Artificial intelligence (AI) is driving one of the fastest increases in electricity demand seen in modern digital infrastructure. So the basic question when it comes to power and AI is, will data centres have enough electricity to handle the computational load? 

The inconvenient truth is no, not without significant investment, grid rejuvenation, the creation of new generation capacity and changes in how AI information is distributed. 

Data centres with modern GPU-based clusters are consuming 40–120kW per rack, and facilities reaching 100–500MW (megawatts) are needed. 

Traditional grids and transmission infrastructure are often unable to deliver reliable and concentrated loads, creating bottlenecks in power delivery rather than total generation. Advanced cooling solutions, on-site generation, and renewable integration partially mitigate these constraints, but reliability remains critical.  

Long-term AI deployment will increasingly depend on strategic positioning of data centres and energy sources, which links a nation’s AI ambitions directly to investment in generation, transmission and grid modernisation. 

Traditional enterprise data centres were designed around relatively modest and predictable loads, typically operating at rack densities of 5–10kW. AI has obsoleted that model. 

As mentioned, modern GPU-based AI clusters must have 40–120kW per rack and huge new centres with power demands reaching 500MW are the future. At this scale, a single data centre can rival the electricity consumption of a medium-sized city. Unlike many industrial loads, AI compute is continuous, latency-sensitive and reliability of power supply is an essential requirement. 

A further limiting factor is often not total energy availability, but localised capacity, and this can be affected by a number of operational factors that are playing an increasing role in challenging power availability. 

Prime amongst these are: transmission congestion, interconnection traffic jams and thermal limitations.  

When it comes to transmission congestion, many countries lack high-voltage transmission capacity to deliver hundreds of megawatts to a single site, and many new data centres increasingly face multi-year delays for grid connection approvals, which increases interconnection queues. 

So the point here is that in many geographic locations, the overwhelming limitation of AI is not total electricity generation, but the ability to deliver power to specific locations. Typically, in the past, power transmission infrastructures were designed for gradual, distributed growth, not sudden multi-hundred-megawatt loads.  

As a result, high-voltage transmission lines are often unavailable or congested near suitable data-centre sites, grid interconnection queues can delay projects by several years, and large, rapid power load additions can stress grid stability, reserve margins and frequency control. 

Data-centre siting decisions are therefore increasingly driven by electrical topology and grid proximity, rather than land cost or network latency alone. 

Cooling and Power Possibilities 

Power consumption scales non-linearly due to thermal management, which means that air cooling becomes impractical beyond 30–40 kW per rack, and direct-to-chip liquid cooling and immersion cooling are becoming standard for AI workloads.  

However, even with advanced cooling, power usage effectiveness (PUE) improvements are flattening, meaning compute efficiency gains are increasingly offset by absolute load growth. 

To try and solve the problems of grid constraints, data centre operators are adopting hybrid energy strategies such as on-site generation involving gas turbines, reciprocating engines and fuel cells to provide stable capacity and reduce grid dependency. 

Operators are increasingly turning to PPAs to resolve power supply problems. A Power Purchase Agreement (PPA) is a long-term contract between an electricity generator and a buyer for purchasing electricity, often from renewable sources, at a pre-agreed price, providing financial stability for projects like wind or solar farms and ensuring green energy supply for buyers. 

Something that project managers like is that PPAs offer predictable costs and support sustainability goals, with options for physical delivery or financial settlement and terms usually spanning 10-25 years, covering operation and maintenance. 

Modern Power Plays 

To get around the constraints of overloaded or inefficient distribution grids, operators are adopting hybrid energy strategies. Amongst these are On-site power generation equipment using turbines, reciprocating engines and fuel cells to provide firm capacity and reduce grid dependency. Hardware and software efficiency are also considerations using specialised accelerators, lower-precision computation, and model-level optimisation to improve FLOPs per watt. 

Strategic Outlook 

In the short term, power availability and interconnection delays will be a hindrance to AI expansion, but in the medium term, AI growth will drive accelerated investment in generation, transmission and behind-the-meter power. For the long term, compute deployment will increasingly follow energy abundance, not network proximity, linking AI scale directly to national energy strategy.  

AI data centres use a lot of electricity, and we know that running them constantly takes massive, consistently supplied power at costs that are controlled within a planned budget. Nuclear energy can provide all of that, but the problem is that nuclear plants take a long time to build and don’t come cheaply. So when it comes to the power game, who’s got the edge globally? 

China currently has roughly 59 active nuclear power reactors and is building a large number of additional ones, and has plans for around 90 reactors. How’s the USA doing? It has about 94 reactors generating electricity across the country.  

What about the EU? Currently, it has 106 operable nuclear power reactors across its member states. Finally, what about Britain?  The UK has 9 operable nuclear reactors generating electricity. 

Location, Location, Location 

So it is abundantly clear who the big power players are, and Britain certainly needs to invest more in that area if it hopes to realise its much-talked-about AI ambitions. But it’s not all about how many nuclear reactors you have, it’s also about location and making sure the positioning of them and data centres coincide efficiently. 

There are other ways of supplying the energy demands of AI, and one of those is by using renewable energy. The big question is, can they provide enough power, given how climate-reliant they can be?  When it comes to volume, the answer is yes. Experts will say that solar, coupled with wind power, will provide sufficient energy for AI data centres, but there are issues which make them risky when it comes to sustainable supply.  

Because of climate unreliability in certain parts of the globe, Britain, for example, data centres must have energy storage and backup systems, and these can consist of UPS batteries, generators, multiple grid feeds and a system called software failover whereby the system can shift loads to other sites if required. 

There is a financial advantage to renewables: their costs keep falling, and they are cheaper and faster to build than nuclear reactors. 

Large-scale Investment 

The data centre industry pundits predict that over the next few years, power availability and grid interconnection delays are likely to slow or redirect AI expansion in mature markets. In the medium term, sustained AI growth will require large-scale investment in generation, transmission, and behind-the-meter energy systems. 

Longer term, AI deployment is likely to follow energy abundance rather than network proximity, tightly coupling compute scale to national and regional energy strategy. AI is not simply increasing electricity demand; it is redefining the role of data centres within the energy system. As compute density rises, the boundary between cloud architecture, grid engineering, and power policy is dissolving.  

Ensuring that AI data centres can handle the massive workloads of the future will depend on the ability to plan, build, and integrate energy infrastructure at unprecedented speed and scale while maintaining supply reliability and controlling costs.