AI boom drives global data centre growth despite soaring costs: Moody's

Global data centre expansion accelerates to meet AI and cloud demands; challenges include rising energy needs, supply chain disruptions, and power infrastructure constraints

The global data centre sector is undergoing rapid expansion, fuelled by surging demand for artificial intelligence (AI), cloud computing, and digital services. Despite ongoing high capital expenditures and operational challenges, the sector’s fundamentals remain resilient, according to latest report by Moody’s Ratings.

 

Surge in hyperscale data centre capacity

Hyperscale data centre (hyperscalers) capacity is surging, driven by the rapid development of AI models and services. A significant portion of this new capacity — either recently completed or underway — is tied to AI-specific infrastructure. These AI data centres, often pre-leased to financially stable tenants, carry lower investment risk but require more power and cooling due to dense GPU-based architectures. An analyst at Moody’s noted, “The capital cost of power generation is much lower than the cost of the data centres and computing equipment they’ll support.” 

To meet soaring energy demands, hyperscalers are shifting toward large-scale AI campuses — centralised hubs with up to 5 GW of capacity. These sites are strategically located where new, low-cost power generation can be rapidly developed, often in remote areas. Projects like Meta’s 2 GW facility in Louisiana and the 5.6 GW Wonder Valley site in Alberta exemplify this trend. 

However, growth beyond the year 2028 is harder to predict due to power infrastructure constraints. Scenarios range from moderate 5 per cent to a very rapid 20 per cent annual growth, depending on how fast new power generation is built. While remote locations offer cost stability, they come with higher construction and operational risks, including limited workforce access and long-term adaptability concerns.

 

Rationalising capacity amid demand shifts

As data centre investment accelerates, hyperscalers are increasingly rationalising capacity to match evolving demand projections. Much of the newly leased or owned capacity is built in anticipation of future needs, but as it comes online, it may not align with current requirements. “These decisions are being driven by long-term demand forecasts and how quickly a site can secure power and become operational,” an analyst at Moody’s said. 

Delays in obtaining power grid connections have made some regions less attractive for new development. As a result, hyperscalers are beginning to slow or pause early-stage projects. However, such projects typically pose limited credit risk, as initial costs — largely covered by developers or hyperscalers — represent only a small portion of total construction expenses. 

Signs of this market rationalisation include fewer land purchases, power requests, and more upgrades at existing facilities. Approaches vary by company: some hyperscalers lease large amounts of capacity globally to support both current services and emerging AI products, while others continue to build their own infrastructure. Outside the US, leasing is more common due to regional complexities. Meanwhile, smaller tech firms are choosing to lease space and GPUs from colocation providers rather than commit capital as hyperscalers do.

 

Financial risks grow for data centre developers

Developers also face growing financial risks as they invest in capital-intensive data centres to support evolving tenant computing needs, especially in the age of generative AI. Compared to earlier tech shifts like cloud computing and 5G, today’s AI-driven workloads demand much greater upfront investment. This challenge is amplified by the growing pace of innovation and uncertainty around future computing demands, making it harder to project long-term capacity needs. 

Designing facilities to meet these demands — often tailored for single-tenant hyperscalers — comes with risks. “Tenants may not need the space beyond their original 10-to-15-year lease”, and lease renewals may involve major capital upgrades. Expiring tax incentives could further increase tenant costs, influencing renewal decisions. 

The surprise emergence of Chinese startup DeepSeek, which trained a competitive AI model with less-advanced chips, highlights the threat of obsolescence. Developers must constantly upgrade infrastructure to stay relevant.

New hyperscale data centres must support high-density, power-hungry AI operations. Average rack density has doubled since 2016, and new AI centres may exceed 200 kW per rack. NVIDIA’s latest GPUs are pushing toward even higher densities, possibly reaching 1–5 MW per rack. 

Overbuilding poses risks, too. “It may be easier to construct a new facility than adapt an existing one.” Joint ventures and pre-leasing help mitigate such capital exposure.

 

Tariffs, rising costs disrupt data centre timelines

Rising trade tensions and tariffs will disrupt data centre supply chains, increasing costs and causing project delays. “Tariffs on essential materials” pressure financials, especially in IT hardware. While demand from hyperscalers ensures new projects proceed, “rising costs of structural steel and electrical components” complicate timelines, leading to possible contractual changes.