Serenity: Nvidia's AI Compute Demand Far Exceeds Supply, Memory Purchase Commitments Rise to $390 Billion
Odaily reports that Serenity stated on X platform that, based on Nvidia's latest earnings call and related regulatory filings, AI compute demand continues to far outstrip supply, with industry chain segments such as memory, advanced packaging substrates, and optical networking expected to benefit sustainably.
Serenity noted that Nvidia disclosed in its SEC filing that its supply purchase commitments increased from $279 billion in the previous quarter to $390 billion, primarily involving multi-year memory supply agreements. Meanwhile, Nvidia raised its year-over-year revenue growth forecast from 70% to approximately 75%, and reached new agreements with Kinsus and Samsung Electro-Mechanics to expand ABF packaging substrate capacity.
On the compute demand side, Serenity cited Nvidia's disclosed information stating that the industry's cloud services backlog increased from $2 trillion to $2.5 trillion within a single quarter. At the same time, Hopper GPU rental prices have doubled, B300 GPU rental prices have risen 60% since launch, and computing resources for hyperscalers and neoclouds have been fully utilized.
Additionally, Nvidia stated that its co-packaged optics (CPO) technology roadmap is progressing well, with quarterly networking revenue exceeding $21 billion, and CPO scale-out switches have entered mass production.
Serenity believes that the above data reflects that AI infrastructure demand is extending from memory and advanced packaging to optical networking and neoclouds, while current market pricing still reflects concerns about a significant industry downturn.
