The sustainability conversation in the data centre industry has developed over several years. Operators have focused on improving efficiency through better Power Usage Effectiveness (PUE), increased use of renewable energy, and efforts to reduce carbon impact.
The rise of AI infrastructure has added a different layer to this. Demand now looks very different, and many of the assumptions behind earlier green data centre designs no longer hold in the same way.
Attention is no longer limited to optimising facilities. It now extends to how they are designed, operated, and assessed when workloads remain high for long periods. As a result, the importance of sustainable AI has increased.
What AI Does to Energy Consumption
Traditional enterprise workloads tend to follow a pattern. They rise and fall, with periods of lower activity that give infrastructure some flexibility. AI workloads behave differently. They tend to run continuously at high density, placing constant pressure on power and cooling systems.
The scale difference is noticeable. A standard server rack may draw between 5 and 10 kilowatts, while high-density AI infrastructure racks can exceed 100 kilowatts. It’s not only a matter of volume, but also how energy is drawn and used across the facility.
This has implications for AI carbon footprint reduction. As density increases, energy use per square metre rises and cooling systems work harder to maintain stable conditions. Under these conditions, maintaining efficiency metrics such as PUE becomes more difficult.
Energy-efficient AI data centre design increasingly focuses on how systems perform under continuous load rather than under ideal or variable conditions.
Why the Existing Sustainability Approach Needs to Evolve
Many energy-efficient AI data centres in operation today were designed for a different type of workload. Earlier environments assumed variation in demand, with some level of relief built into how systems operated.
Retrofitting these facilities can improve performance, but only to a certain extent.
Adjustments to airflow, cooling behaviour, and power distribution still help, though their impact reduces when systems are already operating close to full capacity. With sustained load, infrastructure has less room to adjust. Once a facility is operational, it’s difficult to rely on reactive improvements. So, more such decisions are made during the design stage.
Newer facilities are being planned with this in mind. Cooling systems are sized closer to peak demand, power systems are expected to remain stable, and energy sourcing is considered earlier in the design process.
For operators running a colocation data centre, customer expectations also influence these decisions. Enterprises are working toward both performance and sustainability goals, and providers are expected to support both at the same time.
How Design Is Responding
Cooling is usually where the difference becomes visible first. As density increases, heat builds up more quickly, and air-based systems on their own become harder to rely on.
Liquid cooling is being used more often in these environments. Some setups use direct-to-chip cooling, others rely on rear-door systems, and in certain cases, immersion is used, depending on the workload.
Changes in cooling tend to affect the rest of the system as well. Layouts are reworked, operations become more specialised, and power and water usage begin to follow actual demand more closely.
Energy sourcing is also being addressed earlier. Organisations working toward sustainability targets are factoring this into infrastructure planning from the start. Supporting Powering AI, sustainably, becomes part of that process.
In some newer builds, this is already visible. STT GDC India, for example, is working with higher rack densities while adapting cooling approaches based on workload demand rather than relying on a single method.
Greater use of renewable energy and long-term carbon reduction targets are also becoming part of how these facilities are planned.
The Role of Operations in Sustaining Performance
Design plays a role in how a facility performs, but it is not the only factor. Day-to-day operations influence how efficiently systems continue to run.
AI workloads don’t stay fixed. Demand can increase, conditions change, and systems need to adjust along the way. Without regular monitoring, performance can begin to slip.
How closely systems are tracked and how quickly adjustments are made make a difference. Metrics such as PUE, cooling behaviour, and power usage provide useful indicators, but they matter only when action follows.
At STT GDC India, improvements in carbon efficiency have come from ongoing operational management rather than a single upgrade. Over time, this reflects how consistent system management supports sustainable performance.
Where This Is Heading
Sustainability and performance were once treated separately. That distinction is becoming less clear.
With workloads staying high for longer periods, energy consumption and cooling begin to influence what infrastructure can handle. These factors move from the background into core design considerations.
Newer data centre facilities are being planned with this in mind. Design starts closer to sustained demand rather than average utilisation. Cooling systems are aligned more closely with actual load, and energy sourcing is considered earlier.
As demand continues to increase, differences between facilities become more noticeable. Some continue to perform reliably, while others begin to reach their limits. Addressing those limitations later is not always straightforward.
The link between performance, efficiency, and long-term sustainability becomes clearer over time.
Comments
Log in or sign up to join the conversation.