Nvidia’s Next-Gen AI Rack System Delayed To 2028, Report Says

Nvidia’s upcoming Kyber rack-scale system, designed to house its 2027 Rubin Ultra chips, has been delayed by more than a year to 2028, according to research firm SemiAnalysis. The system was meant to pack 144 of Nvidia’s most powerful chips into a single unit, giving AI companies the computing power needed to train advanced models.

Circuit Board Proves Difficult To Manufacture

The delay stems from challenges producing a specialized circuit board that connects components inside the system. A larger version linking multiple racks together is also expected to face delays or limited production, the firm said. A backup plan to combine two current-generation racks was scrapped after cloud providers pushed back over its complicated and costly design.

Setback Raises Questions About Nvidia’s Roadmap

The delay adds to a series of reported issues affecting Nvidia’s product lineup, fueling concerns that its rapid annual release schedule is running into real manufacturing limits. Analysts say this could open a rare window for competitors like AMD and Google, whose in house chips are already gaining traction with major AI companies.

Current Products Still On Track

Despite the setback, Nvidia’s current-generation Rubin systems remain in full production and are set to begin shipping this fall to major cloud partners, including Amazon, Microsoft and Google. Nvidia’s data center revenue is still projected to come in well above Wall Street expectations for the second half of fiscal 2027. Nvidia shares showed little reaction in early trading following the report.