Why the AI revolution could be heading for a power shortage

Artificial intelligence is changing how the global economy uses energy. This is not a marginal increase in demand but a structural shift. AI workloads require more power than traditional cloud computing, with demand rising quickly.

 

The International Energy Agency projects that data centre electricity demand could more than double by 2030, reaching around 945 terawatt hours. The IEA remains a government-backed body. Its projections carry weight, but they should be read in context. They reflect both technical analysis and the policy priorities of its member states, and depend on assumptions about how quickly AI demand scales.

 

This is not just about building more data centres. It is about whether energy systems are ready for a new type of demand that is intensive, variable and increasingly strategic.

 

A more uncertain global context

 

The wider context matters. The war in Ukraine exposed how fragile energy systems can be. Gas supply shocks translated quickly into electricity price spikes across Europe.

 

At the same time, global relations are shifting. The United States and China are competing for leadership in AI, chips and cloud infrastructure. Europe is seeking to strengthen digital sovereignty while meeting climate targets. Supply chains for semiconductors and critical materials are under pressure. Energy, data and infrastructure are now closely linked. Decisions about digital infrastructure are no longer just technical; they are strategic.

 

Investment and the risk of mispricing

 

There is a risk in how quickly investment is moving. AI is attracting large amounts of capital, often based on expectations of sustained growth. That growth may materialise; it may not. We have seen this before in technology cycles. There is a tendency to scale first and assess later, which carries risk.

 

What is less visible is how underlying costs are treated. Grid reinforcement, network upgrades and in some cases gas capacity are not always fully reflected in investment decisions. These costs are often recovered through tariffs and system charges paid by users, which creates a disconnect. The drivers of demand are not always directly exposed to the full cost of enabling that demand. Financial models may therefore underestimate long-term system impacts.

 

This is a system problem

 

The discussion often focuses on chips, cooling and power supply, yet there is a wider system problem. Energy demand from digital infrastructure is shaped by software, data, networks and how systems are run. Human expertise remains critical. AI is not autonomous intelligence; it is optimisation at scale, bounded by its design and by the data it uses.

 

Yet many systems still rely on brute force approaches. More data, more compute, more energy. The expected productivity gains remain uncertain across many sectors beyond early applications. That trajectory is not sustainable. Large model training requires significant compute over long periods. Widespread use of AI creates constant demand across cloud and edge systems.

 

Data has cost as well as value

 

Data is often described as an asset. While that is true, it also has a cost.

 

Large volumes of unused data still require storage and cooling. They consume electricity and increase system load. At the same time, they create risk. Cyber-security threats are growing. Concentrated data increases exposure. Users are becoming more aware of the value and sensitivity of their data.

 

There is also a gap in how this issue is framed. The focus is often on data centres alone, which are only part of the system.

 

Telecoms infrastructure is critical. 5G networks, fibre systems and backhaul carry the traffic that enables AI services, streaming and data intensive applications. These networks are always on. They cannot easily shift demand or reduce load at peak times. Mobile operators in the UK have recently warned that rising energy costs could force them to consider measures such as data rationing, reduced speeds or peak pricing. The sector consumes large amounts of electricity, yet it has not been treated in the same way as other energy-intensive industries.

 

At the same time, demand continues to grow. Streaming services, messaging platforms and AI driven applications place increasing pressure on networks, but the cost of supporting that demand is not always reflected in how services are priced or how infrastructure is funded.

 

This creates a clear misalignment. Telecoms networks are critical national infrastructure, yet they are exposed to rising costs and constrained investment signals while demand continues to scale.

 

Local constraints are now the issue

 

Infrastructure is becoming more distributed. Hyperscale sites remain important. Regional and edge systems are growing. This changes where energy is needed. Demand is no longer concentrated in a few large locations; it is spread across cities and networks.

 

What I see is a shift from national supply issues to local constraints. A country may have enough generation overall, but it may still lack grid capacity where data centres are being built. This leads to delays and higher costs.

 

Aligning with energy systems

 

Data centres cannot be planned in isolation. They must be aligned with the grid. Location, timing and flexibility all matter.  

 

Renewable energy is essential, but this is not enough on its own. A data centre can have renewable supply agreements and still place pressure on a constrained network. Co-location with renewables helps, but only when it is supported by storage, flexibility and proper planning.

 

Data centres and networks should act as part of the energy system. They should respond to signals and support stability where possible.

 

A question of integration

 

The issue is not whether digital infrastructure can grow, but whether it can grow within the limits of energy systems, telecoms infrastructure and security constraints, and whether costs are allocated in a transparent and equitable way across users. This requires attention from ICT companies, investors and policymakers.

 

If we treat this as a race for scale, we will create new constraints. If we treat it as a system challenge, we can manage it.

 

Digital growth and sustainability do not have to be in conflict. They can be aligned. That depends on the choices we make today.

---