Nvidia’s Kyber AI Rack Delayed to 2028: A Circuit Board Conundrum
Nvidia’s Kyber AI rack, initially planned for 2027, has faced a significant delay until 2028 due to challenges with a crucial circuit board, according to research firm SemiAnalysis. This setback presents a unique opportunity for rivals like AMD and Google.
The Issue at Hand
The Kyber rack is designed to house Nvidia’s powerful Rubin Ultra chips. However, the primary hold-up lies in a multi-layer circuit board (PCB) known as the midplane, which connects all the chips within the rack. SemiAnalysis points out that manufacturing this board on a large scale remains a complex task.
Impact on Nvidia and Competitors
This delay has broader implications:
- Nvidia’s Scalability: The company now lacks a proven method to scale up its most powerful AI systems, as the Kyber rack is a key component in achieving such density for training and running large AI models.
- Rivals Gain Ground: Google and AMD, both with their own AI chips, can capitalize on this temporary setback by attracting work from top AI labs, potentially gaining an edge in the market.
- Market Response: Asian technology and circuit board stocks dropped following the news, as reported by Bloomberg.
Nvidia’s Short-Term Resilience
Despite this delay, Nvidia’s current Rubin systems are in full production and set to begin shipping later this year to cloud partners like AWS, Microsoft Azure, and Google Cloud. SemiAnalysis predicts strong revenue growth for Nvidia in the second half of its 2027 financial year, meeting or exceeding Wall Street expectations.
Manufacturing Challenges
The root cause of this delay is a confluence of factors:
- Supply Chain Limits: The rapid pace at which Nvidia releases new architectures clashes with the capabilities of its suppliers to manufacture advanced circuit boards and memory.
- Competition in Custom Silicon: Rivals are increasingly focusing on custom silicon, betting that speed, not design, is where Nvidia’s vulnerability lies.
In an unusual turn, it’s not the AI chip but the encompassing hardware that presents the biggest challenge for Nvidia in this era of rapid technological advancement.