Beyond Air: How AI Rack Densities Are Rewriting Cooling and Design in the Data Center Solutions Market
Every generation of computing has changed what a data center looks like on the inside. AI is changing it faster than any before. Graphics processors and accelerators pack far more power into each rack than conventional servers, and nearly all of that power ends up as heat. Epignosis Insights sees this thermal challenge as one of the main forces reshaping the Data Center Solutions Market, which is projected to grow at a 15.7% CAGR between 2027 and 2034.
AI Has Changed the Workload Profile
Accelerated servers are the new growth engine
The International Energy Agency (IEA) expects electricity use by accelerated servers, which are mainly associated with AI workloads, to grow by about 30% a year through 2030, compared with roughly 9% for conventional servers. In practice this means that new capacity is being designed around AI clusters first, with general-purpose computing fitted in around them.
The specification is changing, not just the volume
AI is not simply adding more servers. It is changing the mix of equipment in the room: high-density GPU systems, low-latency networking, heavier racks, larger power feeds and more advanced thermal management. Facilities designed for steady, moderate loads now face sustained high loads with sharp fluctuations.
Figure: AI Transformation Indicators
Rack Densities Are Climbing at the Top End
The shift is clearest in rack density. Uptime Institute research shows that the share of facilities with peak rack densities of at least 50 kW rose to 9% in 2025, up from 6% in 2023. At the same time, 82% of facilities still operate below 30 kW per rack. The market is therefore splitting in two: a fast-moving high-density segment built for AI, and a large installed base that remains conventional.
At the leading edge, current AI systems can exceed 40 kW per rack, and some 2025-generation implementations go beyond 100 kW. Each new accelerator generation pushes the ceiling higher.
Why Air Cooling Reaches Its Limits
Air is a poor carrier of heat at these densities. Moving enough of it through a rack requires larger fans, higher airflow and more floor space, all of which add energy cost and noise while delivering diminishing returns. The same Uptime research notes that liquid cooling is generally suited to rack powers above 50 kW, with near-total liquid cooling becoming necessary at around 150 kW or more.
Liquid Cooling Adoption: A Retrofit Wave in Waiting
Adoption is still early. Only 19% of data centers had deployed liquid cooling by early 2026, but a further 36% plan to adopt it within the next 12 to 24 months. That planned pipeline is almost double the current installed base. Efficiency gains have also stalled, with average PUE holding at 1.54 for six consecutive years, which strengthens the case for cooling technologies that remove heat closer to the chip.
Hyperscale AI facilities
Purpose-built AI campuses are designing for direct-to-chip liquid cooling from the start, with coolant distribution units, facility water loops and heat rejection sized for extreme densities. Because these sites are planned around future accelerator generations, designers are building in headroom for higher flow rates and warmer water temperatures, which also improves the potential for heat reuse and lower chiller energy.
Enterprise and colocation sites
Existing facilities face a harder choice. Most will combine legacy air-cooled halls with targeted AI zones, using rear-door heat exchangers, liquid-assisted air cooling or isolated direct-liquid pods that can be added without rebuilding the building. The deciding factors are usually floor loading, access to facility water, available electrical capacity and how much downtime the site can tolerate during installation. For many operators, a phased approach that starts with a single high-density zone is the most realistic path.
The Opportunity Map for Solution Providers
The cooling transition opens demand across several product lines within the Data Center Solutions Market: direct liquid cooling systems, coolant distribution units, rear-door heat exchangers, chillers and heat-rejection equipment, reinforced racks and enclosures, and monitoring and DCIM software to manage hybrid air-liquid environments. Services also matter, because retrofits require careful planning around plumbing, floor loading and uptime.
The Bottom Line
AI is not just expanding data center capacity; it is changing its technology mix and capital intensity. Epignosis Insights expects cooling to remain one of the most dynamic segments of the data center solutions market through 2034, with the strongest growth among vendors that make liquid cooling practical for existing facilities, not only for new AI campuses.