Why AI capacity plans must include data centre infrastructure

AI capacity plans are often expressed in GPUs, servers and cloud commitments. The physical requirement grows with compute capacity: electricity must reach the equipment, heat must be removed, and facilities must keep workloads running as demand changes. Polaris Market Research projects the global data centre physical infrastructure market to increase from an estimated $36.20 billion in 2025 to $128.27 billion by 2034. That puts the supporting physical layer firmly within executive capacity planning.

The infrastructure boom extends well beyond compute

Polaris Market Research forecasts a compound annual growth rate of 15.09% from 2026 to 2034. Its physical-infrastructure category covers systems including power distribution, backup power, cooling equipment, cabling and cabinets. Growth in computing capacity therefore drives demand beyond the servers themselves. Every deployment depends on a facility that can deliver power, manage heat and support reliable operation.

Deployment scale and density shape those physical requirements. Larger facilities and denser equipment can need more supporting infrastructure, influencing where capacity can be installed and the investment it requires. The Polaris forecast measures market growth, making it a useful signal for executives to include physical infrastructure explicitly in capacity and capital plans.

Different digital workloads share a physical layer

Demand comes from several types of digital workload. Polaris associates physical-infrastructure growth with cloud computing and expansion of cloud services across financial services, healthcare, manufacturing and retail. IT and telecom held the leading end-user share in 2025, with investment in 5G networks, streaming services and cloud platforms sustaining demand for the physical systems beneath them. Each workload has a different processing profile but depends on facilities that power, cool and connect computing equipment.

High-performance computing, or HPC, involves intensive processing, while eCommerce can expose operators to fluctuating loads. Polaris also expects healthcare to record one of the faster growth rates over the forecast period as telemedicine, digital health platforms and data-intensive medical applications increase the need for secure, dependable processing environments. Operating requirements vary, but each workload adds demand for supporting data-centre infrastructure.

Smart city programmes extend that demand into distributed infrastructure. Polaris says connected sensors, surveillance systems and smart grids generate data that can require local or regional processing. This can increase demand at core data centres and edge sites, where computing resources sit closer to users or data sources to meet operational or latency requirements. Capacity planning therefore has to cover infrastructure across multiple locations.

For infrastructure leaders, separate business programmes can combine into shared demand for power distribution, backup power, thermal management, cabling, cabinets and facility capacity. Cabling, cable management and cabinets remain part of organised expansion alongside processors and storage. Capital plans need to capture this shared physical requirement across digital portfolios.

AI density raises the importance of power and cooling

Polaris associates artificial intelligence workloads with growing physical-infrastructure investment and identifies facilities designed for artificial intelligence workloads as part of North American growth. Dense computing environments place greater demands on power and thermal design because supporting systems must match the equipment deployed. For CIOs and CTOs, compute capacity and facility capacity belong in the same planning process.

Power systems accounted for 31.9% of the market in 2025, the largest component share measured by Polaris. The category includes uninterruptible power supplies, which provide temporary backup electricity, as well as generators, power distribution units and related equipment. These systems keep facilities operating through grid interruptions and periods of peak demand. Power planning covers supply and resilience.

Polaris also links growth in eCommerce and HPC to the need to handle fluctuating loads while reducing outage risk for digital services and business operations. Changing load profiles affect how computing capacity must be supplied and protected. Servers and accelerators create business value only when the surrounding systems keep them operating reliably.

Cooling creates a related engineering requirement. Traditional air cooling remains part of the data-centre environment, while Polaris identifies liquid cooling and immersion cooling among approaches for high-density environments. It also identifies software tools for managing heat and power. Infrastructure teams can assess these approaches against the density and operating requirements of each planned deployment.

Thermal architecture can become an early capacity-planning decision for dense computing. A cluster that exceeds an existing facility’s cooling design can require facility changes before deployment. This ties AI investment decisions to data-centre operations, facilities and capital planning before equipment is installed. Polaris also identifies total electricity consumption as a central issue as computing demand rises.

Cloud growth leaves a substantial on-premises requirement

Cloud expansion exists alongside continued spending on infrastructure controlled directly by organisations. Polaris reports that on-premises systems accounted for 58.4% of the market in 2025. It associates retained on-premises infrastructure with organisations seeking more direct control over data security, regulatory compliance, customisation and latency. Those requirements can shape where workloads and data are placed.

Hybrid infrastructure combines capacity operated directly by an organisation with capacity supplied by third parties. Polaris says businesses can retain critical workloads on site while using third-party infrastructure for other applications. Technology leaders can assign workloads to environments based on operating requirements, then ensure retained capacity has the required power, cooling, connectivity and resilience.

Regional requirements add another factor to placement and investment decisions. Polaris projects Europe to grow at a compound annual rate of 13.8% over the forecast period, associating investment with data sovereignty rules, edge-computing deployments and a stronger focus on sustainable infrastructure. Asia Pacific is also expanding as governments and companies pursue digital-economy programmes, smart city initiatives and new hyperscale projects. Physical-infrastructure demand spans internal facilities, edge locations and large third-party data centres.

Expansion and efficiency need one plan

Capacity decisions affect electricity use, thermal requirements, resilience needs and workload placement. These factors can shape facility design and operating requirements before a digital service reaches production, especially when dense computing puts new demands on existing infrastructure. Including them in the original capacity decision makes the related physical investment visible earlier.

Polaris identifies renewable energy integration, modular construction, advanced monitoring systems and software tools for managing power use and cooling performance as potential responses. It also highlights digital twins, software representations of physical systems used for planning and monitoring, and smarter power distribution systems as tools that could improve operational efficiency. These technologies give infrastructure teams more options for matching facilities to workload requirements.

Geography also shapes investment. North America represented 40.6% of the market in 2025, according to Polaris, with the United States accounting for 69.2% of that regional total. Polaris attributed North America’s lead to high cloud adoption, heavy investment by hyperscale operators and rapid expansion of facilities designed for artificial intelligence workloads. Digital investment and physical-infrastructure spending can therefore concentrate in the same markets.

Efficiency changes the economics and operating profile of added capacity. A project can meet its compute target while creating significant electricity, cooling and resilience requirements. Assessing those requirements early gives executives useful information while workload placement, facility choices and capital commitments can still be changed.

Key highlights

• Plan beyond compute: AI, cloud, HPC and edge growth all depend on power, cooling, connectivity and facility capacity. Leaders should include these physical requirements in capacity and capital plans from the start.
• Consolidate infrastructure demand across workloads: Cloud, healthcare, eCommerce, telecom and smart city workloads can create shared demand for physical infrastructure. Plan this demand across business portfolios rather than treating each programme separately.
• Match power and cooling to AI density: Dense AI systems can exceed existing power and thermal limits. Assess electricity supply, resilience and cooling architecture before committing to new compute capacity.
• Treat hybrid infrastructure as a physical capacity decision: On-premises systems represented 58.4% of the market in 2025, reflecting continued requirements for control, compliance and latency. Workload placement decisions should account for the infrastructure needed at each location.
• Plan expansion and efficiency together: More compute capacity brings higher electricity, cooling and resilience requirements. Evaluate monitoring, modular infrastructure, renewable energy integration and power-management technologies alongside expansion plans.