Net-zero by 2050 seems further away than ever. Google and Microsoft admitted their emissions are rising from artificial intelligence (AI)-related datacentre activity; while in July, Irish datacentres were reported as accounting for 21% of the country’s metered electricity consumption. What can datacentres do to conclusively turn the tide?
Mark Acton, owner and principal analyst at Acton Consulting, says that the industry needs “quite simply” to face up to the need for change. Too much focus is deflected to factors such as power usage effectiveness (PUE), which is already at the point of diminishing returns in well-designed and well-run facilities.
“Datacentres themselves do not create the energy demand or consume the power. All they do is add an overhead to the delivery of power to the IT equipment they host,” says Acton. “Address the IT load.”
Datacentres typically have low utilisation, excess redundancy and software duplication, with none of that optimised for energy efficiency and environmental impact. However, that demand is driven by “our consumption of digital services and the types of platforms that we choose to use”, says Acton.
Behaviours and choices must change. However, change is typically based on information available – something lacking for the end consumer today.
Never mind Tesco’s clothing “passports”, there could be something like an energy passport tied to consumption, an introduction of new charging models, or a digital services consumption tax, Acton says.
After all, many digital services are currently “free” to use, with the end user unaware of the total cost, including impacts on emissions. What if instead internet use was billed by the hour, or images and uploads each came with a nominal fee?
“We could limit demand,” says Acton. “There could be so many better ways of doing things. Or perhaps an energy-labelling scheme for digital services could enable informed choices, or even a health warning: ‘Consumption of this service may damage your planet.’ That would have a far greater datacentre energy and environmental impact reduction than regulation or tinkering with PUE.”
Software must also be written that’s energy efficient to counter the “massive redundancy” of having loads of systems sitting ready to go just in case of need. A datacentre can be super-efficient, but if there are several of them, the benefit can evaporate.
“Calling that ‘efficient’ is bad accounting,” Acton says. “And yet large corporates are often not interested as it seems like a rounding error to them – that or it’s someone else’s budget.”
Max Schulze, director of green thinktank Leitmotiv Digital, adds that for a long time everybody thought of digital as essentially weightless, and now you have the problem of needing business models where you can make money without harming the environment.
“To date, I have not seen a company really that can do that. Even these phones that you can repair yourself, you still have to make the phone,” Schulze says.
We could also examine how and which kinds of tech continue to guzzle carbon and reconsider carbon taxes and emissions trading, as long as that does not simply allow organisations to avoid actual emissions reduction.
Digital innovation needs to make good economic sense too, especially in the wake of generative AI (GenAI). Avoid innovations where a solid business case remains elusive.
“We cannot subsidise these things in the hope that eventually they will solve all our problems,” says Schulze. “Because it’s highly unlikely, and what we’re essentially subsidising is a waste of energy.”
Regulation could be devised that increases the transparency of environmental costs across each digital service, empowering end user choices. Right now, Schulze says, you cannot choose an AI model based on environmental impact – the metric doesn’t exist. Tools typically only compete on features.
“A philosophical societal discussion is needed on digitisation. If you ask most people, I doubt they’ll say they need 30-second videos in an endless feed delivered on their phone,” Schulze says.
That means that investors need to be incentivising people to do things differently too, therefore calling for a rethink of market design.
“For investors, it would be really attractive if you would have very diverse, mature digital markets,” says Schulze. “We need to design economic systems in which sustainability in digital markets wins.”
Matt Watts, chief technology evangelist at storage vendor NetApp, agrees that we need more action that deeply embeds emissions-reduction efforts in all business aspects.
“The reality is servers consume about 15% of datacentre power – putting in more efficient storage makes a difference, but we’re talking single-digit percentage points,” he says.
Monitoring and measuring are still tough, so that’s where to start, to measure the scope of the problem. Then, break it down into levers.
“Another big thing is data waste,” Watts says. “Our data waste survey suggests about 41% of data is never reused, and analyst estimates have been higher.”
Watts says that making decisions on storage reduction is a “low-hanging piece of fruit” that could move emissions reduction into the double-digits in many organisations. Again, education is key, as change is currently slow, not least because different departments often do not own the problem themselves.
“Those who create data and run applications say it’s the IT team responsibility, while IT says they’re ‘only the custodians’,” he says. “That’s a problem we’ve sat on since datacentres were invented: a data management problem that no one owns.”
Matt Rees, chief technology and operating officer (CTOO) at connectivity provider Neos Networks, notes that IT by nature consumes lots of power, even as it gets smaller and more efficient. Additionally, industries are interconnected – reduction of emissions in one area or organisation can mean an increase elsewhere, so it’s tough to get clarity for an optimal approach.
“At the same time, ultimately, we are in a business of providing more bandwidth in answer to demand,” says Rees.
Certainly, we should be doing more with less wherever possible, which might mean rethinking around-the-clock availability for some organisations.
However, other strategies should continue to be applied, including incorporating renewables, and increasing energy efficiency whether by replacing legacy kit or making more use of efficient cooling, virtualisation and automation.
That is a sentiment with which Fred Lherault, field CTO for Europe, Middle East and Africa (EMEA) at Pure Storage, aligns. For instance, AI power consumption relates closely to GPU requirement “not necessarily designed with power efficiency in mind”, thereby “creating density problems for everyone”.
While doubling down on new tech may increase emissions at first, this should improve over time. “And some of that is about having access to the data in a timely manner,” he adds.
Lherault says some manufacturer data suggests average GPU utilisation of up to 35%. Organisations may be purchasing triple the amount of GPU needed, sometimes because they’re waiting for the data or some preparatory data-science work.
“We need to do the same thing with less resource, because you can’t get all the GPUs you want (so it’s linked to supply chain problems too),” Lherault notes.
John Booth, managing director and principal consultant at consultancy Carbon3IT, says much energy is wasted despite 15 years of code of conduct best practice and the like, with organisations still making the “most basic mistakes”.
“Also, many operators abdicate responsibility for servicing and cooling management to third parties. Often, these players are not datacentre-aware and do things more associated with shopping malls and offices, not precise control for datacentres,” Booth says.
Some accepted their service reports at face value, typically focused on typical business practices and outcomes, including in procurement – such as end-of-life, one-for-one replacement, rather than “big-picture” improvement. Also, many aren’t aware of current relevant standards or options in environmental monitoring.
“At times, they could have just eliminated their chiller completely, and thereby saved pounds in procurement and emissions,” Booth says. “Datacentre design communities are often still doing things they were doing in the 1950s and 1960s.”
Best practice has changed since then and new standards have emerged, but there’s a question over how many have updated their own operations, policies and practices. Separately, development of more ethical and green financing is among the levers available, says Booth.
Around 15-20% of UK energy is imported over interconnectors from Europe, and that money cannot be reinvested into a more efficient, renewables-based UK grid, he adds.
“Billions of pounds that we’re basically giving them to provide us with energy,” he says. “We’re getting energy fairly cheaply and don’t have to build UK infrastructure, but that’s a really short-term view.”