As AI accelerates across industries, it is fundamentally forcing business leaders to rethink how to build and operate data centers. In recent years, Southeast Asia (SEA) has seen an influx of investments from global tech giants, underscoring its growing recognition as a hub for AI innovation and capabilities. Markets such as Malaysia, Indonesia, and Thailand, alongside Singapore, have emerged to become prominent areas poised for data center development, seeing a surge in investor interest. This surge in AI implementation overall contributes to rising demand for compute power and infrastructure across the region.

However, meeting AI’s rapid growth requires more than just increasing capacity. It also means solving the real-world and unique challenges that the technology presents.

In today’s AI-driven era, data centers must support massive compute power, faster builds, and higher-density environments, highlighting the need for infrastructure that can keep pace with AI’s growth.

With that said, optical fiber remains the core of this critical infrastructure.

Meeting the new demands of modern data centers

AI is spreading at a pace much faster than the early days of the internet. What took decades to build web-scale applications is now happening in just a few years with the help of AI.

This is creating unprecedented demands on data center infrastructure.

To keep pace, data center design is evolving rapidly. High-performance graphics processing units (GPUs), essential to AI workloads, require more connections between servers than ever before. At the same time, in traditional Central Processing Unit (CPU)-based data center designs, power and cooling limitations often mean fewer servers or GPU nodes can be accommodated in a cabinet. The result is a sharp increase in the amount of cabling needed to connect everything, far more than in traditional data centers.  Each server needs high-speed links to switches, storage systems, and management tools, which puts enormous pressure on the network.

Matias Peluffo @ CommScope: Scaling AI effectively requires a redesign of how data is moved, power is delivered, and heat is managed — all while maintaining speed, reliability, and scalability.

To put this into perspective, NVIDIA’s DGX SuperPOD, a leading example of AI infrastructure, contains 32 GPU servers connected to 18 switches in a single row. That setup alone requires 384 x 400GE fiber links just to handle data movement across the cluster, not including additional connections for storage and management. It’s a remarkable increase in the volume of fiber cabling required inside the data hall, and a clear example of why traditional network designs are no longer sufficient.

This increased density demands new approaches to energy management and cooling to handle higher power consumption and heat output, while supporting continuous, heavy internal data traffic across highly interconnected systems.

Labor shortages – both in large-scale data centre construction and in day-to-day operations – are also slowing deployments and making it harder to maintain these increasingly complex environments.

Together, these pressures mean data centers can no longer rely on traditional design or incremental upgrades. Scaling AI effectively requires a redesign of how data is moved, power is delivered, and heat is managed — all while maintaining speed, reliability, and scalability.

This is where the network becomes critical. Fiber, at the same time, remains at the heart of that network, providing the essential foundation for AI-ready data centers.

Fiber: The foundation for an AI-ready data center

With the network serving as a key enabler for seamless data movement, fiber plays a crucial role in setting the stage for AI-ready data center operations.

AI workloads rely on high-speed, low-latency connections between GPUs that work together, as well as scalable networks that can support thousands of GPUs in rapidly emerging AI factories. Poor infrastructure design causes slowdowns, higher expenses and limited scalability.

Consequently, evolving and strong fiber networks are essential to satisfy these demands. Today’s data centers require denser, higher-capacity fiber solutions that can handle massive data flows while maintaining the same physical footprint. Fiber infrastructure needs to be faster and simpler to install, allowing skilled labor to work more efficiently amidst pressure to accelerate deployments. In response, the industry is moving toward pre-terminated fiber solutions that offer plug-and-play systems, enabling faster installations and reducing operational disruptions.

Staying ahead means understanding where the market is going, especially in fast-growing regions like SEA, where there is an intensifying race to build AI-ready data centers. Increasingly, data center operators deploying large AI clusters are shifting from point-to-point cabling to more scalable structured cabling systems. The choice between single-mode and multimode fiber remains key, based on the size and design of the facility. At the same time, InfiniBand and Ethernet continue to play critical roles in managing different types of AI traffic.

At the same time, the industry is pushing towards even greater speeds and capacity to support next-generation performance, moving from 400 Gb/s to 800 Gb/s, 1.6 Tb/s, and beyond, utilizing 8 and 12 fiber connectivity. This is complemented by industry developments for higher speeds, which are pointing towards 16-fiber connectivity in the future.

Given the AI climate, fiber is not just part of the solution – it is the foundation for what comes next. The data centers that lead in this era will be those that can move more data at higher speeds, maximize, and scale seamlessly to support future AI technologies and architectures.

As AI continues to redefine industries, a high-performance optical fiber foundation is the critical infrastructure that will ensure data centers are built not just for today’s demands, but for the future of AI-powered innovation.

The views and opinions expressed in this article are those of the author and do not necessarily reflect those of CDOTrends. Image credit: iStockphoto/kynny