Back to articles

Our Future Cities Rely on Microgrids to Power Edge AI Data Centres

Category
AVK thinking
Topic
innovation
Date
20 January 2026
Author
AVK
Read Time
11 min

Microgrid power: The near-term power solution for a new generation of 20MW-30MW+ AI Edge data centres across Europe.

AI edge data centres

How much power will be required for AI Edge data centres if European economies are to take full advantage of the next stage of the AI revolution? According to the UK Government Compute Roadmap, it is estimated that the UK alone could need 6GW of additional power needed across all AI categories by 2030, while studies show Europe’s total data centre power demand could reach 36GW in the same timeframe. 

At the same time, we are witness to regular reports that the coming decade will see access to power as the major blocker to AI data centre developments.

Although different categories of data centres have always existed, today data centres are commonly viewed as large out-of-town campuses of low rectangular, warehouse-like buildings connected to the power grid. It is true that many AI data centres being developed today are wide, flat buildings which are grid attached through 30 or 50 mega volt ampere (MVA) substations.

But as well as being located remotely, data centres have always existed in and around cities. This is where the new AI Edge classification of a microgrid-powered, AI data centre is evolving.

What is an AI Edge data centre?

AI Edge data centres are an emerging digital infrastructure category primarily designed for latency sensitive AI inference applications – where the distance between the data generator, the data user and the data centre is short (see Roadmap below). AI Edge data centres are designed specifically to host liquid cooled GPU based server technologies at high rack power densities. Due to the power intensity of AI, these Edge data centres could require baseload power ranging from 10MW to 50MW and above. They are often vertical in structure with many being developed on brownfield urban sites.

  • The road map to AI Inference and proximity workloads

    How Inference AI is driving the development of Edge AI data centres

    AI Edge data centres are the required infrastructure layer to ensure inference workload success. Inference AI is a term that covers productivity AI, applied AI, Agentic AI, reasoning AI, industrial AI…the list is growing.

    The rapid deployment needs of these new workloads, applications and AI services is driving demand for AI Edge data centres located in close proximity to end users, i.e. population centres. There are many reasons why Inference AI will reside in AI Edge data centres:

    • Much Inference AI will be extremely latency sensitive.
    • The volumes of data required will run into the Zettabytes for proximity AI workloads (video, interactive apps, autonomous vehicles and automated services).
    • This processing, storage and microsecond connectivity means vast amounts of data will need to be located close to major population centres – for consumer AI but also for enterprise alignment of collaborative AI.
    • The exact nature of the top of stack enterprise, B2B, B2B2C and B2C applications is an exciting area of AI software development. These include those that must operate on a ‘real time’ basis – low millisecond or even microsecond round trip networking – such as instant voice translation services where latency must be imperceivable to the user.
    • Real‑time inference on the edge is driving explosive growth in both compute use and data generated locally. Edge‑processed data volumes are expected to see 20–30% CAGR through the late 2020s, with multi‑zettabyte scale data flows across IoT devices, smart systems, and connected vehicles.

It is being driven by the need for location-based AI services which require a jump in scale demanded by AI, combined with the increase in GPU rack densities. However, as with out of town campuses, the major issue for AI Edge data centres is access to power acting as a major hurdle to development.

The UK provides one example of what a looming power shortage could look like in a major economy. With the highest power costs in Europe in 2025, the UK is generating 4000kWH per head of capita – this is half that of France. According to a report commissioned by Drax and conducted by Public First: Mind the gap: Exploring Britain’s energy crunch, by 2028, if new generation is delayed, the UK could see a shortfall in dispatchable and baseload capacity of 7.5GW at peak times.

The UK is not alone. A majority of Europe’s established and evolving metro data centre markets are facing power access and availability issues. This means the scale and number of Edge developments being sought for AI can only happen by addressing the fundamental question, “where will the power come from for this new category of data centre?”

The solution for near term power at scale at the AI Edge is on-site microgrid power generation using a sustainable energy mix.

“AI will use 10% of global energy by 2035.”

– BP’s Global Energy Outlook Report 2025

London calling: AI Edge scale and the need for speed

It is useful to put the scale of Edge AI data centres into context through comparison with other commercial property developments in a city like London.

To take one example, London’s tallest commercial building, The Shard, is a mixed use structure powered by 1.13MW onsite combined heat and power generation using natural gas. To build a medium sized AI Edge data centre, close to or within London (by far Europe’s largest data centre market) could require finding an energy source for a load requirement of 20MW – 30MW.

Such power capacity is unlikely to be available via the grid – with numerous reports and studies saying that developers face waiting a decade or more for a connection. In April 2025, Ofgem – the UK’s energy regulator, issued a grid connection reform report in which law firm Norton Rose Fulbright assessed its impact on data centres.

Amidst the wait for reforms in cities like London, today’s market reality is that the speed of deployment for AI workloads requires the advantages and benefits provided by microgrids.

Among many reasons for data centre developer interest in microgrids, is competition time to market among investors pursuing Edge AI data centre opportunities.

This is pushing companies to seek alternative power generation in large data centre markets such as London, Frankfurt, Amsterdam and Paris where competition for grid connections is high, but also to secondary and tertiary markets across Europe where developments already face being hampered by availability and grid stability concerns.

Yet the AI demand curve is so steep that in London and other European major metros the number of new AI Edge data centres is expected to reach hundreds of multi-megawatt facilities. This is despite power market challenges that span variations in price, low existing available capacity, and extended timelines for long term modernisation and new generation. As overall demand for electricity surges, grid connection times are lengthening across densely populated regions of Europe.

It has been well documented that a data centre project may take 18-30 months, but for new grid infrastructure it currently takes about 12-14 years to new high-voltage transmission capacity from “identification of need”, found by the UK’s Electricity Networks Commissioner.

It is into this environment that Nvidia CEO Jensen Huang predicted that 200 AI centres [will be built] with $300 billion total investment needed by the decade’s end.

Evidence suggests this build out has already begun. A recent study revealed a jump in the number of data centres entering the UK planning process. Construction researchers Barbour ABI analysed planning documents showing the number of data centres driven by AI demand is set to grow by almost 100. Most are due to be built in the next five years.

What is behind the growth of AI Edge data centres?

The top end of the digital infrastructure market sees regular announcements targeted at vast AI developments with hundreds of billions of dollars already earmarked for 500MW-GW+ size campuses for AI training data centres.

There is no denying countries such as the UK, through its $150bn tech deal with the US and France with its home grown AI strategy, have made Europe headlines for this kind of infrastructure alongside announcements and expectations that large AI training data centres will be built in countries with relatively low populations, abundant land and access to relatively low cost, low carbon energy.

This is because in general AI LLM model training is not latency sensitive, meaning large AI companies are comfortable with locations away from the big metro centres. But, the next stage of AI will involve using trained models – so called inference AI with unverified reporting that “75% of AI workloads will be ‘inference’ in nature”.

Whatever the ultimate figure, as inference workloads roll out, a new class of digital infrastructure in the form of AI Edge data centres, will be required. And just as for the campus style data centres, those investing, designing and developing  AI Edge infrastructure must solve the chip energy challenges which begin with finding the power.

Why AI Edge needs microgrids: New types of server hardware and different power profiles

For designers working on AI Edge data centres, the challenge is new power chain topology designs to run varied server architectures at rack densities of hundreds of KWs that will require liquid cooling.

In power provision terms, the key need is to remain flexible to accommodate multiple GPU and CPU architectures, to integrate with cloud computing platforms, to accommodate HPC supported by super fast networking. This is matched in complexity only by the need to deploy flexible power infrastructure without compromising on efficiency, reliability and resilience.

And once thousands of GPU servers are operating running new types of Inference AI workloads – the nature of AI processing means power draw is expected to spike and drop unpredictably. Already people are asking, can the grid respond to drops and spikes of 5MW-10MW as AI compute clusters ramp and drop within seconds?

Do not expect a one size fits all to serve this diverse set of compute requirements. There are a whole spectrum of data centre infrastructure types and strategies that will be in play at the edge. But again the persistent challenge remains: Where will the power come from?

Location location location: Where AI data centres are being built

As noted, AI Edge data centres have already reached the design and planning stage in such major data centre markets as London, Frankfurt, Amsterdam and Paris.

As well as adding to existing primary markets, Edge AI data centres will be an infrastructure category of choice for cities across Europe, in Munich, Berlin, Lyon, Barcelona, Manchester, Birmingham and capitals such as Budapest, Sofia, Athens and to smaller regional cities with smaller populations.

Attractiveness to investors

We may be at the start of the curve. A basket of reports point to the global Edge AI market being expected to climb over the next decade at CAGR of 30%+.

Market growth projections:

In Europe, the AI Edge infrastructure market overall – from traditional small-scale edge nodes to mid‑sized facilities – is forecast to rise from $5.19 billion in 2024 to $28.68 billion by 2034 (CAGR ~18%).

It is expected that investment focus is shifting toward new builds on the outskirts of cities with developers seeking to circumvent being constrained by power availability, strict grid regulations and grid connection delays.

The total market size is expected to be in GW scale. However, instead of 500MW-Gw+ style campuses these locations, many of which will be developed on brownfield sites, and often be vertical in structure, built in blocks of 20-30MWs with many only realised through sustainably-powered microgrid infrastructure.