The conversation around datacentre growth and national initiatives to develop and deploy artificial intelligence (AI) models and applications has dominated the datacentre narrative in 2025.

It has concentrated on the physical size of the facilities, their substantial energy and water demand, and the inadequacy of the available energy infrastructure to support the proposed buildout. Sustainability seems to be an afterthought, if it is mentioned at all.

In fact, the expansion of digital infrastructure offers an opportunity to positively transform its energy and environmental footprint and significantly increase the deployment of carbon-free generation capacity across the global electricity grid.

Datacentre demands for more electricity generation capacity offers an opportunity to increase the deployment of carbon-free energy (CFE) generation assets and reduce the metric tons of CO2 per MWh of generation over time.

In the first phase of the buildout, from 2025 to 2030, wind, solar, and battery systems will be the majority of the new CFE generation. The impact of their intermittent production profiles can be mitigated by permitting their standby generation systems to support the electricity grid during periods of low output, avoiding the need to finance and build natural gas peaker plants.   

The second phase of the buildout will likely begin around 2030, as cost-effective nuclear, advanced geothermal, and other CFE generation assets become technically and economically viable.

These assets will take on a greater share of the generation load and begin augmenting and displacing natural gas generation assets. Many large operators are investing in and supporting demonstration projects and early deployment of these systems to validate the technical and economic viability of these developing technologies.

New facilities can take advantage of modular, high-efficiency cooling systems that maximise free cooling, minimise or eliminate water use in external heat-rejection systems, and deploy direct liquid and immersion cooling technologies to cool CPUs, GPUs, accelerators, or entire IT systems.

These technologies can reduce facility power usage effectiveness (PUE) to less than 1.3 and water usage effectiveness (WUE) to less than 0.4 liters per kWh. These PUE and WUE values are emerging as the maximum acceptable performance values for new facilities. These new, more efficient systems enable 80% or more of the energy consumed in the datacentre to power the IT infrastructure.

The embedded carbon content in equipment and materials will be a harder nut to crack, though this effort has become the centerpiece of many operators’ sustainability plans.

The production of key materials such as steel, concrete, optical fibers, and copper depends on fossil-fuel-driven, high-temperature manufacturing processes. Work is underway, supported by several large datacentre operators and others, to identify and develop alternative, low-carbon manufacturing processes, and limited volumes of material are available at premium prices. However, these efforts are unlikely to yield high-volume production processes for key materials before 2035. Even then, there will likely be significant variations in availability between regions.  

The industry has embraced the challenge of decarbonising the energy and material supply chains, but there is a dawning realisation that this effort will be a journey of several decades. However, the pieces are in place to drive continuous incremental improvements in water and energy use efficiency, reduce Scope 2 emissions associated with energy use, gradually reduce the carbon content of key materials, and increase the use of recycled materials and overall material-use efficiency for new datacentre facilities.

From a pragmatic standpoint, operators are understanding that true decarbonisation of operating facilities will not be achieved in the originally projected 2025-to-2035-time frame (aided by offsets). Instead, 80% decarbonisation by 2035 may be feasible in some geographies and at some facilities, but 95% to full decarbonisation will likely take until 2050 or beyond.  

The missing piece in the datacentre efficiency discussion is the efficiency of the IT infrastructure. There are multiple layers to this effort:

• The deployment of IT infrastructure equipment with high work per watt ratios and the capability to reduce power demand at low utilisation.
• The use of automated (AI or machine learning) power-aware IT workload systems that maximise IT infrastructure utilisation. These systems can minimise the quantity of equipment and increase the work per megawatt-hour (MWh) required to deliver a defined set of workloads, reducing the datacentre footprint.
• The coding of more efficient software that can minimise and optimize the IT infrastructure resources required to execute a specific application.

Datacentre operators need to embrace processes that optimise the performance and efficiency of the integrated hardware/software stack to reduce their energy demand and associated Scope 2 emissions. Avoiding energy use is an effective tool to manage power demand during planned growth; significant capital costs and supply chain stress can be avoided while reducing the operational strain on the electrical grid.

If properly executed, the planned global datacentre buildout can deliver a significant improvement in the sustainability and environmental performance of these facilities.

Integration of the facility and IT infrastructure, along with the deployment of highly efficient software applications, can deliver higher work delivered per unit of energy consumed while significantly reducing resource consumption.

Read more on Datacentre energy efficiency and green IT

• 
  

  Microsoft shares details of plans to reduce water usage of AI datacentres

  By: Caroline Donnelly

• 
  

  Microsoft trials building ‘wooden’ datacentres in pursuit of 2030 carbon-negative goal

  By: Caroline Donnelly

• 
  

  AWS expands datacentre hardware recycling programme to Dublin

  By: Caroline Donnelly

• 
  

  Singtel ropes in industry partners in datacentre push

  By: Aaron Tan