A Circuit Board Problem Just Delayed Nvidia's Next AI System to 2028 And Chip Stocks Are Already Feeling the Fallout - 24/7 Wall St.

Recent reports from semiconductor research firm SemiAnalysis suggest that NVIDIA’s most advanced AI system, Kyber NVL144, has been delayed until 2028, pushing back its originally planned 2027 launch. ... Read original →

The Kyber delay reveals a hard physical ceiling in AI infrastructure: the 78-layer PCB midplane’s signal integrity and thermal constraints can’t be solved by transistor scaling alone. This forces substrate suppliers like Kingboard Laminates (Hong Kong, China) and Ibiden to accelerate low-loss, high-frequency material R&D, while pushing EDA and co-packaging tools to evolve. From a compliance standpoint, yield sensitivity amplifies supply chain concentration risks—especially as advanced substrate capacity remains heavily anchored in Taiwan, China and Japan. Competitors like AMD and Broadcom may pivot to chiplet-based architectures to sidestep monolithic complexity, while Google could fast-track TPU v6 deployments to lock in cloud partners. Over the next 18 months, the industry’s battleground will shift from raw compute to system-level manufacturability; NVIDIA’s 81% market share remains intact, but its moat is now defined less by GPU performance and more by full-stack engineering execution.