February 10, 2023
Morgan Stanley Research
Joseph Moore, Equity Analyst

Stock rating: Equal weight, Target $175

"We have been excited for deep learning for large language
models for several years - it has been the focus of our
constructive view since 2018. ChatGPT can be a catalyst for
expansion. We think that some ests for near term direct
revenues are too high, but the long term opportunity is
significant.

Why we are writing this: We have seen several reports that in our view
incorrectly characterize the direct opportunity for NVIDIA, in particular the
revenue from Chat GPT inference. We think that GPT 5 is currently being trained
on 25k GPUs - $225 mm or so of NVIDIA hardware - and the inference costs are
likely much lower than some of the numbers we have seen.Further, reducing
inference costs will be critical in resolving the "cost of search" debate from cloud
titans.

That being said, we think the broadening enthusiasm for large language models is
a very important development for the longer term prospects deep learning
models for natural language, and we are not trying to be curmudgeons on the
opportunity. But we have talked to several industry participants about these
workloads and do think we have some important context.

Our first instinct when the Chat GPT enthusiasm started to build was that this is
not new, but our views have changed and we do think that this can drive an
inflection in the usage model for generative AI. We have demoed software with
these types of capabilities for some time, but putting them out into the public for
experimentation seems highly likely to act as an accelerant for conversational AI,
commoditizing human activity on a large scale.

Using deep learning for natural language is still at an early stage, but it has been
a key element to the story for sometime; in fact, it was central to our NVIDIA
upgrade in 2018,as deep learning activity was growing beyond vision applications
into large language transformers. At that time, we wrote that expectations for
GPU technology to become a key enabler of machine learning training, and then
inference workloads as well, was the reason to stay enthused about the stock despite
remaining expensive on near term numbers and cyclicality in the consumer-facing gaming business..."