Opinion: Can businesses keep their sustainability promises in the age of AI?

Most ESG targets were set before AI reached the boardroom, explains Matt Johnson, senior director, Foundry at AHEAD. Keeping them will depend less on green credentials and more on being able to cut AI waste, a task the channel is well placed to help with

Hindsight, as they say, is 20:20: anyone can see perfectly what they should have done yesterday with the knowledge that they have today. And so it is with ESG targets and AI.

A few years ago, ESG was the ‘buzzword du jour’. Boards signed up to multi-year carbon targets, announced them with fanfare, and set out to achieve them. The problem is, no one setting those goals had figured on AI arriving as fast or hitting as hard as it has done. Fewer still had any idea of how much power and cooling it would need, and therefore how much energy it would consume.

But then AI did happen, and it happened at a pace not even the most gifted fortune teller could have predicted. Today’s models can do things that just weren’t possible last year, and every month, even every day, new use cases are emerging.

All of which demand more compute and more power, creating more distance from those ESG goals. So, can businesses keep the sustainability promises they made in the age of AI? Potentially, but not in the way they planned, certainly not by accident, and probably not without some help.

ESG: Out of date, not wrong

The ESG targets themselves aren’t necessarily wrong, but they’re built on assumptions from two or three years ago, so they need revisiting. Scrutiny of whether companies meet their stated sustainability objectives is only increasing, particularly in the UK and Europe, and quietly missing a target looks much worse than resetting it with a clear explanation.

The biggest threat to most carbon commitments isn’t necessarily AI itself; it’s wasted AI, which happens in three ways. Firstly, AI is being used for jobs that don’t need AI. Secondly, workloads are running on far more compute than they require. And thirdly, capacity sits idle once it’s bought. All of which draws power, needs cooling and takes up space.

The trouble with “AI first”

A lot of the waste starts in the boardroom with organisational FOMO. Under pressure to show progress, businesses declare themselves “AI first”, which means they go looking for problems for AI to solve. But that’s the wrong way round.

AI is a technology tool, albeit a very powerful one. But that doesn’t mean businesses should use it for everything; no one uses a chainsaw to cut paper when scissors will do. Many tasks now handled by AI models, such as moving data between systems or filling in forms, would run just as well on conventional automation, with less computing power.

Everyone’s an AI user now

Day-to-day use is another source of waste, and this is where AI differs a lot from previous infrastructure shifts. Virtualisation transformed the datacentre, but most users never noticed, apart from when their applications ran a bit faster. But AI isn’t behind the scenes; it’s on everyone’s desktop, and anyone with a browser can try it. And they are!

Individuals are using AI for everything from replying to emails to creating content to building agents. For organisations that pay for AI by usage through APIs, agents or cloud GPU time, costs can spike without anyone intending them to. And every one of those requests runs on a GPU somewhere, drawing power and needing cooling.

But people need to experiment and get hands-on with AI because it’s the best way for them to lose their fear of the technology and find genuinely useful business applications for it. Banning it isn’t the answer; instead, businesses and channel players need to train teams on how the models work and what each one is good at, so people pick the right model or a different tool for the job.

Start with the workload

The “AI first” mistake also has a hardware equivalent: buying the infrastructure before working out what it’s for. This is essentially just guessing. Guess too big, and the business has to power and cool capacity it’s not using. Guess too small, and it ends up buying twice. Either way, the business pays in missed ESG targets, financially or both.

Businesses and channel partners need to start where they always have, with what the organisation is trying to achieve, then test the plan before committing to it. A digital twin of a datacentre, for example, lets a customer see what a new rack layout does to airflow, heat and cooling before a single unit ships. That shows how much extra power and cooling more AI will really need, so the business can weigh it against its targets before it commits, not after.

The AI capacity gamble

The third kind of waste is capacity that doesn’t get used. One side of the argument says buying ahead is worth it because GPU costs are rising at a rate of knots, with everyone in the game raising prices almost monthly.

But buying ahead only pays off if the hardware gets used, and a lot of it doesn’t. Data from cloud optimisation firm Cast AI, covering around 23,000 Kubernetes clusters, put average GPU utilisation at just five per cent. Demand for AI arrives in bursts, kit gets ring-fenced for one team while others queue, and once capacity is in place, it rarely gets scaled back.

That waste costs money and makes sustainability targets harder to meet. Idle hardware still draws power, needs cooling and fills floor space. Worse, a poorly used estate can look full, prompting another purchase and another round of unnecessary spending and consumption.

Instead, organisations need to buy for demand they can actually see, and find out how hard the existing tech is working first. Utilisation isn’t glamorous, but it’s the cheapest capacity a business will ever find.

Back to the day job

This is where the channel plays a crucial, if familiar, role. While AI is changing the game, customers still need the fundamental service channel partners have always offered: understanding how the business works, what it’s trying to achieve and what it already has.

Of course, a big infrastructure quote is always tempting, but a partner whose AI proposition reads like a GPU price list is just one price comparison away from losing the deal. A partner who knows a customer well enough to say “you don’t need that yet” is worth far more than one who can only say yes.

Future gazing

Even with a crystal ball, anyone shaping ESG targets three years ago couldn’t have predicted what AI would become, and nobody can say with any confidence where it will be in three years either.

But what businesses can control is how much of what they run today is wasted: AI deployed only where it’s really needed, infrastructure that’s sized for the job, and capacity that actually gets used.

The targets are still going to be tough to hit, but at least by managing waste today, organisations can put them back within reach without worrying about trying to predict what’s going to happen tomorrow.