AI’s true environmental legacy may be determined less by the energy it consumes, and more by how it reshapes and encourages innovation for sustainable products and processes across industries.
As scrutiny intensifies around the energy and water demands of AI data centres, the environmental debate has centred on infrastructure. Headlines track megawatts. Policymakers question resource use. Critics warn that AI’s sustainability claims risk becoming greenwashing.
AI is no longer confined to data centres. It is increasingly embedded in design tools, engineering workflows and industrial decision-making systems. In those environments, it does not just consume energy. It shapes choices. And those choices carry environmental consequences.
Moving beyond the generic sustainability prompt
I have been examining how AI systems operate inside real design workflows. A recent paper I co-authored explores whether structured AI tools can meaningfully expand sustainability considerations at the conceptual stage, when environmental impact is still fluid. The findings suggest they can.
Ask ChatGPT how to design a sustainable product, and you will get the right language. Use eco-friendly materials. Improve durability. Reduce waste. But the advice is often generic.
The study found that when designers used standard large language models (LLMs) to generate sustainable design guidance, the outputs were broad. Advice focused on high-level principles such as reducing waste or using eco-friendly materials. When the same design challenges were run through a novel two-stage retrieval-augmented generation (RAG) framework, the scope of sustainability considerations expanded significantly. The framework addressed an average of 2.7 times more product design specification areas than standard prompting.
Rather than producing general sustainability language, the system decomposed products into subsystems and linked each component to targeted sustainability strategies. Across case studies including wheelchairs, electric bicycles, kitchen appliances and packaging systems, the approach generated detailed guidance aligned with formal design specifications.
The implication extends beyond product design.
Time to look past life cycle assessments
By the time sustainability is measured, most of the impact is already locked in. Research has shown that up to 80% of a product’s environmental impact is determined at the design stage. By the time sustainability is measured through formal assessment tools, many of the most consequential decisions have already been made. Materials have been selected. Architecture fixed. Lifecycles constrained.
If AI is influencing those early decision-making processes across industries, its indirect environmental impact could be substantial.
Life Cycle Assessment remains a critical tool, but it is usually applied once product architecture and material choices are fixed. The opportunity to fundamentally reshape environmental impact sits earlier, at the conceptual stage.
Rather than auditing a finished design, the system collaborates with designers as they define subsystems and components, prompting structured consideration of modularity, repairability, durability and end-of-life planning from the outset. In testing, this approach significantly expanded the scope of sustainability factors addressed compared to standard AI use.
Sustainability should not be a checklist applied at the end of development. It needs to be embedded in the first design conversations. AI can help surface the right questions while there is still time to change direction.
The implication is clear. Sustainability needs to shape the blueprint, not just evaluate it.
Sustainability without the specialist barrier
Generative AI is already moving into factories and engineering departments. More than 100 companies in Europe and the United States are using industrial copilots to streamline programming and problem-solving. Analysts expect the majority of manufacturers to adopt similar tools within the next few years.
As these tools move upstream into design and production decisions, questions of responsibility and oversight become more pressing. That shift does not replace human expertise. Framing sustainability challenges, weighing trade-offs and defining long-term priorities remain fundamentally human judgements. What AI can do is accelerate how those decisions are explored, tested and operationalised at scale.
At the same time, AI is not environmentally neutral. Large-scale models require significant energy, computing infrastructure and materials. As AI becomes embedded across manufacturing, infrastructure planning and industrial systems, its own footprint will be visible and measurable.
The environmental impact of AI will therefore be shaped by two forces. The first is the resource intensity of the systems themselves, which is unlikely to change. The second is the cumulative effect of the decisions they help inform. The extent to which AI is leveraged across supply chains, product lifecycles and infrastructure planning offsets its operational footprint will be determined by its design, governance and scale of deployment. If not, it risks adding further strain to already stretched systems.
Less visible, but potentially more consequential, will be the cumulative effect of the sustainable decisions they help shape.
Sustainability cannot be a checklist applied at the end of development. It has to be embedded in the earliest conversations. AI has the potential to surface the right sustainability questions while there is still time to act, but only if it is designed and used responsibly.
The tipping point will not arrive automatically. Human expertise defines the direction and sets the boundaries. But as AI moves upstream into the decisions that shape the physical world, the environmental equation may begin to shift.
