AI and Sustainability
Reducing the Hidden Costs
The Invisible Footprint of AI
Artificial intelligence has become synonymous with innovation but behind every model, from generative art tools to industrial systems, lies a growing environmental cost.
Training a single large language model can consume as much electricity as hundreds of UK households use in a year, with additional demands for cooling, data transmission, and cloud storage.
For small and medium-sized enterprises (SMEs), and especially for those in the heritage and creative sectors, this raises a crucial question: how do we embrace AI without inheriting its hidden energy burden?
Sustainability in AI isn’t only about carbon, it’s about efficiency, accessibility, and responsibility. The challenge is to design systems that perform well not just technically, but ethically and environmentally.
Why Energy Efficiency Matters
AI’s energy use is a by-product of scale. The larger and more generalised a model, the more computational resources it needs, often running on remote data centres powered by non-renewable energy.
But the majority of creative or analytical tasks do not require planetary-scale models. Whether restoring a digital artefact or analysing construction data, many applications can be achieved with smaller, task-specific systems that run locally or on low-power infrastructure.
For SMEs, adopting leaner AI means:
Lower energy consumption and hosting costs
Reduced latency and greater data security
Improved environmental reporting (a growing requirement in public sector procurement)
More control over performance and sustainability metrics
In short: less data, smarter use.
The Case for Hybrid and Frugal AI
At Aralia Systems, we’ve seen firsthand how hybrid AI, combining machine learning with classical analytical methods, can reduce both computation and uncertainty.
By embedding prior knowledge (physics, geometry, or domain-specific constraints) into the model, you dramatically cut the number of iterations and energy cycles required to reach accurate results.
This approach, sometimes described as “frugal AI,” doesn’t compromise quality. Instead, it focuses computation where it’s most needed, eliminating wasteful over-processing.
In practical terms, this means:
Using Physics-Informed Neural Networks (PINNs) or smaller transformers fine-tuned on domain data
Running inference locally or on edge devices rather than energy-intensive cloud systems
Caching intermediate results to avoid repetitive computation
Prioritising explainability over raw model size
These steps turn sustainability from a compliance exercise into a competitive advantage.
Sustainable AI in Heritage and the Creative Industries
Heritage and creative organisations are uniquely positioned to lead by example.
Cultural projects increasingly rely on 3D scanning, digital reconstruction, and generative media, all of which can benefit from frugal, domain-aware AI.
A museum digitising artefacts or an archive reconstructing lost heritage needs accuracy, transparency, and repeatability, qualities achievable with smaller, better-targeted tools.
Sustainable AI design also supports funding eligibility, as many Innovate UK and Horizon Europe programmes now include environmental impact criteria. SMEs that can demonstrate reduced compute intensity stand to gain both reputational and financial advantages.
Designing for Longevity
Sustainability isn’t just about the energy used to train a model, but the entire lifecycle: deployment, maintenance, and eventual retirement.
When SMEs choose open, modular architectures and interoperable data formats, they extend system lifespan and avoid the need for costly retraining or redeployment.
Think of it as the circular economy of AI. Reuse knowledge, reduce waste, and design for adaptation.
By making sustainability a design principle, not an afterthought, the creative and heritage sectors can prove that innovation and responsibility are not opposing forces, but two sides of the same intelligent system.
Final Thought
The future of AI doesn’t depend on who can build the biggest model, but who can build the most efficient one.
For SMEs, the path forward lies in hybrid, transparent, and sustainable systems, ones that respect both data and the environment.
In an era where compute power equals carbon, choosing smarter algorithms is not just good engineering. It’s good ethics.
