The Hardest Part of Building a Data Centre Isn’t the Building Itself

Photo Credit: Geoffrey Moffett on Unsplash

Recently, while conducting research for a client engagement, I had the chance to speak with several highly experienced experts across the data centre and digital infrastructure ecosystem. And one observation really stood out for me.

Behind record-breaking investments in the billions, and relentless expansions, lies a more complex reality that seldom makes the front pages. The DC boom is colliding head-on with operating realities and hard physical limits.

Not every data centre will achieve full utilisation. Not every operator will achieve sustainable returns at the velocity they anticipated. And not every market will evolve in the way their business case assumptions were built upon.

In many ways, survival bias exists within the data centre industry as much as it does in any other high-capital infrastructure sector. The barriers to entry are high, and so are the risks.

What struck me most during these conversations was that many of the challenges had very little to do with constructing the facility itself.

In reality, the building itself may turn out to be the easiest part.

In reality, the building itself may actually turn out to be the easiest part.

The Real Constraint Is Power

Data centres have enormous appetites for electricity, and AI is accelerating demand even further. The International Energy Agency (IEA) recently projected that global electricity consumption from data centres could more than double by 2030, driven largely by AI workloads and continued digitalisation.

Even some of the world's largest power systems are beginning to feel the strain.

North America's PJM system, which serves key data hubs like Northern Virginia, has warned that it has "years, not decades" to fundamentally change operations under an unprecedented surge in demand.

At the same time, hyperscalers are securing power in gigawatt-scale quantities. What was once a facility-level consideration is rapidly becoming a strategic infrastructure discussion involving utilities, governments, and long-term energy planning. In some markets, the real challenge is actually finding electricity.

Recent announcements reinforce how power is becoming a key consideration for the industry.

DayOne recently secured more than 1GW of renewable energy capacity through agreements with Tenaga Nasional Berhad (TNB), while Seraya Partners signed an MOU with TNB to explore cross-border green energy opportunities.

Power Is Only The Beginning

"It's not just energy, there's also water!" exclaimed a friend of mine who is deeply involved in the DC sustainability space.

Data centres do not operate in isolation, and they depend on an entire ecosystem of supporting infrastructure. Water is one example.

As computing density increases, so does the challenge of cooling equipment efficiently. Technology providers and operators are actively exploring new cooling approaches such as liquid cooling, direct-to-chip, and immersion cooling to reduce the water intensity of AI infrastructure. But these technologies come with different trade-offs, and the underlying reality remains the same: the heat generated by computing has to go somewhere.

This has brought greater attention to water consumption, cooling efficiency and long-term sustainability. In regions experiencing rapid data centre growth, governments are placing greater emphasis in examining whether supporting infrastructure can keep pace.

Johor, one of ASEAN's fastest-growing data centre hubs, has reportedly begun increasingly scrutiny around water sustainability, infrastructure readiness and long-term resource planning as investment activity accelerates.

Connectivity is equally important. Building a data centre without sufficient fibre connectivity, carrier diversity, cloud on-ramps and access to subsea cable ecosystems is somewhat like building an airport without enough flight routes. And that's only the physical layer. At the logical layer, many AI workloads generate significantly more east-west traffic within data centres than traditional enterprise applications, increasing demands on bandwidth, latency and network architecture.

The facility design can be impressive, but that alone doesn't create digital gravity as customers aren't just evaluating rack space. They're evaluating resiliency, connectivity, latency, interconnection options and the long-term viability of the surrounding infrastructure ecosystem. In many cases, the ecosystem matters as much as the building itself.

Resilience Comes At A Cost

Another angle that became clear from my conversations is that resiliency itself comes at a cost.

Higher availability requirements often meant duplicated systems, reserve power, additional cooling capacity, excess fibre routes and spare operational buffers. These are necessary for uptime, but they also increase land usage, energy consumption and operating complexity.

Environmental considerations are becoming increasingly important too. Site selection today extends beyond land availability and commercial viability. Factors such as water availability, flood risk, heat exposure, energy security and long-term infrastructure readiness are becoming just as relevant.

As AI workloads continue to grow, policymakers, operators and communities are paying closer attention to the broader environmental footprint of digital infrastructure.

Environmental and community concerns are also beginning to surface in some markets. As facilities become larger and more concentrated, questions around land use, water allocation, noise pollution, infrastructure strain and local economic benefit are becoming more common.

Digital infrastructure may feel virtual, but it remains deeply physical in reality.

Data Centres Are Becoming Strategic Infrastructure

Another shift is that data centres are being view more strategically than purely as real estate or infrastructure assets.

Governments are paying closer attention to issues such as data sovereignty, cybersecurity, cross-border data flows, foreign ownership and national resilience with the rise of AI is accelerating this trend.

At ATxSummit 2026, Singapore's Minister for Digital Development and Information, Min Josephine Teo, spoke about the country's next phase of AI adoption: moving beyond pilots towards broader national AI missions across sectors such as healthcare, manufacturing, finance and connectivity.

This reflects a critical perspective that AI is more than an application story. It requires trusted infrastructure, resilient ecosystems, connectivity, governance and deployment pathways capable of supporting adoption at scale.

The discussion is shifting from how many data centres should be built, to whether the broader infrastructure ecosystem can support what comes next.

Not All Demand is Equal

Commercially, not all DC demand is created equal.

AI workloads behave differently from traditional enterprise colocation requirements. Edge computing requirements differ from hyperscale architectures. Regulatory localisation requirements vary across markets.

The demand does not automatically translate into successful investments because the market rarely work that neatly. Timing, ecosystem maturity, power availability, connectivity, operational discipline, infrastructure readiness, and even geopolitics often matter just as much as the broader demand narrative. As capacity becomes concentrated within a handful of massive campuses, concentration risk also emerges, with a larger proportion of digital infrastructure dependent on the same power, network and physical environment.

Data centres are highly capital-intensive and requires significant upfront investment years before meaningful revenues are realised. Developers must secure land, power, financing, permits and anchor tenants while navigating rising construction costs, interest rates and evolving technology requirements. Strong demand does not automatically eliminate execution risk or market changes.

History offers a useful reminder. During the dot-com era, companies such as Exodus Communications and Globix expanded aggressively to meet surging internet demand, only to succumb to debt burdens and changing market conditions before the industry eventually matured. More recently, Cyxtera Technologies filed for Chapter 11 bankruptcy despite operating more than 60 data centres globally. This illustrates that scale alone does not guarantee commercial success. As AI infrastructure investment accelerates, disciplined capital allocation, tenant quality and long-term financial sustainability may prove just as important as demand itself.

Looking Beyond The Building

There is also an emerging recognition that not every workload needs to reside within a full-scale data centre.

As AI inferencing, industrial automation, autonomous systems and latency-sensitive applications continue to grow, edge and distributed infrastructure may increasingly complement hyperscale facilities.

The future may not belong simply to those who build the largest data centres, but to those who best understand how to orchestrate the broader infrastructure ecosystem around them. Data centres are not merely buildings filled with servers. They're deeply interconnected systems sitting at the intersection of power, water, connectivity, sustainability, policy, geopolitics and economics. And in digital infrastructure, the building may ultimately be the easiest part.