The Infrastructure Behind AI Data Centers

Nearly 100 gigawatts of new data center capacity is projected to be added globally between 2026 and 2030, roughly doubling global capacity, according to JLL. AI and hyperscale cloud expansion are helping drive that growth.

But more capacity is only part of the story. AI workloads are changing the physical infrastructure inside data centers, from the power feeding a facility to how operators cool and connect increasingly dense computing systems.

Let’s take a closer look at the infrastructure behind this new generation of AI data centers.

DOWNLOAD: Insight Global’s Guide to Data Centers

AI Has Changed the Blueprint for Data Center Construction

For decades, data center construction has revolved around a familiar set of priorities: dependable power, effective cooling, connectivity, physical security, and the ability to keep critical systems running.

Those principles haven’t disappeared. AI has raised the infrastructure requirements behind them.

AI workloads can require high-density racks that place greater demands on power and cooling. Power availability, high-density racks, advanced cooling, electrical infrastructure, structured cabling, and increasingly sophisticated mechanical, electrical, and plumbing (MEP) systems are among the factors making facilities more complex.

The biggest constraint can begin before construction does: access to power. Power availability and grid infrastructure constraints are already influencing data center site selection and development timelines, making decisions around utilities, feasibility, permitting, and long-lead equipment increasingly important during pre-construction.

Cooling is changing alongside power. The increased importance of liquid cooling for AI data centers means mechanical infrastructure has to accommodate technologies such as direct-to-chip and immersion cooling in high-density environments. AWS, for example, is developing infrastructure specifically for deploying liquid-cooled racks at scale and aligning those cooling systems with future server hardware and data center infrastructure.

A Decade of Data Center Technology Change

Over the last decade, data centers have evolved from infrastructure built primarily around enterprise and cloud workloads to increasingly specialized facilities capable of supporting high-density AI workloads. As computing requirements have changed, the relationship between the hardware and the facility surrounding it has become much tighter.

That shift means construction teams have to think about the AI workloads the facility will support much earlier in the process. Electrical capacity, cooling, connectivity, and rack density increasingly influence decisions across design and construction rather than being considerations reserved for the finished data hall.

The Core Infrastructure of an AI Data Center

Successful AI data center architecture connects several infrastructure systems that have to support demanding AI training and inference workloads.

High-density compute

GPUs and other accelerators provide the computing capacity behind modern AI systems. At large scale, those resources operate as clusters rather than isolated servers. NVIDIA describes AI factories as infrastructure designed to coordinate compute, networking, memory, storage, software, power, and cooling around AI workloads.

Power infrastructure

AI increases both the scale and complexity of the power conversation. Electrical systems have to support the computing environment, while available utility capacity can influence where developers build in the first place. Power availability and grid constraints as factors affecting development timelines and site selection.

Cooling and thermal management

High-density AI hardware is pushing traditional air cooling beyond its limits. Google Cloud says next-generation AI and high-performance computing chips routinely exceed 1,000 watts of thermal design power, which standard air cooling cannot manage alone. In response, Google developed Brazos, a rack-mounted liquid-to-air system designed to bring liquid cooling to high-density equipment in existing air-cooled data centers.

High-speed networking

AI workloads depend on fast connectivity between computing resources, placing greater demands on the fiber and cabling infrastructure connecting them. The Fiber Broadband Association reported in early 2026 that bandwidth purchased for data center connectivity increased nearly 330% between 2020 and 2024, reflecting the growth of hyperscale and AI-related traffic and the need for robust fiber infrastructure.

Together, these requirements create a much tighter relationship between the technology being deployed and the facility being built around it.

Hyperscalers Are Expanding the AI Data Center Model

Hyperscalers sit at the center of this shift because their scale makes them major builders and operators of the infrastructure AI requires.

Consider AWS. The company is redesigning its global infrastructure around increasingly demanding AI workloads, with changes spanning power, cooling, networking, and data center design. Its Project Rainier AI compute cluster alone uses nearly 500,000 Trainium2 chips, while AWS and NVIDIA announced plans in August 2026 to deploy 2 million additional NVIDIA GPUs across AWS infrastructure in 2027 and 2028.

Google Cloud offers another example. Its 2026 AI infrastructure portfolio combines purpose-built computing hardware with specialized networking and storage, including the new Virgo Network, an AI data center fabric designed to connect AI infrastructure at massive scale.

For construction leaders, that matters because advances in compute are increasingly connected to decisions about the buildings, power systems, cooling equipment, controls, and networks required to support them.

AI Infrastructure Is Changing Data Center Jobs

Many roles behind AI data center construction aren’t new. Across the data center lifecycle, here are some of the key roles today:

Data Center Design Engineers
MEP Superintendents
Electrical Engineers
Mechanical Engineers
Controls Engineers
Network Engineers
Commissioning Engineers
Critical Facilities Engineers
BMS/SCADA Engineers

What is changing is the specialization around AI. For example, Apple was recruiting an AI Data & Infrastructure Engineer in September 2026, while Google has AI Infrastructure Customer Engineer and Field Solutions Architect roles focused on AI accelerators and large-scale infrastructure. AWS is also recruiting roles dedicated to liquid-cooling infrastructure, including engineering expertise spanning liquid cooling products, coolant chemistry, materials, and thermal management.

These examples don’t prove that AI created every one of these professions from scratch. However, they do show how employers are developing more specialized roles at the intersection of data centers, accelerated computing, AI, and advanced cooling.

AI Data Centers Support More Than Hyperscalers

Hyperscalers may lead much of the infrastructure expansion, but they aren’t the only organizations using AI infrastructure.

AI infrastructure supports enterprises across many different industries, including healthcare, life sciences, manufacturing, telecom, financial services, and retail. NVIDIA’s 2026 industry research included more than 3,200 respondents and found that 64% said their organizations were actively using AI.

That distinction matters. A hyperscaler may own and operate the physical infrastructure, while customers across industries consume that computing capacity through cloud and AI services. Organizations may also build or operate dedicated infrastructure based on their workloads, security requirements, performance needs, and business models.

Building AI Infrastructure That Can Keep Up

AI is changing the data center from the rack outward. Higher-density compute places new demands on power and cooling. Those requirements affect design and construction. Hyperscalers are pushing that infrastructure forward at enormous scale, while specialized skills are developing around the systems required to build and operate it.

For leaders responsible for data center strategy, the challenge is connecting those pieces across the lifecycle. Insight Global brings talent, consulting, and AI capabilities together to help organizations build the people, technical expertise, and solutions needed across data center construction, engineering, commissioning, connectivity, critical facilities, and operations. Whether you need individual experts, specialized teams, or support solving a defined infrastructure challenge, we help turn ambitious data center plans into execution.