How should we balance sustainability and AI as project managers?

Physicist and programme manager James Lea gave one of the most talked-about sessions at the 2025 APM Conference. His topic? Data centres and energy consumption.

Lea said that “AI is nothing new – it has been around since the 1980s – but what we are seeing today is something quite different." Lea spoke about Nassim Nicholas Taleb’s concept that if something isn’t harmed by change, it can actually be made stronger.

“I think that is largely what we are doing here [with AI],” he said. “Every project is an experiment that tests the hypothesis ‘we know how to do this’.”

What is project science?

Project science is a combination of approaches that looks at people and all the complex change paths we go through – the maths, algorithms, heuristics and data in one huge melting pot, explained Lea.

“If you combine all of that, you can get into a tremendous sweet spot, where you can design the scope of a project, and lead, control and close out projects, programmes and portfolios with real confidence,” he said.

The impact of data centres

The impact of data centres in phenomenal, explained Lea.

“We are seeing demand worldwide for additional nuclear power, and potentially for small nuclear reactors and all sorts of grid connections, to run data centres. Those data centres are powering modern life.”

He quoted a public policy director from a data centre operator, who said that there are three main myths about data centres: that the internet runs by magic; data centres do nothing useful; and they can be built anywhere.

“In our project lives we are using data centres and AI all the time without realising it, and increasingly so,” Lea explained. “There is an explosive growth in the use of data centres and their consumption in gigawatt hours. Data centres are being constructed all over the place.”

Why is there this demand, asked Lea? Because of computation, which is nothing more than algorithms and data, running through data centres, which takes energy.

Project data

“There is so much you can do with project data to uncover patterns and accelerate project delivery,” said Lea. “There is a staggering amount of data, and the algorithms we use have also been growing over time – there has been an explosive growth in computing and the clever stuff that can go on.

“What has really changed, and what is really driving the whole conversation, is the prevalence of machine learning and AI. That is producing a staggering demand for data and our use is growing exponentially.”

He pointed out that AI in projects covers everything thing from generative AI used for video and autocoding; artificial neural nets used for meeting summaries and planning; object detection for site diaries; linear regression used for estimating; and decision trees, used for causal analysis and expert systems. All of these take a lot of energy to run.

Things are heating up

“When you ask an LLM [large language model] a question it is creating real heat in a data centre somewhere around the world,” Lea explained. It can take 700W of power and 460,000 hours to train an LLM, which produces 130 tonnes of CO2. “This is a lot of power and a lot of CO2 being produced,” said Lea.

“So many data centres are being constructed because of our insatiable demand for AI… and it gets worse,” he said. AI tools now start running every time you open your browser and that is generating more CO2, as well as advertising trackers running in the background.

“I disable all the trackers to reduce my environmental impact,” said Lea.

He pointed out that ‘shadow AI’ is being used on projects all the time – where people won’t use corporate systems but will go straight to Gemini or ChatGPT to ask questions.

A balancing act

How can we balance AI with sustainability in a practical sense?

“It’s quite simple – we need to understand in our projects and organisations what systems we are relying upon. Talk to your IT teams. What are we paying for when we buy these services in?” Lea urged.

“The second thing, which is even more fundamental, is to run projects efficiently,” he explained. “Give people the answers they need because then they don’t have to search and use energy-intensive AI.

“As project managers, we should know about the resources that are being used to deliver the change that we are managing. You can challenge your cloud providers to ask them for more information, and your LLM too. Make sure they are being responsible,” said Lea.

You could also try using a power meter to plug into your equipment to measure your energy consumption, and ask your IT providers at work about what the power consumption looks like.

Efficient AI – practical steps you can take:

Say how you will use AI in your project management plans to do projects faster and better – plan it, think about it and manage it across the team. Encourage innovation but encourage sharing results too.
Everyone should ask: is this use of AI helping them to deliver sooner, and to make the project more efficient? If you are, you will reduce the environmental impact of your project as a whole.
Understand algorithmic complexity – essentially how much heat does your computer burn up solving a question. There can be a surprising variety.
Use the simplest algorithms that will do the job. Try and use the right tools to solve a problem.
Cache and reuse insights. Create an FAQ that people can use.

“If we do all of these things, we can improve sustainable outcomes [for projects],” said Lea. We can improve project failure rates, which will have a huge benefit – we can see further ahead and minimise uncertainty over time. We will also get much better at estimating and at learning from the past to predict and shape the future of projects, he argued.

"I also think we can build a circular economy with data. If we can share our data from one project to another that’s a huge opportunity to improve,” said Lea.

Actions you can take today

Take control of data, algorithms and models: own it, rate it
Project plans: state how you will use AI
Recognise how algorithms are being used
Build efficient and responsible AI
Collaborate and build (see APM’s AI and Data Analytics Interest Network and the Project Data Analytics Taskforce)