ISC 2026: KAYTUS Launches Rack-Scale KSManage Ultra for AI Factories

FRANKFURT, Germany--(BUSINESS WIRE)--Jun 25, 2026--

KAYTUS, a leading provider in AI infrastructure and liquid cooling solutions, has launched KSManage Ultra at ISC 2026, a next-generation intelligent infrastructure management platform purpose-built for AI Factories. Designed for the latest high-density AI racks, KSManage Ultra enables unified, intelligent management of key rack-level components, including compute trays, switch trays, power distribution units (PDUs), and cooling distribution units (CDUs). Through end-to-end visibility, performance-level diagnostics, and automated operations, the platform transforms highly coupled AI infrastructure from fragmented oversight into integrated, system-level operations, helping enterprises build more efficient, reliable, and sustainable AI infrastructure.

This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260624682202/en/

Three Key Challenges Facing Traditional AI Operations

Compared with traditional data centers, AI data centers face significantly greater operational complexity, including rack-scale AI system management, intricate network topologies, challenging fault isolation, and liquid-cooling safety requirements. As a result, traditional operations approaches are increasingly constrained by three key challenges:

First, Management complexity is soaring: the basic unit of an AI Factory is no longer an individual server, but a resource-coupled, high-density AI rack. A single rack integrates multiple deeply coordinated subsystems, including computing, networking, power supply, and liquid cooling. Compared with traditional 4U 8-GPU deployments, NVL72 rack-scale systems integrate nearly one hundred accelerators and thousands of high-speed interconnects. In a 100 kW-class rack, power density can be 2–3 times higher¹, while thermal management becomes significantly more complex, involving coolant distribution, CDUs, flow rates, and related safety controls. As AI Factories continue to scale, operational complexity rises sharply, and fluctuations in any single component can affect the performance and stability of the entire rack.

Second, fault identification has moved beyond the hardware layer. AI training and inference workloads are highly sensitive to performance fluctuations, and hidden anomalies can significantly reduce operational efficiency. Unlike traditional downtime-related failures, performance degradation in AI systems often occurs silently. Because these performance issues are closely linked to underlying hardware and infrastructure conditions, identifying the true root cause can be difficult when relying on isolated data from either the workload or infrastructure side alone.

Third, AI Factories face a growing operational efficiency crisis as deployments scale. Traditional device-by-device onboarding is inefficient, slowing deployment and increasing the risk of configuration inconsistencies. At the same time, conventional configuration methods are time-consuming and error prone. With multiple device types integrated within each AI rack, even minor configuration deviations can lead to cluster-wide performance degradation or service interruptions.

KAYTUS Builds an Integrated Intelligent Operations Platform for AI Factories

Against this backdrop, traditional operations models that depend on manual processes or fragmented tools often result in delayed deployment, challenging troubleshooting, and inconsistent configurations, limiting the development and large-scale adoption of AI applications. To help simplify the operation and management of AI data centers, KAYTUS has introduced KSManage Ultra. The platform delivers integrated management across the full infrastructure stack, spanning components, nodes, racks, clusters, and the data center level, by connecting in band and out of band management paths and correlating IT infrastructure status with physical infrastructure conditions. It represents the shift from reactive operations to proactive alerting, helping customers build intelligent operations capabilities for monitoring, diagnosis, fault isolation, and full recovery in complex AI Factory environments.

Single-Pane Global Visibility into AI Data Center Operational Status

KSManage Ultra delivers full-stack unified management across both traditional infrastructure and advanced AI rack systems. The platform provides centralized management for GPUs, CPUs, memory, high-speed switching modules, management networks, power shelves, CDUs, liquid cooling systems, racks, and cluster resources. By breaking down management boundaries between IT and physical infrastructure, as well as between individual components and full racks, KSManage Ultra creates a multi-level resource view spanning components, nodes, racks, clusters, and the entire data center.

Through a unified platform, customers can avoid repeatedly switching between multiple systems and quickly assess resource health, rack availability, and cluster readiness for efficient production deployment and operation.

Integrated In-Band and Out-of-Band System Management for Proactive Remediation

KSManage Ultra consolidates in-band data, including operating systems, drivers, applications, and performance, with out-of-band data, such as BMC, firmware, power, temperature, and hardware logs, together with infrastructure data into a single unified management system. It enables correlation analysis across operating status, hardware health, link topology, power supply, and liquid cooling conditions, shifting operations from reactive response to proactive alerting. When the system detects GPU anomalies, degraded link quality, liquid cooling fluctuations, or declining node health, it can proactively identify at-risk nodes and guide customers to isolate, maintain, or reconfigure resources, helping prevent faulty nodes from entering critical task runs.

Using liquid cooling monitoring as an example, KSManage Ultra supports three-level leak detection at the node, rack, and loop levels. Once a leakage risk is detected, the platform can coordinate safety shutdown, solenoid valve closure, and node isolation, while also triggering email alerts, work order generation, and closed-loop remediation. This helps customers build system-level proactive operations capabilities for AI rack systems.

Real-Time Agile System Health Monitoring and Compute Power Resource Allocation

Designed for multi-rack deployment scenarios, KSManage Ultra provides resource health identification and fault isolation capabilities. The platform continuously evaluates node and rack health based on indicators such as GPU status, memory and PCIe status, network link quality, firmware consistency, liquid cooling conditions, and power supply status. When abnormal nodes or high-risk components are detected, the system can apply intelligent tagging, analyze the potential impact scope, and initiate isolation actions, helping prevent faulty nodes from entering critical task runs.

KSManage Ultra helps customers establish a clear view of available resources, including which nodes should be removed from service, which racks remain suitable for combined use, which resources are ready for training and inference workloads, and which resources should enter in maintenance procedures. As a result, customers can move beyond reactive repairs after failures occur, and continuously maintain a stable compute health-zone, improving AI Factory business continuity and resource utilization.

Minute-Level Onboarding, Configuration, and Full-Stack Automated Operations

KSManage Ultra supports one-click batch scanning and automatic node addition. By intelligently identifying device serial numbers and IP addresses, the platform automatically builds topology mappings between nodes and racks, reducing single-rack onboarding time from the traditional 50 minutes to less than 3 minutes. KSManage Ultra supports one-click batch stress testing at L10 and L11 levels, reducing fault root-cause localization from hours to minutes. The platform also enables rack-level automated initialization and configuration, including driver installation, hardware configuration, and software deployment, all of which can be delivered in batches based on templates. By significantly improving operational efficiency while helping maintain consistent hardware environments across the same cluster, KSManage Ultra effectively reduces the risk of performance fluctuations or task failures caused by configuration drift.

As a comprehensive unified platform for AI Factories, KSManage Ultra features open and highly compatible architecture. Through open APIs, it seamlessly integrates with upper-layer systems such as scheduling platforms and CMDBs, while also providing unified management of lower-layer heterogeneous devices, including servers, networking equipment, power infrastructure, and cooling systems. This enables centralized management across the entire data center environment. KSManage Ultra is designed to help enterprises achieve unified management and intelligent operations for heterogeneous infrastructure, providing a solid foundation for stable and efficient operation of AI Factories.

Source：

1. Traditional HGX H100/H200 4U 8-GPU servers typically support 4 to 8 units per 42U rack, resulting in rack-level power consumption of approximately 40 to 80 kW. In contrast, GB200 NVL72 racks can exceed 120 kW, driving a roughly 2x to 3x increase in power density.

About KAYTUS

KAYTUS is a leading provider of AI infrastructure and liquid cooling solutions, delivering a diverse range of innovative, open, and eco-friendly products for cloud, AI, edge computing, and other emerging applications. With a customer-centric approach, KAYTUS is agile and responsive to user needs through its adaptable business model. Discover more at KAYTUS.com and follow us on LinkedIn and X

KSManage Ultra for AI Factories

CARACAS, Venezuela (AP) — Powerful back-to-back earthquakes that struck Venezuela on Wednesday evening killed at least 32 people and injured at least 700, the nation's acting president said, as communities across the South American country sustained damage.

Acting President Delcy Rodríguez warned the toll was expected to rise as rescuers search collapsed buildings and emergency crews reach devastated areas after the 7.2 and 7.5 magnitude earthquakes that struck shortly after 6 p.m.

Rodríguez declared a state of emergency in an address to the nation late Wednesday and said the quakes caused damage in several states. The casualty figures released early Thursday excluded the state of La Guaira, which Rodríguez described as a “disaster zone” and the area hardest hit.

“Dozens of buildings have collapsed there, about 30 kilometers (19 miles) north of Caracas, and we are currently carrying out intensive rescue operations to save lives,” she said.

The earthquakes, among the strongest to strike Venezuela in more than a century, roiled the region, with buildings evacuated in cities and areas impacted as far as Brazil’s Amazon about 1,700 kilometers (1,050 miles) from Venezuela's capital Caracas.

Venezuela’s state-run VTV showed footage early Thursday of three children, covered in dust but alive, being pulled from the rubble in hard-hit La Guaira. The broadcaster also said a hospital in the city of Tucacas, about 200 kilometers (120 miles) northwest of Caracas, suffered damage and showed images of dozens of people in what seemed like medical garb in front of the building.

The earthquakes damaged and closed Simón Bolívar International Airport near Caracas, the country’s main airport, Rodríguez said, adding that subway and natural gas services in Caracas were canceled. She urged Venezuelans to report any damages through a government app.

Rodríguez said school classes would be canceled for several days. The Ministry of Education said some school buildings would be used as shelters and donation centers.

“We urge our population to remain calm,” said Rodríguez, who asked health care professionals to report to hospitals to assist the injured. “We urge unity.”

The U.S. Geological Survey initially said the first earthquake had a magnitude of 7.1, later revising that to 7.2. Its epicenter was west of Morón on the country’s Caribbean coast about 168 kilometers (104 miles) west of Caracas. The quake had a depth of 22 kilometers (13.6 miles).

The USGS reported an even larger 7.5-magnitude earthquake just a minute later. The second quake had a depth of 10 kilometers (6.2 miles) with an epicenter 16 kilometers (10 miles) southwest of Morón.

In the coastal state of Falcon, Gov. Víctor Clark said 32 people had been hospitalized and there were 15 people trapped in the hours after the earthquake.

Offers of help were made by various governments including the United States, Argentina, Bolivia, Brazil, Chile, Ecuador, El Salvador, Mexico, Panama and Uruguay.

U.S. Secretary of State Rubio said in a post on X early Thursday that the United States is “immediately deploying search and rescue teams, medical resources, and humanitarian assistance to Venezuela.”

Jeremy P. Lewin, the U.S. undersecretary of state for foreign assistance, said the State Department had mobilized a disaster assistance team and task force to coordinate aid in coordination with the interim Venezuelan government.

Rodríguez said Thursday that Qatar had already sent rescuers who were expected in Venezuela the next day, along with rescue personnel from Mexico and El Salvador.

El Salvador President Nayib Bukele, once diametrically opposed to Venezuela’s government, said in a post on X Wednesday night that he had offered aid.

“We send you all our solidarity and our prayers. Stay strong, Venezuela,” Bukele wrote.

Ecuadorian President Daniel Noboa said he had ordered the immediate delivery of humanitarian aid to help respond to the emergency.

“Ecuador will respond with the speed and commitment this moment demands because, despite our enormous differences, humanity must always guide the actions of a leader,” Noboa wrote.

Bolivian President Rodrigo Paz, who less than a week ago declared a state of emergency in his country following weeks of anti-government protests, said his country stood ready to provide any needed assistance.

The administration of Brazil President Luiz Inácio da Silva expressed solidarity and said no Brazilians reported being injured.

Television broadcasts Thursday showed images of rescue workers using power tools on collapsed structures.

During the quakes, people evacuated swaying buildings in Caracas, many visibly shocked as they saw collapsed walls that left furniture visible from the street. Dust columns could be seen in two capital neighborhoods with typically busy restaurants and other businesses.

People remained on the streets for hours, some sitting on the ground hugging pets as dust gathered around them. Collapsed buildings, toppled electric poles and debris blocked streets. Parts of the capital lost power and cellphone signal.

“It started off gently and then gradually grew, and in the end, we all had to leave our houses, go outside and gather together,” Caracas resident Hector Ricci said.

Roberto Gamas, another Caracas resident, said the building he was in “really shook from side to side. Unreal. The force was incredibly strong.”

The lack of cellphone signal in parts of Venezuela deepened the distress of many families, particularly those among the more than 7.7 million people who have left the country during its protracted crisis.

Venezuela opposition leader María Corina Machado, in exile after leaving Venezuela in December, took to X to send prayers and wish strength to Venezuelans.

“May strength, serenity, and solidarity prevail among us in the face of this difficult time,” she said on X.

Venezuela Interior Minister Diosdado Cabello said the quake could be felt in several states and asked motorists to give way to ambulances and other emergency vehicles.

“We understand that some people may be desperate, but we are acting according to protocols to activate aid and rescue efforts to help those who need it most,” Cabello said, urging people to remain outside as aftershocks could further damage structures. “Be very careful with children and the elderly. Call each other and check that no one has been harmed.”

Buildings in Manaus, Belem and Macapá in Brazil's Amazon were evacuated, according to reports on TV Globo. The quakes also were felt in Colombia’s Caribbean and northeast regions, but there were no reports of damages or injuries.

The U.S. Pacific Tsunami Warning Center issued several tsumani alerts in the wake of the earthquakes that were quickly lifted.

Strong earthquakes are unusual in Venezuela. While the country sits near multiple fault lines, its position straddling the South American and Caribbean plates make earthquakes much less common than in other parts of Latin America.

Earthquakes are frequent along the Pacific coast, including in Mexico and Chile, which both sit along the seismically active tectonic belt known as the Ring of Fire, an area that the USGS said is responsible for 90% of earthquakes.

Garcia Cano reported from Bogota, Colombia. Associated Press writers Clara Preve in Buenos Aires, Astrid Suarez in Bogota, Colombia, Mauricio Savarese in Sao Paulo and Anna-Catherine Brigida, Megan Janetsky and India Grant in Mexico City, Cristina Fuentes in Madrid and Maria Teresa Hernandez in Beijing contributed to this report.

A boy comforts his mother after an earthquake in Caracas, Venezuela, Wednesday, June 24, 2026. (AP Photo/Pedro Mattey)

A man holding a dog cries after an earthquake in Caracas, Venezuela, Wednesday, June 24, 2026. (AP Photo/Javier Campos)

Rescue workers search through the rubble after an earthquake in Caracas, Venezuela, Wednesday, June 24, 2026.(AP Photo/Javier Campos)

Rescue workers search through the rubble of a collapsed building after earthquake in Caracas, Venezuela, Wednesday, June 24, 2026. (AP Photo/Ariana Cubillos)

Rescue worker carry an injured man after an earthquake in Caracas, Venezuela, Wednesday, June 24, 2026. (AP Photo/Ariana Cubillos)