Manufacturers need to adopt sustainable AI practices to reduce environmental impact of data-driven operations

While other industries have raced ahead with the adoption of artificial intelligence (AI), it is fair to say that manufacturing has taken a more measured approach to this game-changing technology. 

A report published by Deloitte last year and a more recent study by ING both broadly argue that while there is some adoption of AI in manufacturing, it is still early days. Some people might interpret this as manufacturing failing to keep pace with change. I disagree. 

To me, this merely reflects a more considered response to the operational complexity, risk, and real-world consequences that come with deploying new technologies. We can argue about what those risks are, but for me, one of those concerns has to be the environmental consequence of unfettered progress. 

The role of manufacturers in developing sustainable AI

That’s why it's up to manufacturers and other heavy users of AI to play their part. It’s up to us collectively as an industry to harness the positive aspects of AI while minimising the emissions it causes.

In other words, to address the environmental impact of AI, there needs to be a shift towards ‘sustainable AI’ practices that balance innovation with efficiency.

One of the first things we can do is use techniques such as model distillation, which makes AI models smaller, faster, and more efficient without losing performance. Similarly, we can also employ compute-aware design to ensure that models are no larger or more complex than they need to be.

And it goes without saying that, where possible, manufacturers should also look at where and how their AI is being run, including opting for greener infrastructure, such as cloud providers powered by renewable energy.

Speaking to colleagues across the sector, this decision to source ‘green data’ is something that is gaining increasing interest among manufacturers looking to develop sustainable AI. 

This can also involve approaches like edge computing, where AI is deployed closer to where data is generated. By reducing the need to move large volumes of data between systems, this can help lower both latency and overall energy use.

And finally, we should only scale up to large models when the task really needs it. That way, we don’t use resource-heavy AI if a simpler solution works just as well. Together, these approaches form part of a much more sustainable approach to AI. 

Sustainable AI by design

It also shows how sustainable AI is not a single decision but a series of deliberate choices. From how models are designed to how workloads are managed – and even where systems are deployed – manufacturers have far more control over the environmental impact of AI than they might think.

In much the same way that we use sustainability as a guiding principle for manufacturing and innovation, we should also use a similar mindset regarding our approach to AI. As I’ve just described, that means rethinking conventional practices to develop innovative ways to reduce environmental impact. 

Just as we empower our engineering and design teams to come up with sustainable products and solutions, so too must we build awareness of AI’s environmental impact so that everyone understands not just how to use AI, but also the cost associated with it. 

This could involve, for example, the development of practical tools such as an AI energy calculator that is designed to measure and quantify the energy consumption of different applications. By making usage visible, manufacturers can identify areas of high demand and take steps to improve efficiency.

Alongside this, we also need to promote transparency and accountability. By improving how environmental impacts are measured and communicated, manufacturers can ensure that any claims made are accurate, consistent, and verifiable. 

Taken together, these efforts form part of a broader responsible AI approach, where sustainability is not treated as an afterthought, but built into how AI is understood, deployed, and governed across the organisation.

After all, AI is going to play an increasing role in manufacturing over the next decade, which is why that we – as an industry – must embrace the shift towards ‘sustainable AI’ practices that balance innovation with efficiency.

That means placing sustainability at the heart of AI decision making in areas such as governance, procurement, and project approval processes. By doing do – even at this relatively early stage – manufacturers have an opportunity to lead by example and show that innovation and environmental responsibility can move forward together.

After all, we’ve already made great gains in areas such as product design and production, where efficiency and sustainability go hand in hand. There is no reason that the same approach cannot also be applied to the sustainable use of AI.