The 2026 State of Liquid Cooling Report landed with data that confirms what colocation operators have been hearing from their largest tenants: liquid cooling is no longer a niche capability reserved for hyperscale facilities. It is becoming a baseline expectation for any operator that wants to compete for AI workloads. The report aggregates deployment data from hundreds of facilities across North America and Europe, and five findings stand out as particularly significant for operators making infrastructure investment decisions right now.
Here is what the numbers tell us and what operators should do about it.
1. AI Workloads Now Represent Over 40% of New Colocation Lease Demand
This number has been climbing for three years, but 2026 marks the first year AI workloads surpassed traditional enterprise and cloud workloads in new lease signings at major colocation providers. The report shows that AI-related leases grew 62% year-over-year, driven by model training, fine-tuning, and inference at scale.
For operators, this shift is both an opportunity and a warning. The opportunity is straightforward: AI tenants pay premium rates for high-density capacity. The warning is that these tenants have specific infrastructure requirements that general-purpose facilities cannot meet. Operators without liquid cooling capability are being excluded from RFPs before they even submit a proposal.
The practical implication is that capital investment in liquid cooling infrastructure is no longer speculative. It is the cost of remaining competitive in a market where the fastest-growing tenant segment requires it.
2. The Average Rack Density for AI Deployments Has Crossed 50kW
Two years ago, 30kW per rack was considered aggressive. The 2026 report shows the average rack density for AI-specific deployments has reached 50kW, with leading-edge deployments pushing past 100kW. This is a direct result of GPU manufacturers shipping higher-power processors and AI companies deploying denser configurations to reduce networking costs and improve training performance.
Operators designing new capacity or retrofitting existing space need to plan for these numbers today. Building cooling infrastructure rated for 30kW per rack will be outdated before construction is complete. The report recommends designing for a minimum of 60-80kW per rack for new AI-focused deployments, with the ability to scale beyond 100kW as next-generation GPUs enter production.
This has significant implications for mechanical, electrical, and plumbing infrastructure. Power distribution, cooling loops, structural floor loading, and cable management all need to be designed for densities that were considered extreme just 18 months ago.
3. Liquid Cooling Adoption Reached 35% of New Deployments in 2025, Projected to Hit 55% by End of 2026
The adoption curve for liquid cooling has accelerated faster than most industry projections from even two years ago. The report attributes this to three factors: GPU power density increases that make air cooling impractical, tenant demand for liquid-cooled infrastructure, and improved availability of coolant distribution units from multiple vendors.
For operators who have not yet invested in liquid cooling, this data point carries urgency. Moving from 35% to 55% adoption within a single year means the market is approaching a tipping point where liquid cooling becomes the default rather than the exception. Operators that delay investment risk finding themselves on the wrong side of that shift.
The report also notes that operators who deployed liquid cooling early are seeing 15-25% higher revenue per square foot compared to air-cooled deployments of similar scale. The revenue premium reflects the higher value that AI tenants place on cooling performance and the willingness to pay for infrastructure that supports their compute requirements.
4. Retrofit Projects Are Taking 30% Longer Than Planned on Average
This finding should concern every operator with retrofit projects in their capital plan. The report surveyed 87 retrofit projects completed in 2025 and found that the average project ran 30% longer than the original timeline. The primary causes were supply chain delays for specialized cooling components, unexpected structural and plumbing challenges in existing facilities, and underestimated engineering complexity.
The lesson for operators is to build realistic timelines with meaningful contingency buffers. A retrofit project planned for six months should be budgeted for eight. More importantly, operators should start engineering assessments now rather than waiting for a signed tenant lease to trigger the work. The facilities that can offer liquid-cooled capacity on shorter timelines will have a significant competitive advantage.
The report also found that operators who engaged MEP engineering firms with specific liquid cooling experience completed projects 20% faster than those who relied on general data center engineering teams. Specialized expertise matters when the stakes and timelines are this tight.
5. Water Usage Efficiency Is Becoming a Tenant Selection Criterion
The final standout finding is that 48% of AI tenants surveyed now include water usage efficiency in their facility evaluation criteria. This represents a significant shift from two years ago, when power and cooling capacity were the dominant evaluation metrics.
AI companies face increasing pressure from investors, regulators, and public stakeholders to demonstrate environmental responsibility. Liquid cooling systems that minimize water consumption or use closed-loop designs that eliminate evaporative water loss are increasingly preferred over traditional cooling tower approaches.
For operators, this means cooling technology selection has implications beyond performance and cost. The sustainability profile of your cooling infrastructure is becoming a differentiator in tenant negotiations. Operators that can demonstrate low water usage alongside high cooling performance are positioning themselves for the next wave of AI tenant demand.
What Operators Should Do With This Data
The 2026 report paints a picture of a market that is moving faster than most operators planned for. The key takeaway is that liquid cooling investment is not a future consideration. It is a current competitive requirement. Operators who act on these findings now, by accelerating cooling infrastructure projects, engaging specialized engineering teams, and designing for tomorrow’s density requirements rather than today’s, will capture the AI tenant revenue opportunity. Those who wait risk permanent market share loss to competitors who moved first.
Nautilus Data Technologies partners with colocation operators, AI Companies and hyperscalers to plan and deploy liquid cooling infrastructure that meets current and future AI workload requirements. Contact our team to discuss how the 2026 report findings apply to your facility roadmap.
