2025 Global Data Center Outlook

• Artificial intelligence (AI) will continue to build momentum in 2025. The data center industry stands at the dawn of a transformative era, driven by the relentless advancement of AI. Billions of dollars have been invested in AI over the last couple of years, driving demand for more data center infrastructure.

• 2025 will see an acceleration of SMR announcements with the total amount of gigawatts likely to double. Nuclear power is emerging as a preferred solution to meet growing energy demand. As traditional power grids struggle to keep pace, the sector is exploring both traditional large scale nuclear power and small modular reactors (SMRs).

• A shift to liquid cooling will be essential to accommodate GPU advancements. In new construction liquid cooling infrastructure has quickly become the default installation. For existing facilities transitioning to higher density workloads, liquid cooling will be a viable solution and an opportunity for owners and investors to upgrade their assets.

• $170 billion in asset value will need to secure development or permanent financing in 2025. While investment transaction volumes are likely to record only a modest increase in 2025, due to significant bid-ask spreads and lack of available product, development finance will present a compelling opportunity for investors to gain exposure to the sector.

The data center industry stands at the precipice of a transformative era, driven by the relentless advancement of artificial intelligence. This technological revolution is not merely evolving the digital infrastructure landscape; it's fundamentally redefining it.

The applications of AI are expanding across nearly all industries. Billions of dollars have been invested in AI over the last couple of years which is driving demand for more powerful and efficient data center infrastructure. As result, global data center construction currently stands at record levels. All signals suggest that AI demand will continue to build momentum in 2025.

At the core of the AI revolution is the rapid advancement in semi-conductor technology. A computation task that used to take 32 hours to perform can now be accomplished in just one second with the latest GPU technology. Given the increasing processing speeds, AI programs are able to train on larger and larger data sets. The ability to train, iterate, and improve AI models at faster and faster speeds is making the entire AI ecosystem more valuable. The pace of AI innovation will continue to accelerate with each new generation of GPUs.

Power infrastructure bottlenecks are a major impediment to data center development. Power scarcity garners most of the headlines, but equally as significant are the extended timelines required to build transmission lines. These challenges will continue to intensify as the data center sector expands rapidly into new geographies.

In many markets it can take four years or more to have high-capacity power lines extended to new development sites. Most of this delay is associated with securing easements and regulatory approvals. Supply chains continue to be challenged, particularly for transformers and switchgear, but equipment procurement is not the primary reason for transmission delays. These challenges have led to a shift in site selection criteria, with land now being evaluated based on available power capacity and proximity to transmission lines, rather than pricing or total acreage.

Utilities are now more selective in Purchase Power Agreement (PPA) approvals, using thorough intake forms and application fees to filter out speculators. This is generally seen as positive for the industry as it focuses limited resources on serious, well-funded projects. However, it does not address the fundamental issue of long lead times for infrastructure development.

Nuclear power is emerging as a preferred solution to meet the growing energy demands of data centers, particularly for AI and high-performance computing applications. Tech companies are the largest occupiers of data center space, and they have among the most aggressive net zero targets. Nuclear provides a solution to both challenges.

As traditional power grids struggle to keep up with increasing power demands, nuclear energy is gaining enthusiasm as a clean and reliable alternative energy source for data centers. Multiple nuclear PPAs were signed in 2024 involving active nuclear plants as well as decommissioned plants which will be reactivated around 2028. However, the development pipeline for additional large-scale nuclear facilities is limited. Instead, attention has turned to another form of nuclear energy: small modular reactors (SMRs).

SMRs can provide 1.5 to 300 megawatts of power. They are modular and scalable, making them potentially an ideal solution for data centers at a fraction of the traditional large-scale nuclear cost. It is important to note that SMR technology is still in the early stages of development, with commercial deployment in the U.S. not expected until 2030 at the earliest. However, if SMRs materialize as a credible power alternative, they could provide data centers with abundant green energy. This would have significant implications for site selection, data center design and ongoing operations.

The surge in power density necessitates a shift in thermal management strategies. NVIDIA's latest AI chips consume up to 300% more power than their predecessors, and industry forecasts suggest that global data center energy demand will double in the next five years.

Liquid cooling is becoming essential for high-density racks. A hybrid approach is typically employed today, with 70% liquid cooling and 30% air cooling. Liquid cooling installations are currently taking the form of rear door heat exchangers (RDHx) and direct-to-chip (DTC) technologies. Within new construction, liquid cooling infrastructure has quickly become a default installation. Additionally, RDHx and DTC retrofits are considered a viable solution for existing facilities transitioning to higher-density workloads.

Immersion cooling will likely become a common thermal management strategy as GPUs push above 150 kW per rack. But broad implementation of the technology is still a few years away. The global average rack density is currently only 12 kW and immersion cooling is implemented in less than 10% of data centers today. Deployments of immersion cooling over the next few years will be concentrated in AI facilities and in sections of traditional data centers running AI workloads.

While the technology promises many benefits, including highly efficient cooling, it also faces challenges related to liquid quality, reliability and maintenance. Additionally, immersion cooling introduces new challenges in structural design due to weight considerations. The weight of the largest cooling baths can reach up to four metric tons when filled with equipment and cooling fluid, which requires significantly reinforced flooring. This will need to be addressed in data centers implementing immersion cooling on a large scale.

Data center development financing will achieve another record year in 2025. Across the hyperscale and colocation segments, an estimated 10 GW is projected to break ground globally in 2025. Separately, 7 GW will likely reach completion. This equates to roughly $170 billion in asset value that will need to secure either development or permanent financing in 2025.

Data center development financing is typically arranged at 65% to 80% loan-to-cost while permanent financing is typically arranged at 65% to 75% loan-to-value. The majority of data center development financing has historically been originated by a handful of lenders, but as the deals get larger, the lending pool is slowly increasing and club deals are becoming more common.

A relatively limited number of data centers trade each year. For context, global data center investment sales (excluding entity trades and recaps) have averaged just $7 billion annually since 2020. This compares to an annual average of $241 billion for office assets over the same period.

An increased number of developers will be looking to exit positions and recycle capital in the year ahead. But in many cases, these developers will have challenges making the numbers work. As result, global data center trading volume is likely to record only a modest increase in 2025.

The modest increase, despite significant investor interest, is due to several factors. First, the significant upward move in interest rates in most G10 countries between the time assets were financed a few years ago and today has created a significant bid-ask spread. Second, there are variances in relet assumptions between buyers and sellers which are contributing to bid-ask spreads. Most owners are not in a position where they need to sell, so they are likely to hold on to assets over the next year and wait for financial conditions to become more advantageous.

As we look ahead to 2025, it is clear that the data center sector is on the precipice of enormous, transformative growth driven by the rapid advancement of AI and its increasing compute demands. This growth is creating both opportunities and challenges, the convergence of rising power requirements and data center growth is leading to a generational investment opportunity. The emergence of new technologies provide additional latitude for sustainability growth and value creation, but also highlight the fundamental importance of expert advice and guidance in this rapidly evolving sector.