KAYTUS adds rack-level AI infrastructure management

KAYTUS has introduced KSManage Ultra, an AI infrastructure management platform designed to monitor and control high-density AI racks, clusters, and data-centre systems.

The platform was launched at ISC 2026 in Frankfurt and brings compute, networking, power, rack, and liquid-cooling management into a single operational environment. It is designed for deployments where GPUs, CPUs, memory, network hardware, power distribution units, cooling distribution units, racks, and clusters must be managed as one integrated system.

KSManage Ultra combines in-band and out-of-band management, correlating operating system and application data with hardware health, power consumption, temperature, and cooling information. It can monitor compute resources, evaluate GPU and memory status, check PCIe links, assess firmware consistency, monitor cooling equipment, and identify faulty or high-risk nodes before they are assigned to AI training or inference workloads.

The platform also monitors liquid-cooling infrastructure, with functions for leak detection, node shutdown, equipment isolation, and operator alerts. Deployment functions include automated rack discovery, topology mapping, batch scanning, stress testing, driver installation, hardware configuration, and software deployment workflows.

AI infrastructure has turned data centres into dense electromechanical systems. Processor performance and accelerator availability remain highly visible, but the ability to use that compute consistently depends on power delivery, coolant flow, firmware alignment, switch health, interconnect stability, and workload placement. A fault in one layer can appear as instability somewhere else in the rack.

Power-density work at component and package level is already compressing the electrical and thermal problem around the processor. Lotus Microsystems’ vStrata vertical power delivery platform moves conversion closer to high-current AI processors, shortening current paths while integrating electrical and thermal design. KSManage Ultra addresses the rack-level operating layer that sits above those component choices, where power, cooling, firmware, and compute status have to be coordinated continuously.

Management tools for conventional servers were not designed for AI clusters containing tightly coupled accelerators, high-speed networking, and liquid-cooled racks operating near infrastructure limits. The cost of a failed or underperforming node rises when workloads span multiple GPUs and long training runs. Preventing poor workload placement can therefore be as valuable as detecting a hardware fault after it has disrupted operation.

Liquid cooling adds another set of control requirements. It enables higher rack density but brings mechanical and maintenance risks that need direct integration into the management stack. Leak detection, flow monitoring, safe shutdown logic, and isolation procedures become part of system reliability rather than facilities maintenance alone.

Firmware consistency is another pressure point. AI infrastructure contains large numbers of similar but not always identical components, with performance and stability dependent on driver versions, firmware levels, BIOS settings, accelerator health, and network configuration. Automated discovery and configuration reduce the risk that human process becomes the weak link in cluster deployment.

KSManage Ultra reflects the movement from server management towards infrastructure orchestration. Dense AI systems cannot be managed as isolated boxes when a rack behaves as a single compute resource, a power load, a thermal zone, and a network fabric. Semiconductor performance will continue to draw attention, but usable AI capacity increasingly depends on the systems that keep accelerators powered, cooled, configured, and scheduled correctly.