NTT Data: How AI Energy Use Threatens Net Zero Goals

It is no secret that the AI boom is creating an environmental crisis that could derail global climate commitments. 

As a result, while companies strive to deploy ever more powerful systems, the energy demands of training and running AI models are spiralling upwards, raising serious questions about whether the technology sector can meet its decarbonisation targets.

In response, NTT Data, a technology services provider, has released a report calling for fundamental changes to how AI systems are designed and operated. 

The whitepaper, titled ‘Sustainable AI for a Greener Tomorrow’, projects that AI workloads will account for more than half of all data centre power consumption by as soon as 2028. 

At that point, the technology could consume as much electricity annually as 22% of all US households.

The impact of training AI models

The strain on resources extends well beyond electricity. 

Training a large AI model, which involves processing vast datasets to teach the system to recognise patterns and make predictions, can require millions of litres of fresh water for data centre cooling systems. 

Running between 10 and 50 queries can consume up to 500ml of water.

“The resource consequences of AI’s rapid growth and adoption are daunting, but the technology also can empower innovative solutions to the environmental problems it creates,” says David Costa, Chief Sustainability Business Officer at NTT Data.

The report identifies four key metrics for measuring AI’s ecological footprint: 

• Energy demand
• Global warming potential
• Water consumption 
• Abiotic resource depletion 

The fourth refers to the exhaustion of non-renewable materials such as minerals and metals. 

On emissions, data centre carbon footprints are expected to more than double by 2030, reaching approximately 860 million tons of carbon dioxide equivalent.

Why hardware replacement cycles drain mineral reserves

Digital user devices currently drive 9.4% of global cobalt production and 8.9% of palladium output, pushed by short lifecycles and frequent replacement cycles. 

Data centres compound this problem by consuming vast quantities of copper, aluminium and rare earth elements, with servers typically replaced every few years to meet performance demands.

The report criticises the AI industry’s historical focus on performance metrics such as accuracy and speed at the expense of efficiency considerations. 

Some modern AI models consume over 300,000 times more computational power than their predecessors, creating what the report describes as an increasingly exclusive domain accessible only to organisations with resources to sustain the energy demands.

Inside NTT Data’s tsuzumi model with lower energy needs

NTT Data argues that addressing AI’s environmental impact requires coordinated action across the entire technology ecosystem – from hardware manufacturers and data centre operators – to software developers, cloud providers and policymakers.

The report recommends several interventions, including running AI workloads in locations and at times aligned with renewable energy availability, applying green software engineering patterns, which are coding practices designed to reduce energy consumption – and prioritising modular, upgradeable hardware components to reduce electronic waste.

“AI’s amazing capabilities can help manage energy grids more efficiently, reduce overall emissions, model environmental risks and improve water conservation,” David explains.

The company has introduced its own initiatives to tackle the problem, including remote graphics processing units (GPU) services that shift AI workloads to energy-optimised locations. 

GPU’s are specialised chips that perform the parallel calculations required for AI processing. 

NTT Data has also developed the tsuzumi large language model (LLM) that the company claims requires 250 to 300 times less energy for training than conventional models.

The challenges ahead

However, the report acknowledges that significant barriers remain, including fragmented assessments, inconsistent metrics and a lack of standardised reporting frameworks comparable to those in traditional industries. 

Additionally, many organisations focus narrowly on energy or emissions without considering water usage, rare material depletion and electronic waste comprehensively, according to the analysis.

The report calls for industry-wide adoption of lifecycle thinking and circular economy principles as essential prerequisites for sustainable AI development.

“It’s vital for organisations to recognise the challenge and build sustainability into AI systems from the start,” David adds.