Surprisingly Useful AI Article Enhancements
In the era of cloud adoption and Ai, the demand for data centre bandwidth has skyrocketed, leading to the exponential sprawl of data centres worldwide. However, new data centres are running up against sustainability, space and budget constraints. Policymakers recognise the benefits of data centres to productivity, economic growth and research, but there is still a tension over their impact on local communities, water and electricity use. The best solution is in optimising the data centre infrastructure we have already to unlock more performance while still being mindful of the limits we have.
Our cities, our consumer products and our world is going to become more digital and we need more compute to keep up. Optimising the data centre infrastructure we have already to unlock more performance is the best way data centres can turn constraints into an opportunity for a competitive advantage.
Why Data Centre Optimisation Matters
CIOs and IT leaders increasingly face calls to provide a high-performance foundational compute infrastructure across their businesses and handle new, more demanding, use cases while balancing sustainability commitments, space and budget constraints. Many have sought to build new data centres outright to meet demand and pair them with energy efficient technologies to minimise their environmental impact.
For example, the LUMI (Large Unified Modern Infrastructure) Supercomputer, one of the most powerful in Europe, uses 100 per cent carbon-free hydroelectric energy for its operations and its waste heat is reused to heat homes in the nearby town of Kajanni, Finland. There are many other examples like LUMI showing the progress the data centre industry has made in addressing the need for energy efficiency. Yet energy efficiency alone will not be enough to power the growing demands of Ai which is expected to increase data centre storage capacity.
Ai’s greater energy requirements will also require more energy efficient designs to help ensure scalability and address environmental goals. With data centre square footage, land and power grids nearing capacity, one way to optimise design is to upgrade from old servers. Data centres are expensive investments, and some CIOs and IT leaders try to recoup costs by running their hardware for as long as possible. As a result, most data centres are still using hardware that is 10 years old (Dell) and only expand compute when absolutely necessary.
While building new data centres might be necessary for some, there are significant opportunities to upgrade existing infrastructure. Upgrading to newer systems means data centres can achieve the same tasks more efficiently. Global IT data centre capacity will grow from 180 gigawatts (GW) in 2024 to 296 GW in 2028, representing a 12.3 per cent CAGR, while electricity consumption will grow at a higher rate of 23.3 per cent from 397 terawatt hours (TWh) to 915 TWh in 2028. For ageing data centres, that can translate to fewer racks and systems to manage while still maintaining the same bandwidth.
It can leave significant room for future IT needs but also makes room for experimentation, which is important in Ai workloads. They can use the space to build less expensive proof-of-concept half racks before committing to larger build outs and use new hyper-efficient chips to help reduce energy consumption and cooling requirements, recouping investment more quickly.
What To Look For In An Upgrade
There are many factors to consider in a server upgrade and there is no one-size-fits-all solution. It is not just about buying the most powerful chip that can be afforded. While the significance of a good chip on energy efficiency is clear, each data centre has different needs that will shape the hardware and software stack they need to operate most efficiently.
Leading South Korean cloud provider, Kakao Enterprise, needed servers that could deliver high performance across a wide range of workloads to support its offerings. By deploying a mixed fleet of 3rd and 4th Gen AMD EPYC processors, the company reduced the servers required for its total workload to 40 per cent of its original fleet, while achieving increased performance by 30 per cent and cutting total cost of ownership by 50 per cent.
Much like Kakao Enterprise, IT decision makers should look for providers that can deliver end-to-end data centre infrastructure at scale, combining high performance chips, networking, software and systems design expertise. The right physical racks make it easier to swap in new kit as needs evolve, and having open software is equally important for integrating different pieces of the software stack from different providers.
In addition, providers that are continually investing in systems design and Ai capabilities will be best positioned to accelerate enterprise Ai hardware and software roadmaps. AMD, for example, recently achieved a 38× improvement in node-level energy efficiency for Ai training and HPC over five years. This translates to a 97 per cent reduction in energy for the same performance, enabling providers and end-users to innovate more sustainably at scale.
Advancing The Data Centre
As reliance on digital technologies continues to grow, so does the need for computing power. It is important to balance the demand for more compute capacity with sustainability goals. Making the most out of existing infrastructure can turn this challenge into an advantage. With the right computational architecture, data centres can achieve the same tasks more efficiently, creating space for the future technologies that will transform businesses and lives.
Alexey Navolokin is General Manager, Asia Pacific (APAC) at AMD.
Last Updated on August 13, 2025 by Alexey Navolokin
