Environmental Impact of Generative AI and Babbily’s Sustainability Commitment

Introduction

Generative AI has revolutionized how we interact with technology – from AI chatbots and image generators to advanced language models – but it comes with environmental costs. Running large AI models requires significant computation, which translates to high electricity use and a carbon footprint. Recent studies and industry figures highlight growing concerns about the energy and water consumption of AI data centers. Babbily – a cutting-edge generative AI platform – recognizes these concerns. As a platform that brings together major closed-source and open-source large language models (LLMs) (e.g. OpenAI’s ChatGPT, Google’s Gemini, Meta’s Llama, Anthropic’s Claude) under one roof, Babbily is committed to delivering powerful AI capabilities in an environmentally responsible way.

The Environmental Footprint of Generative AI

Energy Use and Carbon Emissions

Running AI models – especially generative AI that produces text, images, or audio – consumes substantial energy in both training and everyday use (inference). Training state-of-the-art models is extremely energy-intensive: for example, creating OpenAI’s GPT-3 (175 billion parameters) consumed an estimated 1,287 MWh of electricity, emitting 552 tons of CO₂ – equivalent to the annual emissions of ~123 gasoline cars. Even somewhat smaller models can have hefty training footprints; the open-source BLOOM model (176 billion parameters) used ~433 MWh for training, emitting ~30 tons CO₂, which was noted as ten times a French person’s annual emissions. These figures illustrate that training large AI models can be comparable to large-scale industrial or transportation emissions.

Once a model is deployed, each inference (user query or prompt) also uses energy. OpenAI CEO Sam Altman recently revealed that an average ChatGPT query consumes about 0.34 Wh of electricity. That is a tiny amount – roughly what an oven uses in a little over one second, or what a high-efficiency LED light bulb uses in a couple of minutes. In Altman’s terms, the water used per query is also minuscule (≈0.000085 gallons, “roughly one fifteenth of a teaspoon”). Individually, a few watt-hours and drops of water seem trivial. However, scale matters: with millions of users and billions of queries, those fractions of a watt-hour multiply into significant demand. Researchers have warned that the surge in AI adoption could drive AI’s total electricity usage to 85–134 TWh/year by 2027 (a sizable fraction of global data center power), potentially rivaling or exceeding the power needs of entire countries. Indeed, data centers (including those running AI workloads) plus cryptocurrency mining already accounted for ~2% of global electricity use in 2022 (~460 TWh, comparable to the electricity consumption of Sweden or Poland). So while AI’s per-query energy might be small, the aggregate impact of widespread generative AI can be massive if not managed responsibly.

Figure 1: AI Energy Usage Comparison

To put AI energy use in perspective, it helps to compare it with other familiar activities like driving a car. One analysis estimates that a single prompt to an AI like ChatGPT uses about 0.005 kWh (5 Wh) of energy, which is roughly the energy needed to drive an electric car only 0.03 km (just 87 feet). In other words, one AI query is like driving a fuel-efficient car for only a few dozen paces. Another technologist quipped that the impact of one query is akin to driving a gasoline car at highway speed for about one second. These analogies highlight how small one AI prompt’s footprint is compared to typical transport energy usage. Even a “heavy” user of generative AI would use relatively little energy: for instance, 100 ChatGPT queries might consume on the order of 0.3–0.5 kWh (using older conservative estimates around 3–5 Wh per query). Over an entire year, 100 queries every single day would use about 110–180 kWh of electricity – which is about the energy in only 3–5 gallons of gasoline (enough to drive a medium-sized car perhaps 100–150 miles). By contrast, an average car driver (e.g. a daily Mercedes-Benz commuter) burns through vastly more energy. For example, driving 40 miles in a gasoline sedan could use ~2 gallons of fuel (≈67 kWh of energy), which already dwarfs the 0.034 kWh that 100 AI queries would consume in a day.

Figure 2: Carbon Footprint Comparison

Carbon emissions comparisons between AI queries and car use show a similar disparity: asking 10 questions a day to ChatGPT for a whole year might add ~11 kg of CO₂ to one’s footprint, whereas driving an average gasoline car for a year emits on the order of 4,000–5,000 kg CO₂. In one analysis, cutting 50,000 AI queries (roughly 10 queries/day for 14 years) would avoid only ~30 kg CO₂ – negligible next to lifestyle changes like avoiding a single transatlantic flight or living car-free for a year (which save hundreds or thousands of kg).

Figure 3: CO₂ Savings Comparison

To put things in perspective, skipping 50,000 AI queries saves about 30 kg CO₂, whereas common climate actions like living car-free or recycling lead to much higher CO₂ savings.

Energy Usage Across AI Models

The energy consumption of different AI models can vary significantly, from small open-source models to larger, more complex models like GPT-3. Smaller models are far more efficient in terms of energy consumption per query.

Here’s a chart illustrating Energy Usage Across Different AI Models:

Babbily’s Commitment to Sustainable AI

As an AI provider, Babbily acknowledges the environmental impact of AI and is proactively taking steps to limit that impact. “At Babbily, we recognize the significant environmental impact that AI generation can have due to its high energy consumption and increased carbon emissions,” says Chris Crawford, CEO of Babbily. Babbily’s strategy to promote sustainable AI is twofold: reducing/optimizing the energy usage of its services, and offsetting the remaining carbon footprint through investments in climate solutions.

1. Optimizing AI Efficiency (Prompt Caching & Smart Usage):

One way Babbily limits waste is by employing prompt caching and other efficiency techniques in its platform. Prompt caching is a technique that stores and reuses parts of AI prompts and computations that repeat, so the system doesn’t redo the same work unnecessarily. For example, if a user is analyzing a long document with multiple queries, Babbily can cache the encoded document context so that subsequent questions about the same document don’t require processing the entire text again. This can significantly reduce processing time and energy for repetitive or context-heavy tasks. Industry data shows prompt caching can cut token processing (and thus computation) by 50–90% in ideal cases.

2. Carbon Removal and Offsetting (Stripe Climate Partnership):

No matter how efficient AI operations become, there will inevitably be a non-zero carbon footprint from the electricity used. Recognizing this, Babbily has committed to offsetting and removing carbon to counteract the emissions of its AI services. In 2024, Babbily joined the Stripe Climate program – a coalition of forward-thinking businesses funding carbon removal projects. Babbily pledges 1% of its revenue to support carbon removal efforts that pull CO₂ out of the atmosphere. Over time, Babbily’s contributions to Stripe Climate help fund initiatives like direct air capture (e.g., Climeworks) and biomass carbon removal (e.g., Charm Industrial).

Final Notes

The rapid rise of generative AI has raised reasonable questions about its environmental impact. Indeed, running large models at scale consumes significant electricity and water, and the industry must continue innovating to manage this footprint. However, it’s important to keep the numbers in perspective: on a per-use basis, generative AI is far less carbon-intensive than many everyday activities like driving, heating, or even streaming video. As we’ve seen, one AI chat query uses about the same energy as an oven in one second or driving a few dozen feet – a negligible amount for an individual, though not negligible in aggregate. The path forward is clear: make AI smarter and greener at the same time. This means improving efficiency (better algorithms, hardware, and practices like caching), powering AI with clean energy, and offsetting emissions through credible means. Babbily exemplifies this approach by combining technical optimizations with a bold climate commitment. By being part of Stripe Climate’s carbon removal coalition and implementing energy-saving techniques, Babbily is ensuring that the convenience and creativity unlocked by generative AI do not come at the planet’s expense.

Sources: Generative AI energy and carbon data from recent analysesscientificamerican.compolytechnique-insights.comsustainabilitybynumbers.com; Sam Altman’s comments on ChatGPT’s per-query consumptiontheverge.com; comparisons of AI usage to other activitiesmedium.comsustainabilitybynumbers.com; Babbily’s sustainability initiatives and Stripe Climate partnershipapnews.combabbily.com; prompt caching and efficiency techniquesdocs.anthropic.comrequesty.ai; Stripe Climate carbon removal projectsclimate.stripe.comclimate.stripe.com; and Babbily company statements on accessible, eco-friendly AIapnews.com.