Nvidia's Kyber NVL144 AI Rack System Delayed to 2028 on Production Snags

Nvidia's next-generation AI rack system, the Kyber NVL144, has been delayed until 2028 due to manufacturing challenges, according to a report by semiconductor research firm SemiAnalysis. The system, initially slated for a 2027 launch and designed to integrate with the company's Rubin Ultra chips, now faces a significant postponement that raises questions about Nvidia's ambitious product roadmap and its ability to maintain a rapid annual release cadence.

The primary reason for the delay is attributed to persistent manufacturing difficulties with a specialized printed circuit board (PCB) midplane, a critical component at the heart of the Kyber NVL144. SemiAnalysis indicates that this complex midplane, which can feature up to 78 layers, has proven highly susceptible to defects during production. This setback comes just three months after Nvidia CEO Jensen Huang publicly showcased the system at the GTC event. Furthermore, the larger NVL576 system, intended to link eight racks through optical connections, may also face delays or be limited to small production volumes. A fallback design, the NVL72x2 back-to-back rack architecture, was outright canceled following strong pushback from cloud service providers (CSPs) and hyperscalers due to its awkward design and heavy operational burden. The Rubin Ultra roadmap has also been adjusted, with the planned quad-chip version canceled, leaving only a dual-chip design and effectively halving the system's performance. Earlier reports from SemiAnalysis in June indicated delays for Nvidia's 800V DC power delivery architecture and Co-Packaged Optics (CPO) technology until 2028-2029, also citing yield and cost issues.

These delays amplify concerns that Nvidia's aggressive annual product release schedule is encountering significant manufacturing limitations. The market is assessing the potential impact on the pace of AI infrastructure buildout and cloud capacity planning. Analysts suggest that these setbacks could create a valuable opening for competitors such as Advanced Micro Devices (AMD) with its MI500X and Google with its TPU v8i, particularly in the high-end AI infrastructure market. In the short term, funding may reassess the AI capital expenditure cycle, potentially shifting towards stocking existing products like Blackwell rather than aggressive expansion into new, delayed architectures.

Nvidia's encounter with these delays is seen as part of a broader shift within the AI hardware industry. The sector is transitioning from a pure computing power race to one dominated by supply chain resilience and system-level optimization. Capital-intensive investments, particularly under advanced packaging and material constraints, are reshaping the distribution of pricing power. The company's ambitious roadmap is being tested by manufacturing realities, with technical challenges such as high power density and complex system integration highlighting the inherent hurdles in developing next-generation AI infrastructure.

Nvidia has not publicly confirmed the delay and did not respond to CNBC's request for comment regarding the SemiAnalysis report. However, current-generation Rubin systems are reportedly in full production and are expected to begin shipping this fall to key cloud partners, including Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. Analysts are closely monitoring whether these delays represent a temporary engineering hiccup or a more structural challenge related to the maturation of co-packaged optics technology, which could potentially rewrite the entire industry's scaling roadmap.