Nvidia Unveils AI Factory Solutions at Berlin GTC as Jensen Huang Advances Enterprise Agents and Robotics Applications

TradingKey - On October 21, Nvidia (NVDA) CEO Jensen Huang introduced new advancements in AI infrastructure, enterprise agents, and robotics at the GTC conference in Berlin, stating that Vera Rubin has entered full production. The announcements focused on compute deployment and application development, aiming to help enterprises put AI into practical business use.

AI Factory On-Site Deployment Completed in as Fast as 14 Days, Delivery Planned for Early 2027

On the infrastructure front, Nvidia introduced DSX POD, a modular AI factory solution that integrates and tests computing equipment, networking, power, and liquid cooling systems in advance at partner facilities before shipping them to customer sites for installation.

According to Nvidia, when supporting site infrastructure such as power is already in place, the solution can shorten deployment time from approximately six months to as fast as 14 days. Partners include Dell, Hewlett Packard Enterprise, Lenovo, Schneider Electric, and Vertiv, with deliveries expected to begin in early 2027. This timeframe refers specifically to the on-site deployment phase and does not cover the entire process of building a data center from scratch.

From a business perspective, shortening the equipment installation and commissioning cycle helps customers bring computing capacity into operation earlier. However, whether this accelerates return on investment still depends on supporting project progress, compute utilization rates, and actual demand, and cannot be judged solely on faster deployment.

Launches AI Agent Security Architecture to Help Enterprises Connect Internal Operations

In terms of enterprise applications, Nvidia launched the OpenShell secure AI agent reference architecture, which manages agents' access to data and tools through isolated execution environments, access controls, and continuous monitoring.

The architecture assigns model inference, task orchestration, and secure execution to different computing components, aiming to enable enterprises to integrate agents into business processes under controlled access conditions.

Nvidia also introduced the Agent Knowledge Foundation, which equips agents with capabilities such as document parsing, information retrieval, and structured data processing to help them utilize internal enterprise knowledge. Compared with generic Q&A, these features are closer to the foundational tools required for practical tasks like handling customer profiles, business records, and databases.

These announcements show that Nvidia is expanding its enterprise AI coverage through hardware, models, and software tools. However, a gap remains between product launches and large-scale customer adoption, and actual deployments as well as commercial revenue contributions will need to be observed going forward.

Autonomous Driving and Robotics Collaboration Expands, Some Products to Launch Next Year

In the automotive sector, Nvidia showcased partnerships related to assisted driving, autonomous driving, and robotaxis centered around the Hyperion platform. Among them, Volvo Cars adopted the Hyperion and Alpamayo technology stacks, providing a concrete application case for Nvidia's further entry into the automotive computing market.

In robotics, the company introduced the Agile ONE SN humanoid robot reference platform with 31 degrees of freedom, which is expected to launch in 2027. Supporting tools cover data collection, model training, simulation, and operation, helping developers move robots from simulated environments to real-world scenarios.

In the medical field, Nvidia also introduced a collaboration with Siemens Healthineers in adaptive radiation therapy, combining medical imaging, digital twins, and real-time computing to assist in treatment planning and adjustment. While these collaborations broaden the application scope of AI computing, the regulatory, delivery, and commercialization progress of each project still needs to be tracked individually.