[Edaily Reporter Kwon Oh Seok ] E8IGHT Co., Ltd(418620)(E8), a company operating a physical AI platform, announced on the 30th that it has implemented a feature to predict data center cooling environments in its physical prediction solution, “NFLOW Ai.” The system calculates the internal temperature and airflow of server rooms in advance, and the AI then identifies and proposes operating settings that minimize cooling power consumption while remaining within the equipment’s tolerance limits.

After servers, the second-largest contributor to data center power consumption is the cooling equipment used to cool them. A key metric for cooling efficiency is PUE (Power Usage Effectiveness), which is calculated by dividing the data center’s total power consumption by the power consumption of IT equipment, such as servers. A value closer to 1 indicates that less power is being used by auxiliary equipment other than IT equipment.
According to the Uptime Institute’s 2025 Global Data Center Survey, the average PUE among responding companies was 1.54, remaining largely unchanged for six consecutive years. After falling rapidly from approximately 2.5 in 2007 to 1.65 in 2014, the figure has plateaued in the 1.5 range since the 2020s. Some analysts interpret this to mean that the efficiency gains from equipment upgrades have already been largely factored in.
Another limitation pointed out is that PUE represents an average value for the entire facility. Even if some zones are operated at excessively low temperatures while others are operated at temperatures close to the upper limit of the allowable range, no issues are apparent when looking solely at the average value. Since it is a summary of past operating results, it also fails to provide a basis for determining which settings need to be adjusted at the present time.
As a result, facilities often prioritize safety by setting cooling margins on the conservative side. Although power consumption would decrease even with a slight increase in cooling temperature, the lack of clarity on the safe upper limit leads to a structure where “just-in-case” excess power consumption accumulates as costs. In environments where the risk of service interruption is high, the process of directly adjusting and verifying cooling settings also acts as a burden.
According to E8, NFLOW Ai predicts temperature distribution and airflow when users input the server room layout and operating conditions. While similar calculations were possible in the past, they took a long time to process even a single set of conditions, making it difficult to compare multiple scenarios. NFLOW Ai has significantly reduced calculation time by having AI pre-learn the results of calculations under various conditions.
With faster calculation speeds, it has become possible to compare different combinations of conditions. There are so many possible combinations of cooling temperature settings and HVAC operating modes that it is difficult for Saramins to review them one by one. NFLOW Ai automatically explores all possible combinations using AI and proposes operating settings that minimize cooling power consumption while maintaining the temperature range that the equipment can withstand.
The solution does not directly control the equipment but serves solely to predict and make recommendations; the decision to actually change settings rests with the person in charge. The prediction results also include a confidence metric, indicating a high degree of uncertainty for conditions outside the trained range.
An E8 official stated, “Data centers are an area where the thermal flow analysis our company has been conducting so far extends relatively naturally,” adding, “We are observing how this approach, which provides quantitative justification for cooling decisions, is being utilized in the field.”