Niklas Lindqvist, Nordic General Manager at Onnec, explores why data centre operators need a new definition of modularity to adapt to growing demands driven by emerging technology like AI.

Growing workloads in AI, cloud computing, and streaming have all created a huge demand for data centre compute capacity. Operators are racing to meet this demand, but to do so, they must figure out how to extend the life of data centres to be more flexible.

This requires a new kind of modularity – data centres that are designed with flexibility and interchangeability from the outset. But only with a holistic approach to data centre design and planning can operators create modular data centres that can keep pace with changing workload and client demands.

How has the concept of data centre modularity changed in recent years?

Modularity has traditionally meant building data centres from prefabricated modules or containers that included everything from electronics to cooling. This was intended to speed up construction of the sites.

However, these containers limited operators’ ability to scale and get ahead of problems that hinder interchangeability, such as bad cabling, poor installation, and resource shortages. These challenges saw high-profile hyperscalers, including Google and Microsoft, move away from using containers to scale data centre capacity.

Instead, operators must embrace a new kind of modularity, where data centres are designed with flexibility and adaptability from the outset. These sites will utilise plug-and-play hardware to enable frequent and easy upgrades to ensure data centres keep up with the demand for AI. This approach will also make data centres more efficient, adaptable, and sustainable, helping companies achieve their ESG targets.

What pressures do operators face that are driving the demand for modular data centres?

Evolving technology trends, including AI, the IoT, and edge computing, have brought surging workloads and challenges to data centre operators. This change is evident in rack power and density trends, as rack power densities have already risen from 4-5 kW over a decade ago, to 8-10 kW in 2020. The JLL North America Data Centre Report also predicts rack densities will need to increase further, to 50-100 kW per rack.

Surging demand has created three clear challenges for operators: a rapid increase in the need for AI-related data centre capacity, a construction slowdown, and the prolonged operational lifespan of hardware. To overcome these challenges, operators need to conduct rapid hardware refreshes more frequently, specifically to adopt and integrate GPU-powered racks.

How are cabling limitations holding back modular data centres?

Faulty and bad cabling can greatly impact the efficiency of data centres by causing poor connectivity, data corruption, and troubleshooting issues, which all lead to downtime and outages. These problems are caused by badly planned cabling that limits interchangeability. This starts in the design phase, where operators are typically focussed on power and cooling, rather than thinking about cabling or anticipating future changes in demand.

Too often, the cabling backbone installed isn’t designed to help data centres scale. As demand grows, the cabling infrastructure cannot accelerate to meet higher bandwidth needs. This means cabling could become obsolete within three years, instead of a more acceptable 10-15-year lifetime. Not only is this costly, compared to best practice installation from the outset, but it also increases environmental concerns through the production of e-waste. For modularity to become a reality, correct, quality cabling must become a much more important aspect of data centre design.

Why is cabling so important for enabling modularity?

Cabling is key for modularity – it’s the foundation that gives data centres the flexibility that’s needed over their lifespan. When best practice is followed, operators can seamlessly upgrade hardware as needed to deliver AI-compute and cater to evolving workloads. This means sites can stay competitive for longer with less waste and more efficiency. Ongoing maintenance is also easier, and sites can be repurposed as demands change.

But to adopt best practice, a mindset change is needed – operators must think about sites as integrated engineering purpose-built projects, rather than just another commercial property development. This mindset shift helps operators consider how the data centre could evolve as workloads change and develop cabling infrastructure that supports interchangeability. A consistent approach to cabling, brought about by limiting the number of cabling suppliers chosen, keeps operators focused on the data centre vision. This creates a standardised architecture that enables seamless interchange of hardware to achieve modularity. 

How can cabling help data centres be built with modularity at their core?

Data centre design can no longer happen in a silo. Operators must take a more holistic approach to data centre design and planning, and ensure cabling best practices are followed. This will allow operators to adopt a modern approach to building a modular IT infrastructure where interchangeability and connectivity are foundational.

To create modular data centres, it is useful to begin by defining the data centre’s strategic direction and focus on business-driven decisions up front, rather than technical requirements. Teams should also consider how AI can help the on-site team optimise data centre operations, as well as how to invest the right amount of capital in the early design stages to enable long term efficiencies rather than short-term cost savings.

Before operators think about cabling installation, they must create a detailed cabling project plan to lay the foundations for the infrastructure required, which will minimise installation delays. A meticulous installation guide should also be created that outlines how cabling should be installed, providing a working plan for installation, as well as detailed instructions for seamlessly interchanging hardware. Operators must also ensure they have the right skills in the team to install, maintain, and replace equipment.

For existing sites, it can be difficult to create an interchangeable infrastructure. A good starting point is creating a digital twin of the data centre to understand infrastructure, cabling, and complexities at play. With the twin, operators can review infrastructure to identify how to standardise and reduce complexity by introducing shorter cabling and reduce the number of connections. This also allows operators to improve hardware upgrades, scale, and adapt to changing demands more effectively and makes technical teams more efficient.