Industrial AI Cloud – The cloud infrastructure behind our Foundation Model

Our Robotic Foundation Models are built on extensive datasets that enable them to recognize patterns, understand complex relationships, and perform highly accurate predictions and actions. To support this capability, they rely on powerful cloud infrastructures designed to store, process, and seamlessly scale vast amounts of data. As a leading provider of AI-driven automation solutions, we harness the first cloud platform purpose-built for industrial AI - the Industrial AI Cloud by Telekom and NVIDIA.

Large datasets

Data forms the foundation of intelligent systems. At Agile Robots, we make technology more powerful and adaptive by combining real data from industrial processes, synthetic data from simulations, and human data from our data farms. The more diverse and higher quality the data, the more accurate the analyses and decisions become.

Our systems learn from real sensor data to execute movements precisely, apply forces efficiently, and proactively adjust workflows. This enables smoother processes, reduces downtime, and allows production to flexibly adapt to changing conditions.

Digital data, physical results

In practical terms, this means our robots no longer require complex programming - they perform tasks largely autonomously. The range of applications is extensive, including AI-driven server maintenance, which we showcased live at automatica 2025. Visitors could instruct our dual-arm robot H10-W simply using natural language.

After receiving a command, the robot executes it with remarkable precision: it autonomously moves to the server rack, identifies the hard-drive slots, and carefully removes the existing drives with tactile sensitivity. It then inserts the appropriate storage devices into the empty slots with exact placement.

The entire process operates almost exclusively based on foundation models with zero-shot capabilities. Data plays a central role: it serves as the interface between digital input and physical action, while also representing the outcome of extensive training. Long before a robot executes a task, this data is systematically collected to train our Robotic Foundation Model.

This training goes beyond individual actions – it enables a comprehensive understanding of robotic tasks: we leverage imitation learning to bypass the complexities of classical robotics, allowing the system to learn directly from demonstrations end-to-end for specific behaviors. 

Computing power requirements

Processing large and diverse datasets requires substantial computing resources. The larger the datasets, the higher the demands for computing power, storage, and network bandwidth. Every sensor measurement, simulation, and human input must be captured, processed, and analyzed in real time to ensure our models learn accurately and make reliable decisions.

At the same time, the models themselves grow in complexity as the data volume increases, making computations increasingly demanding. Without powerful infrastructure, training times would be significantly longer, analyses delayed, and adaptation to new scenarios more difficult. To fully leverage the capabilities of our Robotic Foundation Models, a highly scalable and efficient computing architecture is therefore indispensable.

Industrial AI Cloud

For this reason, we leverage the Industrial AI Cloud by Telekom and NVIDIA, specifically designed for industrial artificial intelligence. It combines highly scalable GPU clusters, powerful storage, and high-speed network connections to provide our Robotic Foundation Models with real-time access to data. This enables even complex computations and training processes on vast datasets to be carried out efficiently. The Industrial AI Cloud is also used to generate synthetic training data.

As Europe’s first cloud platform specifically focused on industrial AI, the Industrial AI Cloud delivers sovereign computing power and strengthens the technological competitiveness of Germany’s industrial sector. It allows companies to efficiently process large industrial datasets, train complex AI models, and optimize production processes in real time - all while adhering to the highest European data protection and security standards. 

Strong collaboration

NVIDIA and Agile Robots have maintained a long-standing collaboration. Agile Robots was an early adopter of NVIDIA Omniverse libraries, took part in NVIDIA GTC 2025 in San Jose, and demonstrated a project powered by the world-class AI computing platform NVIDIA Jetson at automatica 2025.

The collaboration with Telekom extends this partnership by adding a robust cloud solution. By integrating NVIDIA’s advanced AI computing technology and open robotics platform, NVIDIA Isaac, Agile Robots’ expertise in industrial robotics, and Telekom’s secure, high-performance cloud infrastructure, a comprehensive ecosystem for industrial AI applications emerges.

This combination accelerates innovation cycles, enhances scalability, and advances the transition toward a connected, autonomous industrial landscape.