Nvidia has delayed its next-generation AI rack system, Kyber NVL144, by more than 12 months, pushing its release to 2028. The company has also canceled its NVL72x2 architecture, according to a report from SemiAnalysis cited by CNBC.
The root cause lies in printed circuit board manufacturing problems that have disrupted production timelines. The delay could reshape AI hardware roadmaps, affecting GPU supply chains and potentially altering industry strategies for scaling AI infrastructure.
The Kyber rack-scale architecture was designed to house Nvidia's upcoming 2027 Rubin Ultra chips. The extended timeline means the company's flagship AI system will now arrive a year later than initially anticipated.
This setback may give rivals such as AMD and Intel more time to close the performance gap in AI accelerators. Cloud providers and AI startups reliant on Nvidia's cutting-edge hardware may need to adjust their deployment plans.
While Nvidia remains dominant in AI chips, any prolonged delay risks shifting customer loyalty. The company has not publicly commented on the reported schedule changes.