NVIDIA’s Upcoming Rubin AI Servers Expected to Ship by August 2023, Key Supplier Quanta Reports Smooth Mass Production

NVIDIA is poised to deliver its next-generation Vera Rubin AI racks to customers by the third quarter of this year, although the initial production levels are expected to be limited.

Suppliers Anticipate Smooth Production Onboarding of Vera Rubin Due to Retained Blackwell Components

In a notable announcement, NVIDIA has confirmed that the Vera Rubin platform entered full production in the first quarter of 2026. This milestone exceeded market expectations, showcasing NVIDIA’s commitment to an accelerated product rollout. While the timeline for hyperscalers to receive their first batch of AI racks remains undisclosed, Quanta’s Executive Vice President, Mike Yang, indicated that customers could start receiving initial units as early as August 2026, facilitating complete integration into hyperscaler systems by the end of the year.

The transition from chiplet designs to advanced packaging and high-bandwidth memory (HBM) integration raised concerns regarding NVIDIA’s ability to meet the anticipated “H2 2026″schedule. However, Yang reassured that many components in the Rubin infrastructure are derived from the Blackwell series, thus minimizing production risks at this stage. It is expected that NVIDIA will adopt a similar initial low-volume shipping strategy as seen with Blackwell Ultra, subsequently increasing production early next year.

A person is standing on stage showcasing various open server units with visible cooling systems and hardware components. — Image Credits: NVIDIA

The Rubin lineup was extensively introduced at this year’s CES, showcasing a remarkable array of six chips that enhance both networking and computational capabilities. The featured chips include:

Rubin GPU (336 Billion Transistors)
Vera CPU (227 Billion Transistors)
NVLINK 6 Switch for Interconnect
CX9 & BF4 for Networking
Spectrum-X 102.4T CPO for silicon photonics

Although NVIDIA has announced an NVL144 configuration at last year’s GTC, initial shipments of the Rubin series are expected to be limited to the NVL72 configuration. This strategy aims to keep the chip count manageable, thus controlling thermal output and streamlining the production process. NVIDIA’s experiences during the transition from Blackwell to Blackwell Ultra have highlighted the complexities involved when adapting to architectural changes, especially with the implementation of the Cordeila board design.

Image Credits: NVIDIA

NVIDIA’s Rubin AI series is on track to be fully adopted by hyperscalers between Q4 2026 and Q1 2027, with future models expected to significantly enhance performance. Early adopters of this cutting-edge technology include industry giants such as OpenAI, Microsoft, Google, and innovative cloud players like CoreWeave.

Source & Images