We are starting to trust AI with high stake tasks, especially in execution of automated factories and by guiding military drones via a hostile airspace. But when it comes to managing the data centers that feed this AI revolution, human operators are much more cautious.
According to a new survey of more than 600 operators of data centers around the world by Uptime Institute, an inspection and notation company of data center, only 14% say that they would trust the AI systems to change the equipment configurations, even if they are trained on years of historical data. In the same survey, only one in three operator claims that they would trust the AI systems to control data centers equipment.
Their skepticism can be justified: despite the dozens of billions of US dollars in AI systems, 95% of organizations have so far has no clear return on investment, according to a recent MIT report of generative-AI use. Advanced industries, which include factories and data centers, classified near the list of sectors processed by AI, if applicable.
Operator’s confidence in AI systems
Even before the AI-centered push to extend the data centers, the operators of data centers themselves are known to be a crowd relatively opposed to the change which was disappointed by the technologies of the past, explains Rose Weinschenk, research associate at the Uptime Institute. Operators often have electrical engineering or technical mechanical history, with training in the management of critical installations; Others work on the computer side or the network system and are also considered operators.
Operator Trust in Ai Declined Every Year for the Three Years Following OpenAi’s Release of Chatgpt in 2022. When asked by Uptime if they trusted a trained a ai system to run data-center operations, 24 piernt of responds Said no in 2022, and 42 Percent Said No In 2024. While the public has marveled at the Seemingly All-Knowning Nature of New Large Language Models, Operators Seem to Feel This Type of Ai is Too Limited and Unpredictable for Use in Centers of Data.
But now operators seem to have entered a “meticulous test and validation period” of different types of AI systems in certain data center operations, said Max Smolaks, search for search for availability in a public webinar of the latest survey results. To capture the evolution of feelings, availability asked operators in 2025 which AI requests could serve as a trusted decision -maker, assuming adequate previous training. More than 70% of operators say they would trust AI to analyze the sensor data or predict equipment maintenance tasks, according to the survey.
“Data centers operators are very, very happy to do certain things using AI, and they will never trust AI to do certain other things,” Smolaks said in the webinar.
The unpredictability of AI in data centers
One of the reasons why confidence in AI is low for critical equipment control is the unpredictability of technology. The data centers are executed on “good and old -fashioned” engineering, as programmed if / then logical, explains Robert Wright, the director of data centers of the Ilkari data centers, a data center startup with two centers in Colombia and Iceland. “We say that we cannot run luck, we must run on certainty.”
Data centers are a complex series of systems that feed on each other. Only seconds can pass before catastrophic failures occur which cause damaged tokens, waste -waved money, angry customers or deadly fires. In the environment with high issues of data centers, anonymous posters on the R / Datacenter Reddit forum which responded to a Spectrum ieee The request has generally not seen a reason to justify the risk that AI could bring.
Mistrust can also hide an underlying insecurity. Workers of many industries fear that AI will take their jobs. But the 2025 availability survey revealed that only one in five operator considers AI as a means of reducing average personnel levels.
“Operators believe that today AI will not replace the staff required to manage their facilities,” said SMOLAKS in the help webinar. “This could come for office employees, but data center work seems to be safe from AI for the moment.”
But it is understandable that career operators feel As this technology arrives for their work, said electrician engineer Jackson Fahrney, who has been working in data centers for over eight years. Someone only six months at work can see an AI system as if they had been said: “Here, train your replacement,” he said. In reality, he does not think that AI will replace it or others in data centers. However, the AI offers a more “disturbing” presence in the workplace that automatic learning tools, which has long been part of an operator’s toolbox and aim to help operators when making decisions.
AI could be the Cherry in addition to a tendency on the scale of the industry to reduce the number of operators in data centers, explains Chris McLean, consultant in the design and construction of data centers.
While 60 engineers may have executed a data center in the past, now only six are necessary, says McLean. These six is also necessary, as more and more critical maintenance is outsourced to specialists outside the data center. “Now you compensate all your risks with a low -cost human and high cost AI,” said McLean. “And I have to imagine that it scares for operators.”
That said, there are more data center jobs than qualified candidates, as indicated above by Spectrum. Two -thirds of operators find it difficult to retention or recruit staff, according to the increase in the increase in 2025, similar to the responses of surveys for the previous two years.
Effective AI algorithms for data centers
However, there are useful algorithms built on decades of research on automatic learning that could make the functioning of the data center more efficient. The most established AI system for data centers is predictive maintenance, explains Wright from Ilkari. If the readings of a particular HVAC unit increase faster than that of other units, for example, the system can predict when this unit must be maintained.
Other AI systems focus on optimizing cooling plants, which are, in fact, refrigerator systems that keep the data center cool by circulating water and refrigerated air. The coolers represent a large part of the energy consumed by the data centers. The data on weather conditions, the load on the grid and the degradation of the equipment over time all fuel a single AI system executed on the equipment in the installation to optimize total energy consumption, explains Michael Berger, who directs research and development in the Software of Energy -based Software.
But Berger is quick to note that its AI optimization software does not control the equipment. It extends over the basic control loop and refines the parameters to use less energy while reaching the same result, he said. Berger prefers to call this automatic learning system instead of AI because of its specialization for the needs of a data center.
Others fully adopt AI, both name and technology, such as Joe Minarik, the DataBank Operation Head, a Dallas-based data center company with 73 data centers across the United States and the United Kingdom. He attributes his certainly optimistic attitude towards AI in his 17 years of work for Amazon Web Services, where the software is King. Currently, DataBank uses AI to write software, and it is planned to deploy AI systems for automated generation and monitoring of tickets, as well as monitoring and adjustments of the configuration network by the end of the year. The AI for greater tasks, such as cooling, is temporarily planned for the end of 2026, subject to the time necessary to form AI on enough data, he said.
AI makes hallucines: Minarik watched him give bad information and send his team on the wrong path. “We do it, we see him happening today. But we also see him becoming better and better once we have given him more time, ”he says.
The key is “huge amounts of data points” so that AI understands the system, says Minarik. It is no different from the training of a human data center engineer on each possible scenario that could occur in the corridors of a data center.
Hyperscalers and corporate data centers, the single customer of which is the owner of the data center, deploy AI at a faster rate than commercial companies like DataBank. Minarik hears AI systems that run whole networks for internal data centers.
When the database deploys AI for larger operations of data centers, it will be kept in a tight leash, says Minarik. The operators will always perform final executions.
While AI will undoubtedly change the way the data centers work, Minarik considers operators as an essential part of this new future. Data centers are physical places with an activity on site. “The AI cannot go out and change a spark plug,” he says, or hear a strange rattle in a server rack. Although Minarik says that one day there could be sensors for some of these problems, they will always need physical human technicians to repair the equipment that keeps the data centers in progress.
“If you want a safe job that can protect you from AI,” says Minarik, “go to data centers”.
From your site items
Related items on the web
