Smarter grids for the AI data centre boom

Artificial intelligence and machine learning are set to have a profound impact on many if not all aspects of our lives. However, they bring some unintended consequences for the power system.

Data centres are not only proliferating, with the advent of so-called hyperscalers they are growing in capacity too and their power demand is significant.

A recent report from the International Energy Agency, Energy and AI, projects that electricity demand from data centres worldwide is set to more than double by 2030 to around 945 TWh and is on course to account for almost half of the growth in US electricity demand between now and 2030.

By then the US will consume more electricity processing data than for manufacturing aluminium, steel, cement and chemicals.

New Goldman Sachs Research forecasts go still further, suggesting global power demand from data centres will increase by as much as 165% by the end of the decade compared with 2023. By then, they forecast there will be around 122 GW of data centre capacity online.

However, given the higher processing workloads demanded by AI, the density of power use in data centres is likely to grow as well. According to their analysis power density will grow from 162 kW per square foot to 176 kW per square foot in 2027.

This chimes with the IEA report which notes that demand from AI-focused data centres is projected to more than quadruple over the same period.

As a consequence of this substantial growth in energy consumption, considerably more investment in grid infrastructure is also projected. Goldman Sachs Research estimates that worldwide about $720 billion of grid spending may be needed through to 2030 to support data centre demand.

However, developing grids is not only costly but is also a time-consuming endeavour. As James Schneider, a senior equity research analyst and author of the Goldman Sachs report says: “These transmission projects can take several years to permit, and then several more to build, creating another potential bottleneck for data centre growth if the regions are not proactive about this given the lead time.”

The grid infrastructure challenge could also have significant implications for the growth of data centre capacity.

According to a Slaughter & May report, grid constraints are becoming fundamental barriers for value creation in the data centre sector. They cite delays in securing grid connections reaching up to 15 years in the UK with seven to 10-year connection queues not uncommon across Europe and some projects facing queues of up to 13 years.

These delays threaten project timelines and increase costs. They emphasise that in many markets wholesale electricity prices in grid-constrained can see data centres subject to elevated prices and capacity premiums.

Governments in Australia, Canada and the UK, for example, are looking to reform grid connection rules in a bid to address grid constraints, but analysis from Ember points to the potentially corrosive effect of limited grid capacity.

They argue that as time to market is critical for data centre players, better grid availability is directing data centre investment to the Nordics and countries in Southern Europe where grids are less congested and wait times for a connection are consequently far shorter. 

As a result, they forecast that by 2030 market growth in data centres in these regions is expected to reach nearly double the rate of traditional market leaders such as Frankfurt, London, Paris and Dublin.

This trend, says Ember, is expected to continue and by 2035, half of Europe’s data centre capacity will be located outside of these traditional hubs.

Nonetheless, Ember’s analysis notes that strategic choices by system operators can help overcome grid constraints, speeding up data centre deployment by reducing the required expansion of grid infrastructure.

One attractive solution is to use grid enhancing technologies that can make more of existing assets. Of these, dynamic line rating is already delivering a route out of grid capacity constraints.

As with many aspects of the energy sector, the capacity margins for any power conductor cable span are invariably highly conservative.

For conductors, higher ambient temperatures are often associated with higher conductor temperatures and that in turn makes them more susceptible to sagging. Where conductors sag the clearance with the ground or other objects is reduced, potentially allowing energised lines to contact trees or vehicles, for example. 

Want more content on AI and data centres? Try these:
High-performance utilities: How data and AI power success
Energy Transitions podcast: AI, robots and the future of nuclear decommissioning
Former Scottish steel mill could become ‘green AI data centre’
Rolls-Royce to launch fast-start gas gensets to power data centres

As well, conductor temperatures need to be maintained under a certain limit determined by conductor’s characteristics. As line ratings, the maximum permissible current flow that the line can withstand without exceeding safety parameters, are typically based on worst-case scenarios and modelled to ensure complete reliability. 

When higher ambient temperatures are generally experienced, during summer for instance, line ratings are reduced by a corresponding amount.

In the majority of jurisdictions, so-called static line ratings are used to ensure power lines remain well within limits to reduce the risk of sagging causing problems or of excessive conductor temperature.

However, using a more sophisticated dynamic line rating approach can free up additional transmission capacity based on real-world conditions rather than broad assumptions.

For example, even when solar irradiation might raise the temperature of a conductor, higher wind speeds have a cooling effect that could potentially allow more power to be passed down the line.

This places a greater emphasis on accurately assessing the highly localised impact of wind which can be dramatically affected by immediate variation in terrain or even individual trees for example.

To overcome these issues and free up capacity, Ampacimon uses its patented sensor technology to generate accurate data from individual conductor spans.

Powered using induction from the conductor current itself, these sensors are fitted with accelerometers to measure the frequency of vibrations of the conductor. By carefully analysing these data on the vibration spectrum, an accurate estimate of any sag can be derived along with actual conductor temperature in real time.

This approach gives an unprecedented level of fine detail on conductor conditions but is also coupled with a machine learning model as well as highly accurate weather forecasting data.

Already widely deployed in Europe and America, Ampacimon’s dynamic line rating systems are helping grid operators enhance grid transfer capacity.

A recent example comes from Denmark’s Energinet, an independent public enterprise owned by the Danish Ministry of Climate, Energy and Utilities which owns and operates the transmission system in Denmark.

Energinet had previously announced it will invest DKK 41 billion (US$6.5 billion) in the electricity transmission system between 2023 and 2026 expanding and strengthening the Danish electricity transmission grid with a total of 3,300 km of new conductors.

However, traditional approaches to capacity enhancement such as reconductoring and building new lines take a long time, while capacity is urgently needed. As a result, Energinet is also using DLR to replace its static ratings and relieve constraints on Demark’s overhead lines.

Unlike a system based on static ratings, DLR takes into account the temperature of the conductor and many other environmental factors, such as the cooling effect of the wind. 

In developing an immediate and accurate assessment of the true thermal state of the conductor as well as a forecast for future conditions, grid operators making use of DLR can unlock latent grid capacity. 

This grid enhancing technology thus relieves capacity crunches and supports the deployment of more load, even in constrained grids. 

It does this safely, in a timely way and at far lower costs when compared with the traditional approach of building additional lines or other grid reinforcement measures.

It seems that the answer to the data centre challenge and the growing role of AI systems is get smart and deploy smart grid technology.

The future of AI and data centres will be debated in detail at Enlit Europe in Bilbao. Here’s just some of the can’t-miss sessions taking place:

AI for Energy, Energy for AI
Optimising Green Data Centers
AI in Practice
Gen AI: Monetising the Investment