SemiAnalysis notes Nvidia lacks a proven path to scale Rubin Ultra for large deployments, potentially giving AMD and Google an edge in high-end AI infrastructure.

Nvidia’s Kyber NVL144 rack-scale AI architecture is delayed to 2028 due to manufacturing bottlenecks in the 78‑layer PCB midplane, the core interconnect for 144 GPUs in a single rack.

A larger NVL576 system, which would use Co-Packaged Optics, faces readiness and production uncertainties that could push back mass shipments and create room for competitors.

An even bigger NVL576 variant linking eight racks is also expected to be delayed or limited in shipment.

Nvidia reported 82 billion dollars in revenue for the latest quarter, with 75 billion from data center and free cash flow of 49 billion, underscoring strong demand for AI infrastructure.

There are monetization opportunities in software layers that optimize existing hardware and through consulting for hybrid cloud AI deployments, though migrations must maintain performance parity and be phased with benchmarking.

TIKR’s valuation tools are highlighted as a way to estimate fair value and upside, though the article includes affiliate links and promotional content.

Insiders sold about 410.6 million dollars worth of NVDA shares in the last three months, signaling caution about near-term prospects.

The delay could spur diversified supply chains and open standards for rack-scale AI, emphasizing software-defined infrastructure and resilient sourcing to reduce hardware dependency.

Nvidia’s stock reaction was muted despite the delay, as investors focus on strong current-generation performance and record earnings.

Rivals like AMD and Google may benefit as customers explore alternatives amid Nvidia supply constraints and questions about scaling Rubin Ultra for large deployments.

Market and regulatory implications include potential higher costs for early adopters, export-control considerations for advanced chips, and the need for diversified hardware vendors to mitigate supply risks.