Artificial intelligence (AI) and cloud computing are changing the way data centers operate. This convergence is redefining the role of data centers within an interdependent and constrained infrastructure. This is happening against the backdrop of a global energy crisis, with electricity demand rising, networks becoming constrained, and sustainability legislation tightening. Digital infrastructure can no longer operate in isolation.

The AI adoption

We should first define what “AI” means in this context. Rather than true autonomous intelligence, it currently manifests as optimisation algorithms, predictive modelling, and automated control. While these tools excel at forecasting demand and streamlining efficiency, they are not truly independent. Instead, they function as bounded agents limited by their underlying system design, specific objectives, and the data available to them. Consequently, the hurdles we face are systemic and architectural, not merely computational.

Furthermore, integrating AI is complicated by its significant, fluctuating energy demands, which often clash with the low-latency, high-bandwidth requirements of 5G and immersive services. True latency is determined by the entire network chain, from radio access and transport to the core network. Ultimately, achieving technical standards for speed depends less on where data is generated and more on the physical proximity of computing power to the end user.

Shift to distributed architectures

This change in requirements has led to the adoption of distributed architectures. This approach combines regional and edge data centers with hyperscale cloud infrastructure. By moving computing closer to the user, latency is reduced, enabling applications such as augmented reality, gaming, and industrial control.

Naturally, there are trade-offs to this approach. By adopting a distributed architecture, the local energy demand increases. There is also the issue of increasing bandwidth demand. Multi-gigabit speeds are possible with 5G, placing greater pressure on data centers interconnects and backhaul networks. Rather than treating network and compute as separate layers, it is imperative that they be optimised together.

Telecoms pressures

Evolution continues in telecoms, as operators expand their tower, small cell, and distributed antenna systems. As wireless, fibre, and edge infrastructure converge, there is increased demand for site-specific energy.

These pressures are particularly pronounced in parts of Asia, where rapid growth is driving demand for data centers expansion and 5G deployment. The concern for operators is that infrastructure is normally based in urban environments with limited grid capacity. That said, combining fossil fuels with fast-growing renewables used in energy systems creates low-latency services without being trapped in high-carbon infrastructure.

The shift to local energy constraints

There is a fundamental shift happening in the energy landscape. Data centers consume a significant amount of power, and as the demand for data centers grows, so does the demand for energy. The lack of adequate fossil-fuel backup systems has been exposed by recent energy crises. Increasing costs, tightening regulations and emission targets are all added to the burden.

More complexity has also been added to the pile with the rollout of 5G. Localised demand hotspots can become an issue at edge data centers and telecom sites. Even if a nation has an adequate total power supply, these clustered areas can overwhelm the national grid. As a result, the constraint is increasingly shaped by local network capacity.

Measuring efficiency

The gold standard of measuring efficiency has typically been power usage effectiveness (PUE); however, with new hyperscale sites, it fails to capture the full picture. PUE does not capture utilisation, and low utilisation typically leads to significant energy waste.

PUE is not broad enough in the current data centers climate. It also fails to account for network congestion or for local grid impacts. PUE still has its uses at a facility level; however, it is crucial that the industry recognises that it is not sufficient for measuring the efficiency of a modern data centers alone.

Edge computing: balancing performance and efficiency

Edge computing introduces new layers of complexity to efficiency metrics. Unlike centralised hubs, smaller edge sites often struggle with higher PUEs and unpredictable utilisation, alternating between being idle and overwhelmed.

However, a higher local PUE is not necessarily a failure; if a site reduces overall network strain and latency, it may yield a superior system-wide outcome. Therefore, efficiency cannot be measured in isolation; it requires a holistic view of the entire architecture.

The multi-objective optimisation challenge

Managing the edge is a delicate balancing act where improving one metric often compromises another. To solve this, developers are turning to AI-driven optimisation to manage:

• Latency vs bandwidth: prioritising speed while managing data volume,
• Energy vs utilisation: balancing power consumption with server demand,
• Dynamic allocation: using machine learning to shift workloads, automate cooling, and react to real-time energy price signals.

Emerging systemic risks

Data centers have evolved from passive loads into active participants in the power grid. This shift introduces risks to grid stability, especially in systems with low inertia, where rapid demand fluctuations can be disruptive.

To mitigate these risks, we must adopt whole-system thinking. This involves using digital twins and integrated planning to synchronise the needs of energy, telecommunications, and compute.

The landscape beyond 2030

As we look towards the next decade, the relationship between data centers and the world around them will undergo a fundamental shift:

• Market evolution: data centers will become flexible assets in energy markets, where their physical locations and load-shifting capabilities become valuable commodities,
• Orchestration: compute will move away from static, fixed provisioning toward a model of real-time orchestration governed by environmental and physical constraints,
• Community assets: there is a significant opportunity for data centers and 5G networks to act as “community energy hubs,” integrating renewables and storage to support low-carbon grids.

Data centers are transitioning into system-aware infrastructure. The defining challenge of the next era is harmonising digital demand with the physical limits of our planet. Success lies in treating AI, energy, cloud, and communications as a single, unified ecosystem.

By Aoife Foley, Professor and Chair in Net Zero, The University of Manchester