AI has the potential to be a force for good in many ways, but it also raises significant concerns. One pressing issue is the energy-intensive nature of generative AI, which is driving escalating demand for power-hungry data centers, many of which still rely on burning fossil fuels. Morgan Stanley estimates that data center power demand from generative AI will increase by around 70% annually until 2027. If this trend continues, generative AI could be consuming as much energy annually as Spain did in 2022. Unless addressed, this growing energy use will lead to a sharp rise in global emissions.
Businesses have a responsibility to act sustainably to ensure environmental factors remain top of mind in AI development. IBM is one company that is leading the way in meeting this challenge. Here are six ways in which it is incorporating sustainability into its approach to AI.
1. Hardware innovations
IBM has demonstrated a strong commitment to developing hardware that supports the sustainable development and use of AI. For example, the IBM Telum II processor and IBM Spyre Accelerator, set to power the next generation of IBM Z mainframes in 2025, are specifically designed to reduce AI-based energy consumption. Their advanced IO technology includes a simplified scalable IO sub-system that reduces energy consumption and data center footprint.
In particular, the Telum II's on-chip AI accelerators help to optimize power usage by reducing data transfer overhead and improving system throughput. They are designed for enterprise environments where power efficiency and minimizing cooling needs are critical to reducing data center energy consumption.
Another example is the Blue Vela supercomputer, which IBM uses to train its Granite 3.0 series of Large Language Models (LLMs). Blue Vela is equipped with NVIDIA H100 GPUs and operates entirely on renewable energy, helping to minimize the environmental impact of training LLMs.
2. Energy efficient AI model development
Energy efficiency underpins many aspects of IBM's approach to developing AI. For example, its AI strategy has been focused on developing smaller, more efficient, more specific and less energy-hungry LLMs like the Granite series. With models ranging from 2B to 8B parameters, the Granite family delivers high performance while maintaining a smaller environmental footprint compared to larger models.
When it comes to LLMs, the bigger the model, the more power it consumes - both in training and inference. Smaller models use fewer computational resources, consume, and have less energy and have a lower carbon footprint.
Another way IBM prioritizes energy efficiency is by seeking ways to minimize the data required by AI models. For example, it uses techniques like data pruning, where only the most relevant data is used for training. With less data, you reduce the data processing burden and cut storage requirements and energy consumption.
IBM has pioneered research in iterative scaling of LLMs - an energy-efficient approach that grows large models by combining smaller ones, rather than training from scratch.
Researchers at the MIT-IBM Watson AI Lab demonstrated that models can be trained four times faster - consuming less energy in the process - by recycling parameters from smaller, earlier versions. Their Linear Growth Operator (LiGO) technique analyzes weights from existing small models to efficiently scale up to larger ones, reducing computational costs by up to 50 per cent.
IBM is advancing energy-efficient AI development with InstructLab's LAB (Learn-Adapt-Build) methodology, which simplifies and accelerates AI model customization while being environmentally friendly.
InstructLab uses taxonomy-driven data curation, organizing knowledge into categories (like a tree) to identify gaps in a model's understanding. Starting with small, curated datasets, it generates synthetic data - computer-created examples that mimic real-world data - reducing the need for large, manually labelled datasets and saving time and resources.
Its low-fidelity workflow allows organizations to tests ideas on standard laptops, enabling quick iterations and feedback without requiring expensive hardware. Additionally, the teacher-trainer architecture lets pre-trained models ("teachers") transfer knowledge to new models ("trainers"), cutting computational costs and avoiding retraining from scratch.
When paired with IBM's watsonx.ai platform and iterative scaling techniques, InstructLab provides a sustainable AI ecosystem. Businesses can optimize resources from prototyping to deployment, while community-driven improvements and automated parameter optimization enhance efficiency and accessibility.
3. Sustainable practices and operations
IBM promotes sustainable AI practices and operations in a variety of ways. For example, one of the keys to making AI more sustainable is to ensure AI operations are powered through green data centers that rely on renewable energy. IBM's expertise and support for hybrid cloud - combining public and private cloud with on-premises infrastructure - helps in this regard by enabling organizations to locate their AI processing workloads in areas that are geographically close to renewable energy sources.
In fact, Christina Shim, IBM's Chief Sustainability Officer, suggests that this part of the reason why 74% of the electricity consumed in its own data centers comes from renewable sources.
More broadly, IBM offers solutions that use AI to optimize IT operations for sustainability. Its Maximo Application Suite, combines IoT data, analytics and AI to streamline asset operations to promote sustainability across the asset lifecycle, reducing energy consumption and waste.
Maximo enables data-driven decision-making that considers environmental factors in several ways. It uses information captured by sensors and technicians to support decisions to extend the life of assets and minimize their carbon footprint. Through asset investment planning, it assesses when replacing aging assets with more efficient ones is financially prudent. Maximo also effectively tracks greenhouse gases and controls continuous and fugitive emissions.
4. AI for climate action
AI itself is a powerful tool to help tackle climate change impact and IBM is leading significant initiatives in this area. In partnership with NASA, IBM has developed advanced geospatial AI models that analyze satellite imagery to accelerate environmental insights and solutions.
These models can be fine-tuned and applied across multiple areas to help reduce climate impact. They represent a game-changer, according to Alessandro Curioni, IBM Fellow and Vice-President for Accelerated Discovery, saying: "they allow us to better understand, prepare and address the many climate-related events affecting the health of our planet in a manner and speed never before seen".
The Kenyan government has formally partnered with IBM to implement the geospatial foundation model in its National Tree Growing and Restoration Campaign, enabling precise tracking and visualization of tree planting in specific water tower areas. Local developers can create fine-tuned models incorporating localized information to monitor forest restoration and measure carbon sequestration.
In another partnership with Mohamed Bin Zayed University of Artificial Intelligence, IBM has deployed a fine-tuned version of its geospatial model to analyze Abu Dhabi's urban environment and how the regional landscape influences urban heat islands - areas with significantly higher temperatures compared to surrounding locations.
5. Collaboration and advocacy
IBM champions AI ethics and sustainability through multiple initiatives. As a founding member of the Responsible Business Alliance, IBM commits to working with suppliers who maintain high standards of ethical, environmental and social responsibility. In 2022, IBM exceeded its commitment to promote AI ethics practices by training over 1,000 ecosystem partners, and has set up a new goal to train 1,000 technology suppliers in ethics by 2025.
The beauty of the open source approach to AI that IBM champions is that it enables a diverse global community of developers to collaborate on building AI models, sharing insights and driving improvements. This has the potential to unearth new ideas and innovations to make AI more energy-efficient and sustainable, which might not emerge in closed systems.
At the same time, IBM's focus on ethical AI includes promoting sustainable software development practices that apply to AI, such as optimizing code for lower resource use, using less intensive algorithms where possible, and encouraging software reuse to reduce the need for new resource-intensive builds.
6. Education and outreach
IBM's initiatives in AI education and training, particularly IBM SkillsBuild, directly address sustainability through comprehensive courses that combine technical skills with environmental principles. The program equips the next generation of AI professionals with both technological expertise and a sustainability mindset, preparing leaders for the green economy.
In November 2023, IBM launched a first-of-its-kind roadmap of sustainability skills courses, available for free through IBM SkillsBuild, to empower future climate action.
This initiative provides training that connects cutting-edge technologies to ecology and climate change to help generate a pipeline of sustainability talent. Using AI-powered recommendations, the interdisciplinary coursework of the training aims to combine topics such as ecology and biodiversity with practical technology training in AI and data analytics. The overall aim is to equip the next generation of leaders with skills for the green economy.
As generative AI adoption accelerates, its environmental impact demands immediate attention. Businesses must follow IBM's leadership in harnessing AI's potential while implementing sustainable practices that protect our planet.
This blog was originally published on the IBM Community.