As digital services become more prevalent in every area of life, the digital infrastructure that supports them, with data centres at the heart, is adapting. What we're seeing is this infrastructure becoming more use-case specific, as location, design and hardware are bespoke to support different use cases. It is therefore more important than ever that businesses understand the different types of data centres and what each one is designed to deliver. In this first article in our new series on data centres, we set out the five most common types of data centres and how they vary, while also exploring what comes next. The series takes a closer look at a range of topics relating to various aspects of data centre operation, contracts, planning, financing and power.

1. Enterprise data centres

An enterprise data centre is a private data centre that supports a single organisation. These are best suited to companies with unique needs or those large enough to benefit from economies of scale. Enterprise data centres are often custom built for the business (known as 'build to suit'), designed to be compatible with the company's distinctive processes and applications and remove risk related to third party control over a company's data and confidential information.

These data centres are usually located on the premises of the organisation or off-premises at a site with suitable connectivity, power and security. Regardless of the location, an enterprise data centre will generally have the company's IT department managing the IT equipment and infrastructure, known as the 'white space', and will likely outsource the back-end components and equipment, known as the 'grey space'.

2. Colocation data centres

The key differences between colocation and enterprise data centres is that a colocation data centre (also known as a multi-tenant data centre) is operated by a third party data centre operator, who provides services to multiple customers who want an offsite space to host computing hardware and servers.

The benefit for businesses here is that all the components required in the running of a data centre, such as power, cooling, security and networking equipment, are provided as services by a specialist data centre operator, without requiring the capex investment by the business. This type of data centre is used by the majority of enterprises who lack the space, skills or staff to operate a data centre by itself and follows the trend in modern IT service provision, which is for business to buy in IT as a service, rather than trying to operate this itself. Customers will therefore pay a premium for having a service delivered and recognising the shift of risk onto the data centre operator and capital investment required by the operator. However, it also gives customers flexibility to simply buy more capacity as their business grows, rather than having to incur heavy costs up-front to try to future proof.

There are two types of colocation facilities – retail and wholesale. Retail colocation facilities sell smaller spaces with less power to a larger number of customers. Wholesale facilities can lease out a whole data hall or even the entire data centre to a single customer.

3. Hyperscale data centres

Hyperscale data centres are designed with large-scale IT infrastructure in mind i.e. as used by the likes of Google, Microsoft, Amazon and Meta.

A hyperscale data centre could either be an enterprise data centre (operated by the hyperscaler and dedicated to them) or wholesale colocation (operated by a third-party data centre operator). The main characteristic of a hyperscale data centre is its sheer size and power capacity. The equipment that a hyperscale data centre may be required to house is vast, including at least 5,000 servers and quite possibly miles of connection equipment. As such, hyperscale data centres can easily encompass millions of square feet of space and tens, or even hundreds of megawatts of power. The technical design of the facility is also different, given the complexity of hyperscaler systems and their requirements for resilience, risk management and connectivity.

As artificial intelligence (AI), automation, data analytics, data storage and data processing demand are all increasing, the demand for hyperscale data centres is also on the rise, meaning that the number and scale of this class of facility already accounts for 37% of total capacity (according to Synergy Research Group), and could double in the near future.

4. Edge data centres

Edge data centres, also called micro data centres, are small and located near to the processing, analysis and action of the relevant data, typically near to urban areas. Their purpose is to enable low-delay communication with smart devices, which uses lots of bandwidth. By processing data as close to end users as possible, transmission and processing is as fast as possible, which is essential for real-time services such as traffic controls or operating machinery. Typically, edge data centres support critical loads of no more than 100-150 kilowatts (kW).

Micro data centres are ideal for edge applications, especially in distributed, remote, or unconditioned locations. Because the entire system is enclosed into one IT rack, micro data centres can be used in existing network spaces or small server rooms, as well as in open office spaces, retail stores, clinics and even ships.

5. Modular data centres

A modular data centre, also known as a container data centre, can be in both temporary and permanent deployments. It is usually a module or shipping container with ready-made data centre components. This means they are easily deployed, energy efficient and customisable. They are therefore ideal for construction sites or in disaster areas, but do have multiple purposes, such as a backup or support for an existing business, or for rapid expansion with zero interference due to the modular, 'premade' design.

What's next: AI-ready data centres?

With the explosion of generative AI in the last two years, the data centre market is racing to provide a sufficient supply of 'AI-ready' facilities. 'AI-ready' could mean a number of things, depending on the type of AI the customer wants (large language model (LLM), neural network, genAI) and what part of the AI-lifecycle they need the facility for (training, fine-tuning, inferencing). Using a data centre for AI purposes is likely to influence:

  • Location – no need for proximity to urban areas if you are training an AI model.
  • Space required – much less now that rack densities can support up to 100kW of IT load.
  • Power required – AI prompts require exponentially more power than cloud computing.
  • Cooling infrastructure – direct to chip liquid cooling is required to get the best performance from AI-specific chips and much higher rack densities requires reconfiguration of cooling systems.
  • Connectivity – networking is far more important for AI services than previous high-performance compute.

This latest digital trend perfectly demonstrates the evolution of use-case specific data centres, as the design of IT solutions clearly now extends down to the digital infrastructure layer of the tech stack.

What does the future hold for data centres?

As the digital landscape evolves, so does the infrastructure that supports it. From enterprise data centres built for bespoke organisational needs to hyperscale facilities powering the world's largest tech giants, the demand for more specialised, use-case specific data centres is growing. With increasing technological advancements and the rise of AI, we can expect further innovations in data centre design and operation, driven by the need for greater performance, efficiency and flexibility. The future of data centres will undoubtedly be shaped by these ongoing shifts as businesses and service providers continue to adapt to an increasingly data-driven world.

To discuss any of the points raised here further, please contact Jocelyn Paulley, and look out for other upcoming articles in our data centre series.