Can green data centres solve the AI carbon conundrum?

As far back as 2018, in the years BC (Before Covid), New Zealand had already begun to acknowledge the economic potential of artificial intelligence with the establishment of a national AI Forum. Bringing together industry, government and academia, the forum recognised AI’s potential for positive disruption, and New Zealand’s – and in particular Auckland’s – potential to be at the forefront of this.

Fast forward through the pandemic and the AI conversation has self-generated to focus on Large Language Models like GPT and BARD, bringing AI chat into the mainstream. For Auckland, already renowned for its start-up ecosystem, harnessing the possibilities provided by these models will be fundamental to realising its leadership vision of developing cutting-edge companies producing and exploiting AI tech by 2030.

But these gains in the virtual space require processing power in the physical, which is driving up demand for data centres in and around Auckland.

Beyond meeting our own needs, given our cooler climate, renewable energy sources and improving connectivity (such as through the Datalink Hawaiki cable), areas of wider New Zealand have been identified as strong potential locations for “green” data centres to service overseas demand. However, supporting the vast data streams they feed on and capitalising on the growth in demand through expanded data centre infrastructure must not come at the expense of us sticking to our 1.5C climate targets – particularly given our slow progress to date against the Paris Climate Agreement, considered “Highly Insufficient” by climateactiontracker.org.

David Lee is Project Director and head of operations at JLL NZ.

Data centres already account for 2 per cent of global energy consumption, and it’s estimated that this figure will rise four-fold over the next decade. Here in NZ, Transpower recently advised that they were projecting to fall below the 200MW buffer that it aims to maintain due to cold weather, leaving significant questions as to how we might meet the forecast power requirements of data centres, which are set to grow from 81MW last year to 303MW by 2030.

While the most modern “sustainable cloud” centres have been built with our climate goals firmly in mind, with no regulation in place there still remains a strong element of uncertainty and opaqueness around data centre-related emissions. Indeed, globally speaking, less than a third have energy utilisation visibility that could help drive transformational change. The suspicion, therefore, is that there is significant work to be done. But the good news is that as the technology develops, so too does our ability to mitigate its environmental impact.

Data centres are already becoming much more efficient, with recent studies showing data storage has risen 25 times with only a three-fold increase in energy use. To capitalise on these advancements, the most obvious solutions lie in the adoption of the circular principles of reduce, re-use and recycle – something that the AI tech itself can enhance.

As cooling accounts for around 40 per cent of a data centre’s total energy use, airflow management can make a significant difference. Google has already implemented a fully automated AI-controlled cooling technology that helps them save up to 40 per cent of their cooling system’s total energy use. But this should be just the start.

With data centres located in or close to urban areas, AI can help with heat, cooling and water recovery and re-use that can provide a wider community benefit – for example providing heating or cooling to surrounding homes or businesses, or servicing rooftop gardens and greenhouses to avert waste and rainwater from sewers.

We humans also have a key role to play – and this starts by acknowledging that there is a very real trade-off at play here as we feed the power-hungry Large Language Models to deliver sustainable solutions. We should view these models as partners rather than panacea, and by encouraging new avenues of research that don’t solely depend on developing larger and more complex models, we can ultimately promote more socially responsible and “greener” AI.

While we may consider data centres as third-party black boxes, the reality is that most are colocation facilities supporting a variety of businesses. Therefore, our business leaders and their own stakeholders can play a significant role in shaping the future of the sector through demanding alignment with their own ESG expectations. This can be through seeking greater visibility over energy use, actively encouraging a transition to green or renewable energy sources for back-up power and ensuring the centre has a plan to identify and act on opportunities to enhance efficiencies.

In Europe the data centre industry is confronting the challenge head on with the Climate Neutral Data Centre Pact, through which more than 100 companies and industry associations have pledged to ensure their data centres are climate neutral by 2030. Government also has a role to play. With NABERS energy targets set to be mandated for commercial buildings here from next year, why not consider extending this legislation to cover data centres?

The evolution of AI has created a uniquely modern conundrum in that while it can massively accelerate our understanding, management and preparedness for the worst impacts of climate change, without a proper mitigation strategy, its enormous appetite for data-processing power could help us further contribute to it in the short term.

It’s therefore vital that we consider the social and environmental risks, costs and benefits alongside the technological. For the next generation of data centres, enabling this begins with good real estate decisions, to ensure they are strategically and sympathetically located and built in a way that considers embodied and operational carbon throughout the entire building lifecycle.