The unseen price of intelligence: Can GenAI be truly sustainable?

As the generative AI tools like ChatGPT, DALL·E and Bard fundamentally change the landscape of industries, a quieter dialogue is developing, focusing on their footprint on the environment. These models offer an array of possibilities—to improve healthcare, predict climate, but also consume extreme amounts of energy, water and rare minerals. It is easy to accept generative AI as sustainable amidst climate anxiety and ESG commitments, but can it be?

In this conversation with Padmashree Shagrithaya, Executive Vice President, and Head of Insights & Data Global Business Line, Capgemini, we explore generative AI's contradictory nature: its potential to quickly advance sustainability objectives and its potential to thwart those objectives. In almost three decades working in data science and AI, Shagrithaya provides a realistic perspective of the trade-offs, the innovation that is to come, and why responsible AI must now factor in mainstream accountability for the environment.

With GenAI’s rapid evolution, how is it shaping the sustainability dialogue within tech?

GenAI presents both an opportunity and a challenge. On the positive side, GenAI will enable rapid innovation across climate modeling, materials science, energy use optimization, and sustainable supply chain management. On the negative side, the training and operation of large GenAI models is, by definition, resource- and energy-intensive. If we don't think this through, we could be solving some environmental issues while creating others.

What are some lesser-known environmental impacts of GenAI?

Most people are focused on emissions from training models, but there's quite a bit more to look at. We must also consider both the water used for cooling data centers (which is especially critical when data centers are located in semi-arid areas), and the rare earth elements used for chips, which arguably generates end-of-life electronic waste. We can't lose sight of the emissions generated during inference—every time a user queries a model, there is an associated energy use. While model usage is limited now, it will scale and with scale comes impact.

How can tech companies innovate while staying environmentally responsible?

It starts with self-regulation. Companies should:

• Opt for energy-efficient AI models via distillation or sparsity
• Choose cloud providers powered by renewable energy
• Include carbon, water, and e-waste KPIs in AI governance
• Be mindful of hardware sourcing and disposal
• Build sustainability into the full AI lifecycle—from design to decommissioning

Frameworks like the EU AI Act address ethical concerns, but we urgently need regulations around environmental accountability too.

Why is it difficult to measure GenAI’s environmental impact?

Should sustainability become a core pillar of AI governance?

We have a Responsible AI ideology that encompasses the environment. We build tools like Hugging Face’s carbon trackers, and we facilitate reuse using open source, to avoid redundant training. The enterprise needs to have a Green AI First mindset - where carbon and water metrics are as common as fairness or accuracy.

How is Capgemini supporting clients in embedding sustainability into GenAI?

Through our RAISE platform (Reliable AI Solution Engineering), we help clients:

• Reuse models and centralize pipelines
• Optimize models for lower emissions
• Track carbon outputs with built-in calculators

Our campuses in India run entirely on renewable energy, and our GenAI-powered Energy Command Center optimizes campus energy use. We also support clients through our Sustainability Data Hub, which helps report ESG performance.

Can you share a real-world example?

We encourage using Small Language Models (SLMs) over large ones and are backing innovations like Liquid AI, which mimics nature to reduce compute costs. We also promote edge AI to minimize data transmission energy. These methods have helped clients significantly lower their AI-related carbon emissions.

Does Capgemini use internal sustainability metrics for GenAI?

What trends or innovations could make GenAI greener in the next few years?

Key areas include:

• Clear, enforceable global regulations
• Lightweight and domain-specific models
• Energy-efficient AI chips like TPUs
• Rethinking architectures beyond transformers
• Growth of green, open-source foundation models

These innovations, combined with intent and awareness, will be game changers.

What gives you hope for a sustainable GenAI future?

Human intent. From energy reuse in data centers to rising customer demands for transparency, we’re seeing real change. But we must act fast. Only when sustainability becomes part of our decision-making DNA—what we measure and manage—can we truly scale GenAI responsibly.

Watch the full conversation: