Datacenter operators face an increasing challenge in managing their resources due to the expansion of AI technology.
The Uptime Institute's 15th Annual Global Data Center Survey reveals significant trends in the datacenter industry, particularly in the context of AI workloads.
As operators grapple with increasing costs and complex capacity planning in 2025, a surprising finding emerges: 73% of respondents trust adequately trained AI models to run automated analytics on sensor data, yet only 14% would allow such a system to make configuration changes. This cautious approach to AI adoption is reflected in the industry's primary interest in tools that increase facility efficiency or reduce human error.
The report also sheds light on the ongoing staffing challenges. The growing number of new facilities and the retirement of experienced senior staff is creating a knowledge gap, leading to staff shortages. Senior roles in operations management are becoming increasingly difficult to fill, while junior and mid-level operations roles were previously more challenging. The crisis continues, with 46% of respondents struggling to find qualified candidates for vacant roles and 37% having trouble retaining staff.
Current trends in capacity planning for AI workloads include rapidly increasing power demand, higher rack density, complex cooling needs, and the need for sophisticated forecasting and resource management tools. AI workloads—particularly training—are expected to more than triple between 2025 and 2030, driving unprecedented growth in data center scale and power consumption.
Key challenges include power and cooling demands, cost and infrastructure constraints, capacity forecasting uncertainty, supply chain and talent shortages, and balancing hybrid workloads. AI servers with GPUs generate intense heat and require new cooling methods such as liquid cooling. Rack densities now routinely exceed 30–50 kW, far surpassing traditional data centers.
In response, AI-powered capacity planning tools like Epicflow help by forecasting future workloads, visualizing bottlenecks, running simulations, and optimizing project rescheduling to maximize resource utilization.
Regarding on-premises versus public cloud infrastructure usage, 45% of IT workloads still run on-premises in corporate datacenters, with another 16% in colocation, and only about 11% in public cloud infrastructure as of 2025. Hyperscale cloud providers currently control 44% of global data center capacity, projected to rise to 61% by 2030, largely driven by AI and cloud growth. On-premises and colocation centers grow more slowly and often serve as capacity offload or temporary solutions while hyperscalers build large AI-optimized facilities nearby.
In summary, the scale and complexity of AI workloads are reshaping data center capacity planning by driving the need for new design approaches, advanced capacity forecasting, and hybrid infrastructure strategies that incorporate both public cloud and on-premises resources. The public cloud will grow substantially, but substantial on-premises capacity remains crucial for many organizations in managing AI workloads.
Forecasting future capacity requirements is the next biggest worry for datacenter operators, as more of them plan to perform AI workloads. Despite increasing power loads, only nine percent had racks greater than 50 kW per rack. The industry continues to grapple with these challenges, seeking solutions that balance efficiency, cost, and the ever-growing demand for AI capabilities.
- AI-powered analytics, running on adequately trained models, are increasingly trusted to analyze sensor data, with 73% of respondents endorsing this practice.
- unexpectedly, only 14% of respondents are willing to allow such AI systems to make configuration changes.
- Amidst these insights, the survey highlights the ongoing staffing challenges in the datacenter industry, with 46% of respondents facing trouble finding qualified candidates for vacant roles and 37% struggling to retain staff.
- For capacity planning of AI workloads, the report emphasizes the necessity of AI-optimized tools, such as Epicflow, which forecast future workloads, visualize bottlenecks, run simulations, andoptimize project rescheduling to maximize resource utilization.
