JP Morgan reports that Nvidia plans to sell fully assembled AI servers, starting with its Vera Rubin platform. This move consolidates supply chain roles, allowing Nvidia to produce integrated compute trays while partners focus on assembly and support. The change could enhance efficiency but also strengthen Nvidia's market dominance.

Nvidia introduced its Vera Rubin architecture, promising significant efficiency gains in AI workloads by reducing inference costs and GPU requirements. The system features six new chips, including advanced networking components, designed to enhance performance through improved GPU connectivity and in-network computing.

Nvidia's CEO Jensen Huang announced new AI server systems, called Vera Rubin, at CES in Las Vegas, set to launch later this year. The company is accelerating chip development to meet the growing demand for powerful processors needed for advanced AI training and simulations.

Nvidia has unveiled its vision for gigawatt AI factories at the 2025 OCP Global Summit, featuring the Vera Rubin NVL144 architecture, which supports advanced liquid-cooled servers and modular expansion for AI workloads. The architecture aims to enhance data center efficiency and scalability, while Nvidia's Kyber server architecture and Spectrum-X Ethernet switches will further optimize performance and energy efficiency in AI infrastructure. Notably, Meta and Oracle are set to adopt these innovations to improve their data center operations.