Are data centres uncool? Questions over AI's ecological trade-off are growing louder but solutions may be emerging

Resistance to AI, till now confined to economics, is spilling over to the environment. Since the outset, AI has been identified as an economic disruptor that will result in large-scale job losses. But a new pushback is emerging over AI's sustainability frontier. Data centres are consuming country-scale electricity, and their computing intensity is exerting enormous pressure on freshwater for cooling needs.

Last year, residents of the executive-style, single-family subdivision of Annelise Park in Fayetteville, Georgia, were outraged when they discovered that the nearby data centre was siphoning nearly 30 mn US gallons (113.56 mn litres) of water - equivalent to 44 Olympic-size swimming pools and far exceeding the peak limit agreed to during the data centre planning process - for its cooling needs. This was after residents noticed that their water pressure had become unusually low. If this can happen in an affluent part of the US, one can be forgiven for being concerned about the way Indian authorities may go about with data centre infrastructure in a country facing extreme water stress due to overextraction of groundwater.

The hardware that runs AI infrastructure is built by exploiting critical minerals, and it generates increasing amounts of e-waste. A specific argument is made about the uneven distribution of costs and benefits of AI. Communities burdened with supplying specific natural resources need to be convinced about equally specific breakthroughs the tech is capable of in managing those resources.


That is the green paradox of AI. It can improve efficiency of electricity grids by predicting demand and integrating volatile RE sources. AI improves our ability to model climate and forecast floods and droughts, lower the water intensity of agriculture, and make water distribution smart. In critical minerals, AI has demonstrated capability of accelerating prospecting and mining. It also improves purity and yield. Robotic sorting and recovery makes e-waste recycling a breeze.

Beyond these specific counter-arguments to the issue of sustainability, AI offers an array of gifts to local communities that can dull opposition. Explaining AI's breakthroughs in the treatment of malaria can be a compelling message in towns around Zimbabwe's lithium mines. Precision livestock farming that AI enables could wear down some of Dublin's resistance to energy-hungry data centres. Predictive policing and digital forensics could persuade US states and counties to drop their moratoria against AI investments.

OECD reckons AI could raise annual labour productivity growth by 0.4-1.3 percentage points in G7 economies with high exposure to AI-led knowledge-intensive services and due to widespread AI adoption. Gains could be up to 50% less in other G7 countries due to their sectoral composition, and slow AI dispersal. Advanced economies cannot afford to miss their biggest productivity boost since their respective industrial revolutions.

Emerging economies will seek to speed up their relatively faster growth through AI. Developing economies will turn to tech-enabled solutions to attain human development metrics.

And the ecological trade-off may not be as severe as is being made out to be. Data centres profit by maximising the computational output per unit of power consumed. This aligns business objectives with energy efficiency goals. Modern data centres are already using closed cooling loops where the coolant never evaporates, eliminating water loss.

Transitioning to smaller, domain-specific models and optimised networks allows developers to achieve high computational power without building gigantic data centres. Tech surrounding AI infra is evolving rapidly, and negative externalities could be limited to acceptable standards.

Tech companies also have plans to take data centres far away from our backyards. SpaceX, which houses xAI, has plans to build them in space powered by uninterrupted solar energy - in keeping with Elon Musk's catchphrase, 'It's always sunny in space' - and cooled by infrared radiation of waste heat into vacuum. Sounds like science fiction but Musk argues there is no new physics involved. And investors are buying into his vision.

Other companies are pivoting to advanced nuclear and specialised RE. Small reactors and giant batteries are nearer our technology horizon than orbital data centres. Terrestrial solutions will face some form of pushback, because AI will be seen as a zero-sum resource game till benefits of the tech become widely available.

Distributing economic benefits of AI is a challenge to the global trade mechanism. Frontier AI models and massive data sets are concentrated in the US and China, and the rest of the world is dependent on their infrastructure. Export restrictions on critical inputs like semiconductors block the flow of hardware to economies that do not have the wherewithal to produce them. Multilateral initiatives are trying to ensure fair participation by countries locked out of the AI arms race.

Political resistance to AI infra is being driven by its concentration and resource strain. The way out would be to spread tech development across a broader geographical expanse to ease pressure on both counts. Assets need to be democratised, and countries allowed to build local capabilities instead of remaining passive consumers.

Fortunately, the competitive intensity of the AI race ensures open-source models become available close on the heels of expensive proprietary breakthroughs. Need for sovereign AI capability will aid dispersal and spread infrastructure demands among countries.

Widely shared costs is a more sustainable financial pathway for AI than the current investor frenzy over the tech. Consumers are more logical about what they pay than investors chasing good ideas among the bad.