How Your AI Queries are Putting Carbon Back in the Air

Nowadays, it is a trend to use AI [artificial intelligence]-powered language models such as OpenAI’s ChatGPT, Meta’s LLaMA and Perplexity AI in almost everything we do. Whether it is a simple search query for an individual user or analysing complex data for corporates, these models are everywhere. They make complex data processing simple, searching multiple sources possible in a few seconds and in today’s era of rush, speed and fast-paced life, these are precisely the attributes people cherish and demand more of—efficiency, instant results and the ability to multitask seamlessly. But what’s hidden behind the scenes is the environmental costs of these quick results—the carbon emissions these queries generate.

While seemingly harmless, each query these AI models process contributes significantly to energy consumption and carbon emissions. Every question asked releases CO₂ into the atmosphere, contributing to global warming.

The Energy Behind AI Queries

 AI-powered models such as ChatGPT, Meta AI and Perplexity AI consume 10 times more energy than traditional search engines like Google. The US Energy Information Administration, Electric Power Research Institute (EPRI) reported that each query made to AI models consumes around 290 Wh [watt-hour] of electricity. India is one of the largest user bases for AI-powered tools, accounting for about 6-7% of the global AI user base.

With over 10mn users actively interacting with AI models, generating around 10-15bn queries annually, the energy required to power these queries is mammoth—over 30mn kWh [kilowatt-hour] annually. To put this into perspective, this is enough energy to charge over 6bn smartphones or to power 7mn 10-watt LED bulbs running for 12 hours a day, every day, for over a year. These queries cumulatively generate between 13,000-15,000 metric tonnes of CO₂ annually, equivalent to driving 50mn km in a typical petrol car, which is enough to circle the Earth over 1,200 times! To absorb this quantum of CO₂ would need over half a million trees each year.

AI models need 10 times more energy than traditional search engines due to the computational complexity of AI models, which need to run deep neural networks across billions of parameters to generate responses. Unlike traditional search engines that retrieve information from pre-indexed web pages, these AI models create responses from scratch, based on user input, requiring considerably more processing power.

The energy consumption is mainly driven by running the servers used for processing and keeping the data centres hosting these servers cool. Some large data centres can cover over 1,00,000 square feet, the size of a few football fields, housing tens of thousands of servers. To maintain optimal performance, temperatures in server rooms must be between 18-27°C; despite the constant heat generated by operating equipment. Cooling such spaces requires industrial-scale HVAC (heating, ventilation and air conditioning) systems that far surpass any typical residential setup—almost equivalent to the cooling power of 10,000 household air conditioners running simultaneously. On top of that, even more energy is consumed during the training phase of these models, and every few months, a new version gets released!

As the global temperature continues to rise, AI models are inadvertently contributing to the problem. According to the Intergovernmental Panel on Climate Change (IPCC), rising CO₂ levels are closely linked to global warming, leading to more extreme and disruptive climate events. Every query AI models process adds to this rise in CO₂, exacerbating the risks of heatwaves, flooding and unpredictable weather patterns. In India, for instance, recent extreme climate events, such as record-breaking heatwaves and severe monsoon flooding, can be linked to the broader trend of global warming driven by human-induced CO₂ emissions. The use of AI tools is predicted to grow exponentially, which would further add stress to the planet that has already breached a 1.5°C temperature rise.

Reducing the Carbon Footprint

Reducing the carbon footprint of AI models requires both individual efforts and system-level changes. Users must adopt mindful AI usage. This means using AI tools efficiently, avoiding unnecessary queries and being conscious of the cumulative environmental cost. Users can also offset their emissions by participating in carbon-offset programmes, such as tree planting or investing in renewable energy initiatives.

One of the most impactful system-level changes is the shift toward renewable energy to power these data centres. Today, renewable energy is cheaper than fossil fuels, and storage now provides a reliable 24-hour green power alternative. Powering data centres with renewable energy sources like wind, solar or hydroelectricity can drastically reduce CO₂ emissions. Companies like Google and OpenAI are already investing in green energy for their data centres, but these efforts must be expanded. By adopting more energy-efficient cooling technologies, data centres can also lower their power usage. Further, developers can significantly reduce the energy consumed per interaction by developing more energy-efficient models that require fewer computational steps per query.

The carbon footprint of AI models may seem minimal on a per-query basis, but when aggregated across millions of users, the environmental cost becomes substantial. Each query puts more CO₂ into the atmosphere, contributing to global warming and increasing the frequency of climate disasters. However, by shifting to renewable energy sources, we can mitigate the environmental impact of these powerful tools. Being mindful of AI usage and supporting sustainability initiatives can make a significant difference for individuals. As the use of AI continues to expand, it is crucial to remember that small actions—like minimising unnecessary queries or supporting green energy—can collectively contribute to a more sustainable and climate-resilient future. The question is, what action are you taking on this?

The author is a renewable energy and climate tech expert, and founder of Mynzo Carbon & SolarArise. Views are personal.