THOUGHT LEADERSHIP

Manish Kumar, EVP, Secure Power and Data Centres at Schneider Electric

The AI infrastructure race is accelerating rapidly. What is fundamentally changing in how data centres are designed and powered to support AI at scale?

The average rack size in 2020 was roughly 10 kilowatt. The size of the rack today is hitting 100 kilowatt. We also anticipate that this will continue to rise with NVIDIA projecting rack sizes up to 1 Megawatt.

The rise of AI is transforming data centres into highly interconnected, high-density ecosystems where power, cooling and digital operations must work seamlessly together. In this increasingly complex environment, Manish Kumar, EVP, Secure Power and Data Centres at Schneider Electric, explains that success will depend on the ability to simulate, monitor and optimise infrastructure continuously, turning data centres into intelligent, adaptive assets rather than static facilities.

When you have that level of density, it impacts everything, meaning you have to reimagine the entire power and thermal architectures. Traditionally, air cooling was sufficient, but now liquid cooling has become a must in cooling designs.

As a result of the increasingly complex nature of these new designs, you need to have a digital operating system to manage the whole centre, making sure you are able to look at every data point and its efficiency across the lifecycle.

At Schneider, we are working very closely with the leading compute players, such as NVIDIA, to make sure we focus on the right system architecture. This enables all the designers to simulate, build, operate and maintain end-to-end efficient architecture from a system standpoint.

Why has deployment speed become just as critical as performance and reliability in AI infrastructure?

The transformational impact of AI is being seen across every industry. However, it is profoundly driving demand and intensity in the data centre space. This is encouraging new native AI players to come into the space and invent applications and software stacks that are better than those that already exist. The demand itself is creating a huge pressure on the system, continually asking for more speed to stay ahead of the game. In order to leverage deployment speed efficiently, prefabricated systems are becoming ever-popular with leading designers as they shrink down the time to design and deploy those architectures.

Power and cooling are now defining challenges for AI data centres. How are AI workloads reshaping infrastructure requirements?

The exponential increase in rack densities, alongside their projected potential, is the driving force for the reconfiguration of power and cooling architecture.

Direct liquid cooling is already becoming the must have over air cooling. We already see 47 % of operators transitioning to liquid cooling architecture specifically for biodensity environments. Cooling is the most critical component in the architecture because if it fails, your physical

14 www. intelligentdatacentres. com