement, as well as a shadow of worry on whether we have created a Frankenstein that we may not be able to control. What is considerably less talked about is how AI and GenAI can serve to accelerate global warming and climate change.

A lot has been written and talked about Blockchain and Crypto being an environmental disaster; but so are AI and GenAI. There are three ways in which GenAI degrades the environment:

• One, training gigantic GenAI models requires enormous amounts of energy, leading to vast carbon emissions. For example, training a 213mn GenAI model just once can spew the same carbon emissions that 125 New York-Beijing round flight trips would! This is a small model, GPT3 is 800 times bigger with 175bn parameters and GPT4 and Google’s Bard are possibly even larger. Besides, most of these models are trained multiple times to achieve optimum accuracy.
• Two, most of these models live on the cloud; for instance, GPT is on Microsoft Azure. The ‘cloud’ is nothing but hundreds of data centres around our planet, guzzling water and power in massive quantities. A recent Guardian article revealed that data centres currently consume 200 terawatt-hours per year, roughly equal to what South Africa uses today. With the emergence of GenAI, it is estimated that by 2030, this electricity consumption will soon cross that of Japan, the world’s third-largest economy! A global semiconductor conference predicted that the computing demand worldwide could outstrip the total world electricity power generation within 10 years.
• Three, the bedrock of GenAI models are the chips produced by the likes of TSMC and Nvidia. The colossal chip fabrication plants, or fabs, require huge amounts of electricity and pure water. A typical fab might require up to 5 million gallons of water and 30 to 50 megawatts of peak power a day – about the amount a town with 50,000 residents would use each day.

As Big Tech companies across the world, race to build larger and larger models, cloud providers build ever more powerful clouds, and semiconductor companies strive to build hundreds of fabs (fabrication facilities) across the world to satiate the demand of GenAI and other technologies, the carbon footprints increase exponentially, clean power struggles to keep up, and the water tables plummet. It is a bleak picture, and what can we do about it? 

We must build awareness among communities and societies by bringing technology and its impact on climate change into conversations at the large climate summits around the world. This will create an environment where Big Tech can participate in the dialogue and be responsible while they continue to create cutting edge technologies that make our world a better place.

A lot of tech companies are aware of this, and Microsoft, Google and others have pledged to make their clouds ‘green’ in the coming decades. However, there are other things that the industry can do. Each time a model is created or trained the producer must estimate and budget the carbon footprint, so they know what they are dealing it. Many use cases do not require a huge model, a smaller model using far less resources could do the job. Just throwing more compute power will not make a better model, there are other finetuning techniques which are much less power intensive.

The good news is that the Big Tech companies have realised this, and are looking at smaller models, training them in places with cleaner energy, using specialised lower-consumption data centres, and tweaking training protocols to reduce the footprint. Indian companies also have stepped up on this; for instance, NTT and Airtel have set up captive solar plants to power their own data centres and Hyderabad-based CtrlS has proudly become the world’s first LEED certified Platinum Green Data Centre.

About Jaspreet Bindra

Jaspreet, a seasoned corporate leader, held key roles at Mahindra Group and Microsoft India. As Founder of The Tech Whisperer, he advises global organizations like ThoughtWorks and PWC on tech, AI and Gen AI strategy. A respected author, coach, and educator, Jaspreet, with an MBA and recent Master's in AI from Cambridge, drives The Tech Whisperer Ltd. which is about Thought Leadership in AI, Digital Transformation, and other frontier technologies like Web3 and the Future of work.