AI Expansion Sparks Data Centre Overbuild Fears

The AI boom is pushing data centres to their limits as hyperscalers invest heavily to meet growing computing demands.

According to a report from Moody’s Ratings, hyperscale data centre capacity could expand by 20% annually after 2028.

Yet, the same report cautions that the sector faces complex risks, including overbuilding, supply chain volatility and escalating financial pressures.

Data centres, once predictable in their development cycles and operational demands, are now being rapidly re-engineered to support the intensive needs of AI.

As developers race to keep pace with surging AI-driven workloads, questions arise about the long-term viability of this accelerated growth, especially when future computing demands are not yet fully known.

Moody’s warns that power infrastructure delays or shortages could force capacity growth back to lower projections of 5% or 10%.

In this context, the challenge is twofold: delivering enough capacity for near-term AI growth while avoiding stranded investment in underutilised facilities.

The report says: “Surging growth in hyperscale data centre capacity will eventually level off, but identifying that inflection point has become increasingly difficult as AI data centre campuses emerge as another key growth driver.”

New pressures redefine the data centre blueprint

AI workloads have changed the scale and nature of what a data centre needs to deliver.

These campuses, now tailored to machine learning and large-scale model training, consume as much electricity as a medium or large city.

With that comes a need for reliable, low-cost power, which often pushes hyperscalers to remote locations where such infrastructure can be built.

These locations, however, pose construction risks.

Skilled labour may be hard to retain year-round and transport costs add further complications.

This adds a layer of uncertainty to project timelines and financial planning.

Unlike legacy facilities, AI data centres require entirely new infrastructure to manage increased power density and heat.

Cooling becomes a major technical challenge, with traditional systems falling short in high-density environments.

As Rajesh Sennik, Head of Data Centre Advisory at KPMG UK, told Data Centre Magazine at the end of 2024: “The industry is now facing unprecedented demand for new infrastructure solutions to efficiently power, cool and support this next generation of compute and as a result, AI is fundamentally reshaping the architecture of IT infrastructure.”

Hyperscalers like Meta and OpenAI are betting on centralised mega-projects to address these demands.

Projects such as the 5.6GW Wonder Valley site in Alberta and Meta’s 2GW campus in Louisiana are set to deliver massive compute power by 2029.

These investments reflect the urgency to support AI, but they also risk outpacing the actual demand or misaligning with evolving tech specifications.

Moody’s acknowledges that capital reallocation is inevitable.

Hyperscalers continue to reassess whether current developments match real-world needs, particularly as AI innovation continues to shift computing priorities.

The report says: “As investment continues to pour into new data centres, some level of capital reallocation and retrenchment is inevitable.

“The hyperscalers that have been spurring the market's expansion continuously right-size their newly leased and owned capacity under development because much of this new capacity is being built in anticipation of future needs.

“As it becomes available, it may exceed or fall short of a hyperscaler’s current needs.”

Financial risks deepen as AI scales up

AI’s hunger for power and compute also comes with steep capital requirements.

Gen AI models demand more processing capacity and infrastructure than past technologies and that means greater upfront investment.

Data centres must now be designed with precise load requirements and many are developed as turnkey facilities for a single tenant — bringing further risks if future needs change.

Moody’s points to growing rack densities, averaging 12kW per rack in 2024, driven largely by the ongoing evolution of GPU racks.

As Nvidia continues to release new versions, data centres must plan for power densities that may soon reach between 1MW and 5MW per rack.

At the same time, unexpected developments can disrupt forecasts.

The emergence of Chinese start-up DeepSeek earlier in 2024 demonstrated that even less advanced chipsets can power competitive AI models.

Such breakthroughs, while welcome, create risk for hyperscalers who have already committed large capital investments in next-gen infrastructure.

Geopolitical tensions also add financial strain.

In the US, tariffs on key materials used in data centre builds are driving up costs.

Structural steel, electrical systems and rare earth minerals are all affected, while limits on exports of certain materials introduce further delays and uncertainty.

Developers must now factor in fluctuating tariffs when planning project timelines and budgets.

Moody’s underlines that while well-funded hyperscalers are likely to weather the storm, profit margins could narrow, particularly if trade policies shift again.

The report adds: “Abrupt changes in tariff policies increase uncertainty, making it challenging for data centre developers to forecast costs and timelines accurately.”

The data centre industry now stands at a crossroads.

While AI continues to create demand for more capacity, the infrastructure required is evolving faster than ever before.

Without a clear understanding of future workloads or supply chain conditions, balancing investment with long-term stability remains an urgent challenge for the global technology sector.

Explore the latest edition of Technology Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.

Discover all our upcoming events and secure your tickets today.

Technology Magazine is a BizClik brand