Optimizing Data Centers for AI Growth and Sustainability

In the era of cloud adoption and AI, the demand for data center bandwidth has skyrocketed, leading to the exponential sprawl of data centers worldwide. However, new data centers are running up against sustainability, space, and budget constraints. Policymakers recognize the benefits of data centers 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 optimizing the data center 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 are going to become more digital, and we need more compute to keep up. Optimizing existing data center infrastructure to unlock more performance is the best way data centers can turn constraints into an opportunity for a competitive advantage.

Why data center optimization 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 centers outright to meet demand and pair them with energy-efficient technologies to minimize their environmental impact.

For example, the LUMI (Large Unified Modern Infrastructure) Supercomputer, one of the most powerful in Europe, uses 100% carbon-free hydroelectric energy for its operations, and its waste heat is reused to heat homes in the nearby town of Kajaani, Finland. There are many other examples like LUMI showing the considerable progress the data center industry has made in addressing the need for energy efficiency. Yet energy efficiency alone won’t be enough to power the growing demands of AI, which is expected to significantly increase data center storage capacity.

This is especially true in India, where hyperscalers are now joined by regional and mid-sized cloud service providers that serve the fast-growing small and medium-sized enterprises. The growing use of hybrid cloud infrastructure in the country (combining global cloud platforms with local data centers) is also increasing the urgency for optimized compute infrastructure.

AI’s greater energy requirements will also require more energy-efficient designs to help ensure scalability and address environmental goals. With data center square footage, land, and power grids nearing capacity, one way to optimize design is to upgrade from old servers. Data centers 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 centers are still using hardware that is 10 years old (Dell) and expand compute only when absolutely necessary.

While building new data centers might be necessary for some, there are significant opportunities to upgrade existing infrastructure. Upgrading to newer systems means data centers can achieve the same tasks more efficiently. Global IT data center capacity will grow from 180 gigawatts (GW) in 2024 to 296 GW in 2028, representing a 12.3% CAGR, while electricity consumption will grow at a higher rate of 23.3%, from 397 terawatt hours (TWh) to 915 TWh in 2028. For the aging data centers, 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 especially necessary in AI workloads at the moment. They can use the space to build less expensive proof-of-concept half racks before it leads to bigger buildouts and use new hyper-efficient chips to help reduce energy consumption and cooling requirements, recouping investment back more quickly.

What to look for in an upgrade

First, working with a chip provider that consistently prioritizes energy efficiency as its core design principle is key. AMD, for example, recently achieved a 38x increase in node-level energy efficiency for AI training and HPC, which equates to a 97% reduction in energy for the same performance compared to systems from just five years ago.

An example of leveraging existing server infrastructure to deliver more with less is leading South Korean cloud provider Kakao Enterprise. The organization needed servers that can deliver high performance across a wide range of workloads to support its expansive range of offerings. By deploying a mixed fleet of 3rd and 4th Gen AMD EPYC processors, the company was able to reduce their server fleet for the same workload by 60%, boost performance by 30%, and halve total cost of ownership.

But of course, it is not just about buying the most efficient or powerful chips that can be afforded. There are many factors to consider in a server upgrade, and there isn’t a one-size-fits-all solution to data center needs. Each data center has different needs that will shape the hardware and software stack they need to operate most efficiently.

IT decision-makers should look for providers that can deliver end-to-end data center infrastructure at scale, combining high-performance chips, networking, software, and systems design expertise. For example, the right physical racks make it easy to swap in new kit as needs evolve, and having open software is equally important for getting the different pieces of the software stack from different providers to work together.

In India, where competition is fierce and speed to market is pivotal, prioritizing x86-based infrastructure is equally critical to ensuring interoperability and compatibility across existing software stacks (reducing integration challenges and improving developer agility).

Advancing the data center

As our reliance on digital technologies continues to grow, so too does our need for computing power. It is important to balance the need for more compute real estate with sustainability goals, and the way forward is to make the most of the existing real estate we already have. This is a big opportunity to think smartly about this and turn an apparent tension into a massive advantage. By using the right computational architecture, data centers can achieve the same tasks more efficiently, making room for the future technologies that will transform businesses and lives.

Authored By~  Manik Kapoor, AI and Data Center Lead, AMD India

More For You

AI-Driven Cybersecurity: Trust the Intelligence, But Train the Human

AI vs fake documents the future of verification

The Best AI Image Generators of 2025

How to Use Google Nano Banana AI to Create Free 3D Figurines That Are Going Viral in India