It would be impossible to fix a city’s traffic congestion by widening a single busy highway. Yet today’s conversation around IT efficiency risks falling into the same trap, focusing on the most visible pressure point while overlooking the broader system at play.
Over the past year, headlines have zeroed in on the environmental impact of artificial intelligence. While these concerns are valid, this limited focus loses sight of the broader picture. In reality, the bulk of data centre energy demand does not come from AI-centric computing, but from the vast base of mainstream IT that underpins corporate IT estates.
In energy-constrained markets like South Africa, this broader view is essential. While the country has achieved more than 365 consecutive days without load shedding, a significant milestone that has stabilised the grid, that stability brings renewed pressure. As economic activity rebounds and digital transformation accelerates, the demand for electricity is rising again, placing fresh strain on an already overburdened system. This is underscored by a recent report from Eskom, which highlights strain across local distribution networks and regional transmission lines.
At the same time, government is pushing to expand digital infrastructure, classifying data centres as critical infrastructure alongside electricity and transport networks. Already, South Africa’s live data centre footprint carries a combined critical IT load of 390MW — enough to power a small city. With major new investments on the horizon, including a proposed data centre in Durban that could consume up to a quarter of the city’s electricity, the strain on energy supply is set to further intensify.
Putting AI into perspective
Concerns around AI’s contribution to this growing energy demand are valid and growing. As organisations scale their use of AI, the resource intensity of training and inference is rightly coming under scrutiny. This is especially relevant in South Africa, where data centre electricity consumption is projected to grow significantly, with per-capita usage expected to exceed 25 kWh by 2030, more than 15 times higher than the continental average.
However, this narrow lens risks overlooking the primary challenge. Research from the Uptime Institute shows that AI-centric computing currently represents only a relatively small share of overall data centre energy use today. Looking ahead, it is expected to account for somewhere between 30% to 35% of total demand by the end of the decade. However, that does not mean that the new AI infrastructure will replace traditional data centres. On the contrary, its power demands come on top of still growing consumption of traditional infrastructure, making it clear that the industry cannot afford to focus on one or the other. It needs to consider the broader equation.
End-to-end IT efficiency by design
For datacentre operators and IT decision-makers, this means efficiency cannot be treated as an afterthought. It needs to be built into every layer of the IT estate, from hardware architecture and cooling, to how workloads are scheduled and managed.
The question is where to begin?
A critical first step lies in creating headroom within the current power envelope. That means eliminating inefficiencies in existing IT infrastructure while embedding energy efficiency into new – traditional and AI – data centre deployments from the outset, as a design principle. The focus here should be on identifying opportunities for efficiencies at the IT hardware and software levels — where most of the power consumption occurs in data centres. The goal is to deliver more compute, storage and connectivity in exchange for the lowest input of energy.
From there, focus on equipment efficiency. Ageing infrastructure can draw significant power without delivering equivalent performance. Modern systems, even when more power-dense, are often far more efficient when measured in performance per watt. Consolidating workloads onto fewer, higher-performing assets, retiring redundant applications and equipment overhead, while rationalising data storage can reduce energy consumption, waste and improve output.
Additionally, cooling technologies can help further improve efficiencies. Traditional air-cooled environments are inherently energy-intensive, whereas hybrid approaches like adaptive cascade, leveraging direct liquid cooling for parts of the infrastructure, can dramatically reduce both environmental impact and operating costs. In fact, 100% fanless systems can cut cooling-related carbon emissions and utility costs by up to 90% for high performance computing systems.
Efficiency also depends on how well resources are orchestrated. By scheduling and managing workloads to optimise resource utilisation, running non-critical tasks at locations and times with higher renewable energy availability, and identifying idle periods for equipment to enter low-power modes, organisations can significantly reduce power consumption and minimise environmental impact.
Software is a critical enabler. Efficient code on optimised platforms reduces the resources required to perform the same tasks, while intelligent tooling can automate workload placement, dynamically adjust capacity and improve visibility.
Data efficiency is the final piece. Not all data needs to be collected, moved or stored in the same way. Keeping data closer to where it is generated and used helps reduce energy-intensive transfers, provided it is supported by low-latency infrastructure. More deliberate decisions around data collection, processing, consolidation, transfer, storage, insights harvesting, backup, retention, storage, and tiering can unlock meaningful efficiencies without sacrificing insight.
True IT efficiency goes beyond energy reduction, it’s a strategy for resilience, growth, and competitiveness, enabling tighter cost control, faster deployment, and confident scaling. But, as with that congested city, meaningful progress will not come from widening a single lane. It will come from rethinking how the entire system works. The IT leaders of tomorrow will have to look beyond the headlines — optimising not just AI, but the full breadth of mainstream compute that keeps the organisation moving.
- Jostein Birkeland, Principal Technologist, Sustainable Transformation, HPE
