Nvidia and Generative AI
Yes, Nvidia’s data center business will continue to thrive when the company reports results in late November as Generative AI and the large language models (LLM) that underpin Gen AI require a multi-quarter build for companies like ChatGPT/ Azure/ MSFT and Google. That’s not to say that ChatGPT, Google and the various LLM startups that supply AWS and Oracle will stop building LLMs. No. LLMs are continually refined and constantly evolving. It’s just that I believe this initial LLM ramp cycle will start to slow beginning now and will become apparent as we enter 2024. Eventually the lack of demand from end users will filter back to Nvidia, likely in 1H 2024.
The initial LLM build phase is bound to slow early next year in my view. I suspect that many enterprise customers will not have experienced a great ROI on their Generative AI experiments (this is more of a function of not properly defining the project scope, not understanding data on hand, and not defining the expected ROI). Therefore, corporate customers are likely to slow spending on Generative AI projects, which translates to less demand for Generative AI capabilities from ChatGPT/ Azure/ MSFT, Google and AWS/ Amazon. This of course means less pull demand for Nvidia’s Generative AI chips. I believe we will hear less about Gen AI as Q4 moves along (I hear less about Gen AI from investors and talking head media types) and as we enter 2024.