Nvidia shifts focus from chips to full AI factory deployments, introducing containerized designs and secure agent architectures at GTC Berlin.
Key points
- Nvidia introduced DSX POD, a reference design for complete Vera Rubin AI factories built in containers.
- The company launched the OpenShell Reference Architecture to help organizations run AI agents safely.
- CEO Jensen Huang emphasized that deployment speed and power constraints now drive AI economics.
Nvidia announced new infrastructure products and architectures at GTC Berlin to address deployment bottlenecks in artificial intelligence. The company is shifting its strategy from selling individual components to providing complete operating models for AI factories.
What happened
At GTC Berlin, Nvidia Chief Executive Jensen Huang focused his keynote on deployment challenges rather than new silicon. According to SiliconANGLE, Huang argued that AI constraints have shifted to deployment speed, power, security, and data utilization. He stated that AI is now a manufacturing industry producing intelligence through infrastructure. Nvidia introduced DSX POD, a reference design for complete Vera Rubin AI factories built by partners in containers.
This design aims to hide complexity from customers and embed construction work into the factory. The company also launched the OpenShell Reference Architecture to run agents safely. Additionally, Nvidia expanded its Agent Toolkit with an Agent Knowledge Foundation and Kumo structured-data models. Huang noted that a one-gigawatt AI factory costs about $50 billion and generates roughly $40 billion in annual revenue. He warned that delaying deployment by a quarter could result in $10 billion in lost revenue.
Why it matters
Large AI projects are complex construction efforts involving specialized labor and extensive cabling. Delays in assembly directly impact revenue potential for operators. By containerizing factory designs, Nvidia aims to reduce the time hardware sits idle. This approach addresses the economic risk of slow deployment in high-cost infrastructure projects. Teams usually need to coordinate hardware, power, and network setup simultaneously. Standardizing these components can streamline the process for enterprise customers.
The fix
- Review current AI deployment timelines for potential delays.
- Assess whether existing infrastructure aligns with containerized reference designs.
- Evaluate security protocols for running AI agents in production environments.
- Monitor vendor announcements for updates on factory construction standards.
Why it matters
Organizations building AI infrastructure must consider the total cost of deployment, not just hardware procurement. Idle equipment represents a significant financial loss in large-scale projects. Standardized reference architectures can help teams manage complexity and reduce assembly time. Security considerations for AI agents are becoming critical as these systems handle sensitive data. Teams should evaluate how new tools integrate with existing security frameworks. Planning for power and cooling requirements early can prevent costly delays during construction.
What to do and how to stay safe: Nvidia
- Audit your AI infrastructure deployment schedule for potential bottlenecks.
- Evaluate the security posture of any AI agents you plan to deploy.
- Consult with vendors about containerized factory designs to reduce assembly time.
- Review power and cooling requirements for large-scale AI projects.
Step-by-step guide: What is password policies and how they protect systems
General security guidance from the Crypt Weekly newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is DSX POD?
DSX POD is a reference design for complete Vera Rubin AI factories built by partners in containers, introduced by Nvidia at GTC Berlin.
Why is Nvidia focusing on deployment speed?
According to SiliconANGLE, delays in AI factory deployment can result in significant revenue losses, with a quarter's delay costing about $10 billion.
What is the OpenShell Reference Architecture?
OpenShell is a reference architecture launched by Nvidia to help organizations run AI agents safely within their infrastructure.
