Daikin And NTT DATA Test AI Data Center Cooling
AI infrastructure teams are starting to optimize not only models and chips, but the cooling systems that make dense AI servers viable. Daikin and NTT DATA Japan said they will begin a July 2026 proof of concept for a data center cooling optimization system that uses AI to predict internal server thermal conditions from indirect data such as power consumption and temperature. The companies plan to coordinate HVAC, chiller, and liquid-cooling equipment from those predictions, validate the system at an NTT DATA data center in Japan during fiscal 2026, and target commercialization in fiscal 2027. For operators, the news points to a practical response to AI server heat loads and workload swings: treat thermal control as a data-driven, integrated system rather than a set of fixed cooling rules.
Why it matters
The AI infrastructure bottleneck is no longer only GPUs, networking, or power procurement. Dense AI servers create fast-changing heat loads, and cooling systems that were tuned for steadier enterprise workloads can waste energy or react too slowly. Daikin and NTT DATA's proof of concept is a useful signal because it applies AI to the physical operations layer that determines whether high-density AI capacity can run efficiently.
What changed
Daikin and NTT DATA Japan said they will begin a July 2026 proof of concept for a next-generation data center cooling optimization solution. The system is designed to predict internal server thermal conditions using indirect inputs such as server power consumption and temperature information, including in environments where operators cannot directly access detailed internal server data.
The companies plan to use those predictions to coordinate HVAC, chiller, and liquid-cooling equipment. They will validate the system at an NTT DATA data center in Japan during fiscal 2026 and aim for commercialization in fiscal 2027. Refrigeration Industry and Data Center Dynamics both reported the same project context around AI-driven cooling control for data centers.
Operator read
For data teams and infrastructure buyers, the important part is the control loop. If thermal prediction becomes reliable without direct chip-level telemetry, cooling vendors can help colocation and enterprise facilities manage AI racks with less intrusive integration. That could reduce overcooling, improve response to bursty inference or training workloads, and make liquid cooling easier to operate alongside existing HVAC and chiller systems.
This is still a proof of concept, not a deployed industry standard. The practical question is whether the model can generalize across different rack layouts, server generations, and safety constraints. But the direction is clear: AI data centers are becoming cyber-physical systems where software, telemetry, and cooling controls need to be designed together.
Key Points
- 1Daikin and NTT DATA will test AI-based thermal prediction for data center cooling in July 2026.
- 2The system uses indirect server power and temperature data to coordinate HVAC, chillers, and liquid cooling.
- 3If validated, the approach could reduce overcooling and make dense AI server operations more responsive.
Scoring Rationale
This is a solid infrastructure development because AI-server cooling is becoming a material operating constraint for data centers. The project is still a proof of concept, so the impact is below major deployment or hyperscale capacity announcements.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
