Summary
Highlights
Nvidia's fiscal third-quarter forecast met consensus but fell short of highly optimistic estimates, causing some concern about waning explosive growth. The stock dropped almost 7% in after-hours trading, largely due to a lack of precise reassurance from management regarding Blackwell sales, despite strong year-to-date performance.
Jensen Huang clarified that the Blackwell issue was a production mechanism problem, not a design flaw. A 'mask change' was made to improve manufacturing yield. Functional samples are currently being distributed, and volume production is set to begin in Q4, with billions in revenue expected from Blackwell.
Nvidia is highly credited for its full-system approach, integrating GPUs, CPUs, wiring, cabling, and firmware, creating a strong moat against competitors. While hyperscalers still dominate as core customers, making up 45% of data center business, demand is broadening to include internet service providers, sovereign AI initiatives, and various enterprises.
Jensen Huang explains that sovereign AI refers to deals with regional service providers, often government-funded, in countries like Japan, India, Canada, UK, France, Italy, Singapore, and Malaysia. These nations view digital data as a natural resource and are investing heavily to build out their AI infrastructure to process and transform it into national digital intelligence.
Nvidia addresses the increasing energy demands of AI by improving the performance and efficiency of its next-generation chips, like Blackwell, which offers significantly higher performance at the same power usage as its predecessor. They also support liquid cooling for greater energy efficiency and anticipate AI training will increasingly occur elsewhere, with smaller models being deployed at the edge.
Nvidia anticipates improved supply sequentially through Q1 next year to meet the tremendous demand. The growth of foundation models, which are increasing in size and learning multiple modalities (languages, images, sounds, 3D graphics, proteins, chemicals, physics), drives this demand. The generative AI market is also diversifying beyond internet service providers to include startups, enterprises, and countries.
Nvidia's GPU cloud is designed to be the best version of Nvidia's cloud built within existing cloud infrastructures like GCP, Azure, AWS, and OCI, rather than being a standalone cloud compute provider. This strategy ensures optimal performance and TCO. Nvidia also uses its cloud internally for chip design, software development, self-driving cars, robotics, and Omniverse, and offers it as an AI foundry service for other companies.