As AI scales, how can energy systems keep up?

AI infrastructure is scaling fast. Every week brings fresh announcements about new cloud regions, sovereign AI stacks, AI factories, data centre campuses and the fibre and network infrastructure needed to connect them. On the surface, it can look like a familiar infrastructure story; line up land, fibre, capital and customers, and the rest will follow. But as these projects become larger and more power-intensive, one question is moving to the front of the conversation; where will the electricity come from?

Some recent developments bring this issue into sharp focus.  

Microsoft’s collaboration with G42 (Abu Dhabi-based AI company) on a planned data centre project in Kenya brings this issue into sharp focus. The project reportedly ran into concerns because, at full scale, it could have required around 1 GW of power, close to a third of the entire country’s installed electricity capacity. President William Ruto was quoted as saying that powering the project would mean switching off power for half the country. That is an extreme example, but it is useful because it makes the trade-off visible. AI infrastructure does not sit outside the energy system, it draws from it.

The same pressure is visible across a broader set of markets, each showing a slightly different version of the same constraint. In Ireland, the Central Statistics Office says data centres used 22% of metered electricity in 2024, up from 5% in 2015. In Germany, the Federal Ministry for Economic Affairs and Climate Action says installed IT power demand exceeded 2,700 MW in 2024 and could reach 4,800 MW by 2030. In the US, The Economist has reported local backlash in parts of Iowa, Michigan, Virginia and Texas over power use, water demand, noise and transmission infrastructure. In Malaysia, ISIS Malaysia, a Malaysian policy research institute, highlights Johor’s rapid data centre growth alongside concerns about power, water and local infrastructure readiness. In Japan, METI (Ministry of Economy, Trade and Industry) and MIC’s (Ministry of Internal Affairs and Communications) Watt-Bit Collaboration report links rising AI and communications traffic with the need to coordinate electricity and telecoms infrastructure, while South Africa’s Department of Communications and Digital Technologies says 24/7 data centre power use may require operators to self-provision electricity and water.

The relationship between AI and energy is not a new debate. However, the latest demand projections suggest the conversation is shifting. The question is no longer simply whether AI consumes more energy or improves efficiency, but whether electricity systems can expand fast enough to support the scale of AI infrastructure now being announced.

What are the numbers already telling us?

The numbers make the issue evident. According to the International Energy Agency (IEA), data centre electricity demand could rise from around 415 TWh in 2024 to around 945 TWh by 2030, with AI as a major driver. This is not just a global demand number; the challenge becomes sharper when growth is concentrated in specific markets. The examples above point to why these matters for grid planning, not just technology strategy.

What happens when electricity demand grows faster than power generation, grid connections and transmission capacity can expand? Projects can be delayed, resized or pushed toward dedicated power arrangements. Governments may also have to make harder choices about who gets access to reliable electricity first: households’ daily essential needs, industry, AI infrastructure or other public priorities. In that sense, the AI race is increasingly becoming an energy-planning race.

Production, consumption and the missing energy check

To understand the issue clearly, one simple check is needed, and it is often missing from AI infrastructure announcements.

Before asking how many AI factories or data centres a market can attract, we should ask:

• How much electricity a country produces

• How much it consumes

• Whether it is broadly self-sufficient, dependent on imports or constrained by grid capacity

This gives a simple but powerful view and helps to answer the most important question – is there enough headroom for new, always-on demand?

If a country is only just meeting existing demand, depends on imported electricity, or has limited spare grid capacity, then adding large AI data centres becomes a structural issue. It is not enough to say that a facility will be efficient or powered by renewables. The bigger question is whether the wider system can absorb the load without creating pressure elsewhere.

This is the part that is often not discussed when new AI infrastructure is announced.

The bigger picture for operators and policymakers

These examples are not exceptions; they are a signal. 

For operators, cloud providers and governments, the next phase of AI infrastructure planning will need to connect digital ambition with energy reality. That means looking beyond land, fibre and capital, and asking earlier questions about power availability, grid connection timelines, generation mix and resilience. AI may still scale quickly, but the places that scale it best will likely be the ones that plan the energy layer as carefully as the compute layer. The limiting factor may not be demand for AI services, but the ability of electricity systems to support them at speed and scale.