Infosys Q&A: Balancing AI & Environmental Responsibility

As companies rapidly adopt AI technologies, Sunil Senan, Global Head of Data, Analytics and AI at Infosys offers insights into implementing environmentally conscious AI solutions without compromising business objectives. 

Sunil is Senior Vice President responsible for the Data & Analytics service line at Infosys.

In this role, he works closely with Infosys’s strategic clients on their data and analytics led digital transformation initiatives. He is passionate about how data and analytics is creating economic impact in society and how enterprises and governments can engage in driving this transformation.

He has written the “Data economy in Digital times” paper articulating how the new data economy presents a set of new possibilities for enterprises, governments to serve their citizens and consumers.

Drawing from Infosys' successful client partnerships across multiple industries, he reveals practical strategies for reducing AI's carbon footprint through model efficiency, edge computing, and renewable energy integration. Sunil's expertise illuminates the path forward for organisations seeking to harness AI's transformative power while advancing their sustainability agendas.

AI systems are known for their significant energy consumption, which can impact sustainability goals. How can companies implement AI in a way that minimises environmental impact while still achieving business objectives?

Companies can minimise the environmental impact of AI while achieving business objectives by prioritising model efficiency, utilising computationally lighter open-source approaches, and optimising the training processes. Additionally, optimising data center efficiency and utilising renewable energy sources to power AI infrastructure is fundamental. Beyond infrastructure, strategic decisions regarding AI models are essential. While closed models may offer faster implementation, open models often provide specialised capabilities and, crucially, reduce the data and computation required for training, leading to a smaller carbon footprint. 

• Data processing and insight generation at the edge: By processing data locally, companies can decrease reliance on large data centers, thereby lowering their carbon footprint. Techniques like "tiny AI" models for edge networks, quantisation, pruning, knowledge distillation, and the use of pre-trained models can further minimise energy consumption during both training and deployment. Decreasing the amount of required computation for model training is another way to decrease the environmental impact of AI. For example, consider open-source AI models. The recent DeepSeek announcement significantly validated open-source AI development. These open-source approaches require much less computational power for training, leading to a smaller environmental footprint compared to training large models like ChatGPT, which rely on energy-intensive data centers.
• Model compression: Reducing the number of parameters without significant loss of accuracy can be achieved through three main techniques.
  • Knowledge Distillation: The “teacher” model trains the “student” model using only the key layers and weights. This helps to create customised small language models that requires less energy for inference when compared to LLMs.
  • Network Pruning: AI architects can remove parameters that contribute the least to model accuracy.
  • Quantisation: Model weights can be stored in quantised units of eight bits or less. This makes the model four times smaller but only two % less accurate.

These steps allow organisations to reduce the size and complexity of their power-hungry AI models that will minimise environmental impact while still achieving business objectives.

What strategies can businesses employ to identify AI opportunities that enhance sustainability initiatives?

Businesses should focus on AI applications that drive both operational efficiency and environmental benefits. For example, in supply chain management, AI can analyse data to optimise logistics and inventory management, reducing transportation emissions and lowering the carbon footprint. By targeting areas where AI can streamline processes and minimise resource consumption, companies can simultaneously achieve efficiency gains and advance their sustainability agendas. 

The AI for sustainability algorithms continuously maximise renewable energy integration into traditional sources, predicts energy demand fluctuations, and optimises energy distribution management systems. Through real-time monitoring, businesses are accelerating towards a future of sustainable energy.

Businesses can identify AI opportunities for sustainability by analysing their operations and value chain, focusing on areas with high environmental impact like energy consumption, waste and resource use.

Beyond improvements in logistics, supply chain and manufacturing, AI can boost overall workforce productivity, enabling companies to achieve increased efficiency that can shrink a company's carbon footprint by reducing resource consumption related to office space, employee commutes and other operational needs.

Analysing data from various sources and staying informed about AI advancements are crucial for identifying and capitalising on emerging opportunities. 

What are the challenges organisations are facing in balancing AI and sustainability?

The initial surge in generative AI has prioritised exploring numerous applications and rapidly scaling them. However, corporate leadership will soon demand continued AI adoption at the same accelerated pace, but with a renewed focus on sustainability. The emerging mandate may well be "do more AI with less emissions," forcing businesses to navigate the complex intersection of technological advancement and environmental responsibility. There's no easy solution. 

• High energy consumption: Training large AI models demands substantial energy, leading to increased carbon emissions. Most businesses rely on cloud providers like Azure and Google for the computational resources needed to train AI models, giving them limited control over the sustainability practices of these providers unless they train models on-premise. It's possible that more sustainable cloud-based AI services could emerge, potentially at a premium price. 
• Data center efficiency: AI operations often rely on data centers, which consume about 2% of the world's electricity. Based on these recent statistics, it's clear that cloud computing and storage already consume a significant portion of global electricity. The rapid expansion of AI is only expected to accelerate this energy consumption.

• Model complexity and size: Larger AI models offer advanced capabilities but require more computational power, leading to higher energy consumption. The only counter to this driver of energy consumption is the use of more computationally efficient models and techniques, usually achieved by leveraging open-source AI innovation.  
• Ethical data practices: AI learns from historical data and makes inferences based on the training sets and models it has access to. If inherent biases exist in our system, AI takes it as the natural way to take things forward. AI isn’t naturally aware of morals or ethics to decide otherwise. Therefore, the liability of removing biases or bringing in ethics is on the people creating and implementing AI systems.
• Integration of sustainability in AI strategy: Many organisations struggle to embed sustainability into their AI strategies effectively.  Sustainability is often discussed within organisations, but real invested capital and management attention is not always devoted to sustainability programmes.  

How can enterprises develop their data infrastructure to promote sustainability?

It’s important that enterprises urge their cloud and AI computing providers to power their data centers with renewable and carbon free energy sources. This might mean paying a premium for sustainable cloud services. Not just “green-washed” with purchased credits and offsets, but the data centers that are directly powered with renewable energy.  

Enterprises must create a data and AI environment that is "responsible by design." This entails building a robust data foundation that complies with trust, ethics, security, privacy and regulatory requirements. By ensuring that data practices are ethical and secure, companies can support scalable AI solutions that align with sustainability principles. 

Some countries are leading the way in developing sustainable data infrastructure. Norway, and a few other select locations for example, have successfully combined AI growth with sustainability efforts.

Since Norway's electricity is primarily generated from hydroelectric power (96%), the rapid expansion of data centers in the region provides economic benefits while also ensuring that AI models deployed in those data centers are inherently more sustainable.

France also demonstrates leadership in carbon-free computing, boasting a 90% carbon-free electricity mix.

Could you provide examples of companies that have successfully integrated AI to drive both business value and sustainability? What can we learn from their approaches? 

Our customers have successfully integrated AI to drive both business value and sustainability by leveraging Infosys Sustainability Intelligence Cloud Solution, an AI-enabled cloud solution empowered by a strong partner ecosystem that provides ease of integration with client and supplier data to provide transparency, and to derive actionable insights on sustainability parameters across the enterprise and extended supply chain.

Infosys partnered with a leading investment management firm to build an ESG (Environmental, Social, and Governance) KPI calculation engine. This firm, like many in the financial sector, was facing significant challenges with data – specifically around quality, accessibility and standardisation. They needed a way to efficiently aggregate and analyse ESG data to meet increasing regulatory requirements and to make informed investment decisions.

Our solution involved migrating their existing data infrastructure to the cloud and implementing AI-driven tools to optimise data sources and automate the calculation of over 100 ESG KPIs, incorporating methodologies from MSCI and TCFD. This dramatically improved their ability to generate comprehensive and reliable ESG reports, enhance data quality and gain deeper insights into the ESG profiles of their investments.

Ultimately, this allowed them to not only meet regulatory obligations but also to better understand and manage ESG risks, benchmark portfolios and drive long-term value creation. This is a concrete example of how AI can transform ESG data from a reporting burden into a strategic asset, driving both compliance and business value.

In another instance, Infosys partnered with a large beverage company on automating ESG metric validation. The program encompassed 1,500+ metrics across ten categories with 100 pilot metrics focused on the initial phase. We identified validation methods, pain points like manual operations and data limitations and recommended best practices for optimisation.

The goal was to provide actionable ESG insights, simplify regulatory compliance and build trust through transparent and accountable reporting, all while adhering to relevant sustainability standards and regulations. We successfully recommended the next steps with responsible by design at core. 

We worked with a major retailer to streamline their utility data management, which was a complex process involving emissions reporting and compliance with regulations like SB253 and SB261.

We recommended data-driven solutions, including AI-powered automation, to streamline their ESG reporting which leads to cost savings, improved compliance and a clearer understanding of their environmental impact. This will allow them to target their sustainability efforts more effectively.

Explore the latest edition of Sustainability Magazine and be part of the conversation at our global conference series, Sustainability LIVE. 

Discover all our upcoming events and secure your tickets today.

Sustainability Magazine is a BizClik brand