NVIDIA Leads AI Infrastructure Capex Surge Amid H100 Bottlenecks

NVIDIA remains the bellwether for AI infrastructure spending, and recent announcements reinforce the narrative of sustained heavy capex from cloud providers and tech giants. A new $3.4 billion five-year contract with IREN could add approximately $700 million in annual recurring revenue as the firm targets $3.7 billion by end of 2026. This deal signals continued strong demand for NVIDIA's high-end accelerators and underscores the company's pricing power in an environment where supply of advanced chips remains constrained. The Bottleneck Map framework, recently updated, highlights that advanced packaging, HBM makers, and materials suppliers are benefiting alongside NVIDIA as each successive layer of AI infrastructure complexity requires new solutions. However, the share rally has also attracted skeptics who question whether peak capex is approaching. Western Digital has outperformed NVIDIA over the past month, suggesting that some investors believe the best gains from pure semiconductor upside may already be priced in and that broader storage and infrastructure names offer better risk-reward. Jensen Huang's inclusion in the Trump China delegation has fueled speculation about potential easing of export controls, which would unlock even larger addressable markets. If H200 chips become available to Chinese AI labs, NVIDIA's total addressable market expands materially. The debate hinges on whether capex growth sustains or plateaus; if AI model training reaches efficiency saturation, spending may moderate. Long-term, the installed base of GPUs continues to drive inference workloads and ASP upside, supporting a bull case. Near term, geopolitical risks around US-China relations and potential export restrictions could trigger a correction if the summit disappoints.