Listen to the podcast
To hear the full episode, please follow the links to Apple Podcasts or Spotify, or use the player below.

The backbone of the digital economy

Data centres are the backbone of the digital economy. Whether simple or complex, if a task requires computation, there’s a facility somewhere to store, process and transmit the data that makes it happen.

Hyperscalers—major companies operating the largest data centres—operate vast campuses containing thousands of servers and storage systems with a power capacity of up to 100 megawatts (MW). 

With the rapid evolution in artificial intelligence (AI), and the additional processing power it requires, there is now a greater focus on how much energy these facilities use. The power consumption for these facilities is expected to more than double between now and 2030. By 2026, they will consume more in a year than Germany or France.

AI is changing the game

There are now more than 11,000 large-scale data centres globally. Hyperscalers, third-party companies (usually cloud service providers), operate these large data centers and rent space to customers. Meanwhile, retail or enterprise colocations offer the physical infrastructure, such as cooling and bandwidth, for up to 200 tenants.

In addition, many businesses instead own and operate smaller-scale data centres kept on site. This can be anything from a single rack to a fully stocked computer room.

Now, AI is changing the landscape, with large language models (LLM) requiring huge amounts of data. As an example, training ChatGPT4, one of the best-known LLMs, took 90 days and required 15 MW of power capacity. That’s equivalent to the annual energy consumption of around 2,000 European single-family homes. 

Using any LLM is around 10 times more energy intensive than a conventional web search, and their arrival is driving a fundamental shift in data-centre architecture. Instead of more conventional facilities, which use central processing units (CPU), operators require graphic processing units (GPU) and application-specific integrated circuits (ASIC). 

Location, location, location – where to build a data centre

With power consumption such an important aspect, access to low-cost – and, if possible, low-carbon – energy is a key driver for data-centre developers. But there are several other factors to consider, including access to land, climate, connectivity, the overall regulatory regime and any incentives that may be on offer.

With abundant and cheaper energy, the US remains, by far, the biggest market, with around half of all operational centres globally (and 56% of those still in the pipeline). Many of these (around 10% of global installed capacity) are in North Virginia’s “data-centre alley”, where tax exemptions and grants are available. However, there are also significant operations in Europe, which has around 18% of the current operating capacity. The cooler climate in Northern Europe has attracted some developers. Facilities are easier to cool, while the excess heat generated is recycled to support local heating grids. Asia is also a popular area, making up around one-third of both operating capacity and the development pipeline.

A closer look at data-centre power consumption

Around 60% of power consumption comes from the IT infrastructure, such as servers and storage systems. But a significant amount of energy is also required to remove or counter the substantial heat generated by the individual chips, critical to ensuring the hardware runs correctly. Around one-third of power consumption is related to cooling and ventilation.

While most centres still use air cooling, which is a relatively cheaper option, more providers are moving to liquid-cooling systems. While this is more expensive, reducing – or entirely removing – the need for fans in IT equipment can cut the power consumption by between 4–15%.

There are also other areas that could improve a data centre’s energy efficiency: advances in chip technology meaning that more powerful does not necessarily equal more power hungry; overall data-centre architecture, such as visualising workflows; and more consolidation as smaller on-premises data centres move to cloud-based solutions. Further energy efficiency could also be achieved with greater integration of servers, cooling, and power systems. 

Opportunities for investors

Data centres are an integral part of the ongoing energy transition as the world moves away from fossil fuels towards greater electrification. So how should investors approach this theme?

Beyond providers of server equipment, or network and cloud-service providers, other industries include those delivering cooling and power systems for buildings or industrial processes, or companies offering power-plant or grid equipment should profit from the heightened demand for data centres. 

That said, there is consensus that the shift to electric mobility will account for the lion’s share of growth, adding around 20% to 30% to today’s consumption, followed by home electrification at around 15% to 25%, and data centres potentially around 10% after 2030 but before 2040. 

This means that, for investors, there is undoubtedly a broader range of investment opportunities, while acknowledging the complexities that come with this evolving theme.