2026 Trends in Data Center Services & Infrastructure

hyperscalers and leading generative AI firms have engaged in what can only be described as an AI arms race. OpenAI, Google, xAI, Microsoft, Amazon, Oracle, Meta, Alibaba and others have committed hundreds of billions of dollars to train large language models in the hopes of gaining first-mover advantage and being recognized as having the “best model.”

This approach is exceedingly resource-intensive: Training a top LLM can use tens of thousands of graphics processing unit chips, each requiring 5-8 times (or more) the energy of other chips. These must be housed in data centers, along with data storage gear, other computational hardware and networking equipment. This AI demand has been particularly strong in North America so far, as most generative AI models have been developed in this region. However, firms have been laying the groundwork to develop large-scale AI data centers outside the US.

Although some “prospector” data center builders are included in this forecast, as well as some duplicate plans for expansion that may not materialize or be delayed, we expect strong data center growth to continue globally in the immediate future.

Multiple factors determine data center location, and several of these are facing growing constraints. The first is related to AI use cases. There are relatively few location requirements related to AI workloads — AI model training can generally be carried out quite a distance away from where end users are located, for example. However, we are tracking AI use cases to monitor the need for AI inference or training that requires rapid connectivity or specific data constraints (e.g., due to data sovereignty requirements) and would need to be in a specific location, such as near urban centers.

Another location factor is available power. Firms expecting to benefit from AI infrastructure demand, such as hyperscale cloud providers, are seeking to lock in as much power for future data centers as possible to avoid having their growth restrained by energy availability and potentially losing out to competitors that have access to more power. However, it is unclear exactly how much power will be required. This creates challenges for energy utilities, particularly if they need to build additional generation capacity and transmission lines (which can take years) to meet this expected demand.

Many of the most desirable data center markets already face constraints related to power, including potential shortages of generation capacity or challenges in building transmission lines. These include PJM (which covers top data center markets such as Virginia, Ohio and Pennsylvania) as well as ERCOT (Texas). Data center developers increasingly face a choice between waiting years for a grid connection in a desirable location, seeking a less desirable site with faster access to grid power, or using behind-the-meter power (e.g., constructing a natural gas-fired power plant).

Additional location factors include electricity costs, taxes, regulations, local resistance to data centers, availability of fiber, type of power available (e.g., from renewable sources), water access, suitable land and the ability to plug into district heating systems or other off-takers for heat. We are developing a model to weigh the trade-offs among data center locations to build scenarios for growth by area and the resulting market for behind-the-meter power options.

Organizations have invested massive amounts of capital in AI infrastructure so far, with continued eye-watering amounts expected, even without factoring in the costs of AI chips, servers and electrical infrastructure, such as transmission and generation. In the US alone, we project average spending on data center construction of more than $70 billion per quarter from 2025 to 2028.

Numerous firms are hoping to enter the sector, and those with access to capital and experience in securing land, power and permits will be better positioned for success. New entrants can add competition in some locations, but they can also offer data center providers the opportunity to sell facilities that are already leased — with predictable revenue and appeal to investors — freeing up capital to build new facilities elsewhere. Vertically integrated models are also emerging, with energy companies, for example, entering the data center industry. Business models are evolving to become more specialized in managing capital, obtaining permits or power and serving particular customer segments. We will continue to explore how these data centers will be funded, how the industry is evolving and the business models that are emerging as a result.

Top AI and cloud firms have maintained their renewable energy targets, as have many of the firms that build or lease data centers to these clients, even as their annual electricity consumption rises. Although many technology firms claim to offset 100% of their energy use by procuring renewable energy, they also have data centers that rely on carbon-emitting power generation. This is expected to increase in many locations, as added electricity demand requires utilities to delay the retirement of coal generation or add gas-fired power plants.

Additionally, data center firms themselves sometimes build their own gas-fired power plants if the grid cannot accommodate their needs. As a result, there has been a clear trend toward greater renewable procurement among surveyed companies, according to the “Global Datacenters Clean Energy Sourcing Strategy – 2025 update,” with the share of power sourced from renewables estimated at 58%, versus about 50% in the 2024 report.

Power purchase agreements remain the primary method for renewable procurement, supplemented by unbundled certificates and green tariffs. While solar is a major source of supply, there is a growing emphasis on hybrid and nuclear solutions.

However, in many areas, even if sufficient renewable or low-carbon generation is planned, a mismatch may occur between when the power-hungry data centers need that energy and the time it takes to add that capacity (e.g., from nuclear power). We will continue to work with teams across S&P Global to track data center growth and its impact on both utilities and renewable energy.

As data center power demand continues to rise, the tech industry may provide financial support for expanding generation across the board, including both renewable and non-renewable resources. But the need to offset increased emissions could drive investment into higher-cost technologies such as nuclear, carbon capture and storage, and battery energy storage systems.

AI workloads typically require high-density infrastructure at the data center level; however, this high-density equipment can generate a large amount of heat, testing the limits of standard data center cooling systems. Until recently, most IT equipment has been cooled using fans to pull cold air past the hot elements to remove the heat. However, as these components become smaller and hotter, they can reach the limit of what fans and cool air can do. Liquid is a more efficient cooling medium than air and has been used for some high-performance computing and advanced modeling, but it has not been broadly adopted in the data center industry for a variety of reasons. That may be changing. Our survey of enterprise data center decision-makers reveals that 21% of respondents plan to shift to liquid cooling over the next year, up from 13% in 2024’s survey, with another 25% planning to switch over the next two to four years.

It helps that technological innovations have made liquid cooling easier to use and retrofitting operational data centers less costly. Most importantly, the increase in high-density workloads, combined with requirements that data centers be as efficient and sustainable as possible, is set to make liquid cooling a more logical approach than air cooling. After years of adoption for specialized uses, the combination of higher-density servers, liquid cooling innovation and sustainability benefits may finally lead to broad adoption.

In addition, the industry is looking at other technological advancements in backup systems, such as using alternatives to lead-acid or lithium-based batteries, using diesel alternatives for generators or replacing generators altogether with hydrogen fuel cells or renewable power sources. Data center firms are also exploring the reuse of heat by plugging into local district heating systems, as well as investigating ways to reduce embodied carbon. We plan to examine and compare many of these proposed technologies in the year ahead.

AI inferencing, the process that a trained AI model uses to analyze new data, creates an unprecedented opportunity for edge computing. Inferencing must run somewhere, and every computing venue is a candidate, from processors on devices such as cell phones to large-scale public cloud venues. While some inference will certainly run on public cloud, some will also likely run closer to end users, at the edge. Much will depend on the use cases for AI: Applications that need to respond in real time will be more likely to run edge inferencing. However, the edge may also be attractive for enterprises that want more control over data. Enterprises have already noted in our surveys that top reasons for processing AI tasks at the edge include security and data sovereignty, not just performance requirements. Key questions for enterprises will be: Which venue will provide the necessary AI inferencing performance at the lowest cost and with the least complexity?

Network growth and innovation will also affect what data will be processed and stored at the edge and what will be moved elsewhere, as will changes to the infrastructure required to support these deployments (e.g., compute, storage, content delivery, accelerators, power, footprint, noise level, ruggedization).

Edge vendors or operators will need to satisfy requirements for low maintenance, space constraints and power conservation. Still, the use cases will vary enormously, so it may be hard for vendors to gain scale in an atomized market. The ecosystem of vendors, operators, financiers and network providers at the edge is evolving rapidly, so we will continue to follow this market.

The cryptocurrency mining industry continues to add data center capacity, in many cases taking advantage of renewable energy or stranded energy, thanks to the relative flexibility of its workloads. Most miners build and operate their own data centers, and some are considering offering these capabilities to other enterprises, particularly those with similar computing profiles and requirements, such as firms with high-performance computing (HPC) and AI workloads. However, when mining companies build facilities specifically for AI or HPC workloads, they are typically closer to more traditional data center builds (e.g., with backup power and some redundancy of equipment) than mining facilities, largely due to client requirements. Thus, cryptocurrency firms are in some cases becoming competitors to traditional leased data center providers. We continue to monitor the cryptocurrency industry and its requirements, as well as blockchain, and the impact both have on the data center industry.

S&P Global Energy Horizons 451 Research provides essential insight into key trends driving digital transformation across the entire technology landscape. By offering a combination of expert analyst insight and differentiated data, 451 Research enables the industry with the information and perspectives they require to make more effective decisions.

Reports such as this offer a holistic perspective on key trends and themes driving the technology space over the coming year. These markets evolve quickly, so 451 Research offers a wide range of research services that provide critical marketplace updates on an ongoing basis. These reports, datasets and perspectives are published frequently, in numerous short- and long-form formats. Forward-looking M&A analysis and perspectives on strategic acquisitions and the liquidity environment for technology companies are also updated regularly, backed by industry-leading databases such as the 451 Research M&A KnowledgeBase.

Our research is organized into channels that align with the prevailing key issues driving digital transformation. These channels are: Applied Infrastructure & DevOps; Cloud & Managed Services Transformation; Customer Experience & Commerce; Data, AI & Analytics; Data Center Services & Infrastructure; Fintech; Information Security; IoT, Edge & Digital Industries; and Workforce Productivity & Collaboration.

For more information about 451 Research, please go to: spglobal.com/451research.

‘Bring your own power’ is an option for those that cannot wait, November 2025

2026 US Data Centers and Energy Report, November 2025

What happened to the data center slowdowns? October 2025

The Nordics: Sweden Leased Data Center Market, October 2025

State-level data center policies, then and now: Part 1, the major markets, October 2025

India: Leased Data Center Market, October 2025

Asset-backed securities pave the way for a new data center financing model in Europe, September 2025

Air cooling remains prevalent, but liquid cooling is gaining momentum – Highlights from VotE: Datacenters, September 2025

Beijing-Tianjin-Hebei datacenter market remains one of the largest globally, July 2025

Truths about how the power sector can (and cannot) respond to datacenter needs, April 2025