.png)
2 days ago
2
Being able to forecast future capacity requirements is a growing concern for datacenter operators as they face conflicting factors such as rising costs, power constraints, and meeting the demands of AI workloads.
A report from Uptime Institute on the results of its 15th Annual Global Data Center Survey shows that the industry is beset by challenges, although some might say that having to expand to meet increased demand is a good problem to have.
Outages are gradually becoming less frequent, but that has to be set against increasing regulation and efficiency requirements, while operators also face staffing challenges and supply chain delays.
Cost issues remained the top concern for the next 12 months among respondents to the survey, while forecasting future capacity requirements is now the next biggest worry as more datacenter owners and operators plan to perform some AI training or inference workloads in the future.
There's a lot of uncertainty facing operators in 2025 as they balance increasing costs due to factors such as energy prices and market volatility with increasingly complex capacity planning, the report states.
Uptime also throws up some interesting findings in light of the relentless growth of public cloud, and reports of sites fitting racks for increasingly heavy electrical loads to cope with AI servers crammed with hot and hungry GPUs.
According to the report, 45 percent of IT workloads are still operating on-premises in corporate datacenters, with another 16 percent in colocation. Only 11 percent are in public cloud infrastructure, although another 10 percent are listed as hosted private cloud, and six percent is made up of software-as-a-service.
Meanwhile, more than 80 percent of respondents indicated that their highest server rack power density was below 30 kW, despite increasing power loads. Although some organizations said that they had cabinets exceeding 100 kW, only nine percent had racks greater than 50 kW per rack.
A small number of high-density facilities distort the average figures, with the average density excluding these outliers coming to 7.5 kW, up from 6.8 kW in 2024, with the trend being upwards.
This puts recent reports like Google planning for rack infrastructure supporting 1 MW of IT hardware loads into perspective.
In terms of outages, Uptime reports that serious ("impactful") outages continue to decline relative to the growth in IT, despite the rising number of media headlines covering significant datacenter failures. On a per-site basis, the frequency of impactful outages is decreasing, but the pace of improvement is slowing.
About Half of operators report no outages in the past three years, while only three percent experienced the most severe kind of incident on Uptime's one to five scale. That still means that 50 percent of bit barns had at least one outage at their facility over the past three years, but this is down from the 53 percent figure in last year's report.
- Slow down on building power plants for all those new AI datacenters, report warns
- Hyperscalers to eat 61% of global datacenter capacity by decade's end
- Dot com era crash on the cards for AI datacenter spending? It's a 'risk'
- What happens when we can't just build bigger AI datacenters anymore?
Staffing is an evergreen issue in the datacenter world, and this remains the same with 46 percent of respondents saying that they are having difficulty finding qualified candidates for vacant roles, while 37 percent say that they are struggling to retain their staff.
However, a new development is that more senior roles, such as operations management, are hard to fill, while junior and mid-level operations roles topped the list previously.
The staff shortages may be due to not only the growing number of new facilities, Uptime says, but also a loss of experienced senior staff through retirement. More worryingly, the report says that a management shortage means orgs are failing to pass on knowledge to more junior staff as senior workers leave the workplace.
Meanwhile, when it comes to using AI, operators are cautious in their adoption, and are mostly interested in tools that can increase facility efficiency or help to reduce human error.
Uptime found that 73 percent of respondents say they would trust an adequately trained AI model to run automated analytics on sensor data, but just 14 percent would allow such a system to make configuration changes. ®
