Nvidia's Kyber AI Rack Delayed to 2028 Due to Manufacturing Issue
Nvidia's Kyber AI rack, a server cabinet housing 144 of its most powerful GPUs, has been delayed to 2028 due to a manufacturing issue with a single circuit board. The delay impacts Nvidia's ability to scale up its most powerful Rubin Ultra systems, potentially creating a window for competitors.
Key points
- Nvidia's Kyber AI rack, a server cabinet housing 144 of its most powerful GPUs, has been delayed to 2028.
- The delay is attributed to a manufacturing issue with a single circuit board, the PCB midplane, which remains challenging to build at scale.
- The delay impacts Nvidia's ability to scale up its most powerful Rubin Ultra systems, potentially creating a window for competitors.
- Nvidia had a backup plan to bolt two current-generation racks together, but it was scrapped due to customer feedback.
- SemiAnalysis says Nvidia now has 'no proven solution' to scale up its most powerful Rubin Ultra systems.
Nvidia's Kyber AI rack, a server cabinet designed to house its 144 most powerful GPUs, has been delayed to 2028 due to a manufacturing issue with a single circuit board. The Kyber rack is crucial for training and running the largest AI models, and the delay is a significant setback for the company.
The issue lies with the PCB midplane, a multi-layer circuit board that ties the entire system together. According to research firm SemiAnalysis, the board remains challenging to build at scale, which has led to the delay. The knock-on effects of this delay are likely to spread to other systems, such as the NVL576, which links eight racks with optical cables.
Nvidia had initially planned to bolt two current-generation racks together as a temporary solution, but this idea was scrapped due to customer feedback. Cloud customers were not pleased with the 'odd design and heavy operational burden' of this setup, and Nvidia ultimately decided to abandon it.
The delay is a rare setback for a company that typically ships a new architecture every year. As a result, Nvidia now has 'no proven solution' to scale up its most powerful Rubin Ultra systems, according to SemiAnalysis. This creates a window of opportunity for competitors to fill the gap and potentially gain an edge in the market.
Sources
The WireByte editorial team synthesises technology news from multiple primary sources, verifies the facts, and links every source. Articles are produced with AI assistance and reviewed under our editorial policy.