Making AI sustainable is crucial to ensuring its accessibility in all countries and reducing the digital divide. Understanding a country’s AI ecosystem is essential for taking a holistic approach to enabling sustainable AI.
The AI ecosystem consists of the physical foundations (data centres and related infrastructure), the data and the model innovation layer (foundational models), the application layer (built on top of the foundational models), and lastly, the end users.
Access Partnership proposes the following principles to drive Sustainable AI:
1. Responsibility for All
Sustainability is the responsibility of every player in the ecosystem – from the physical foundation to the end users. This includes a commitment to operational efficiency in data centres at the physical layer, efficient model inference at the model innovation layer, adherence to green software standards at the application layer, and commitment to minimising redundant and unnecessary AI usage by end users.
2. Human-centred & Purpose-driven AI
AI should be deployed thoughtfully, ensuring that its use cases justify the environmental cost. Instead of automating every process, AI should be strategically applied where it delivers the highest value with minimal environmental trade-offs.
3. Efficiency by Design
Sustainability should be embedded in AI systems from the design phase rather than being an afterthought. This includes optimising algorithms for energy efficiency, reducing model size where feasible, leveraging low-power hardware, and utilising techniques such as federated learning or pruning models to reduce computational overhead.
4. Data Efficiency & Minimisation
The AI ecosystem must emphasise data efficiency by reducing reliance on massive datasets and instead leveraging techniques such as transfer learning, synthetic data generation, and self-supervised learning to lower data acquisition and storage energy costs.
5. Prioritising Renewable Energy & Green Infrastructure
AI sustainability efforts should prioritise the use of renewable energy sources to power data centres and AI infrastructure. Cloud providers and enterprises should commit to green energy procurement, while policymakers should incentivise renewable energy adoption for AI computing facilities.
6. Adaptive & Circular AI Systems
Sustainable AI should encourage the reuse and adaptation of models rather than continuous retraining from scratch. Legacy models should be repurposed and datasets should be shared responsibly to prevent redundant efforts and unnecessary emissions from training new models.
7. Life-cycle Approach
Sustainability needs to be considered across the life cycle. From planning and development to operations, end-of-life, and risk management ecosystem players must prioritise sustainability at every stage. For example, in the model innovation layer, this includes ensuring efficient model training and development (planning & development stage), committing to efficient model inference (operations stage), and retiring inefficient legacy models and re-using training data (end-of-life stage)
8. Accountability & Transparency
Clear sustainability reporting and disclosure on AI energy consumption, carbon footprint, and efficiency metrics should be standard across the industry. This includes third-party audits, standardised carbon measurement frameworks, and public commitments toward reducing AI’s environmental impact.
9. Sustainable AI Governance & Regulation
Governments and regulators should introduce sustainability-linked AI policies, such as carbon footprint thresholds for AI training and incentives for low-energy applications. The regulatory ecosystem should also ensure that AI sustainability aligns with broader environmental, social, and governance (ESG) objectives.
10. Edge AI & Decentralisation
Whenever possible, AI inference should be performed on edge devices rather than centralised data centres. This reduces latency, minimises data transfer energy costs, distributes energy peaks, and lowers the carbon footprint of AI applications.
11. Resilience & Future-proofing
AI sustainability should include resilience strategies to anticipate and mitigate future challenges, such as climate-induced disruptions to computing infrastructure, geopolitical energy supply constraints, and material shortages for AI chips.
12. Close Collaboration
Sustainable AI will require close collaboration and coordination between the private and public sectors. The public sector should define guidelines and provide the right financing (and energy), while the private sector can engage in proactive self-regulation and collaborate with other stakeholders to establish these guidelines.
As the AI ecosystem evolves, the responsibility for sustainable solutions is shared by all players, from data centres to end users. The principles above are key to ensuring AI’s growth benefits both business and the planet.
At Access Partnership, we provide strategic guidance to help companies navigate sustainable AI, from compliance to long-term efficiency. Contact Abhineet Kaul today at [email protected] to learn how we can help you build innovative, sustainable AI initiatives.
