Analyzing the Data Center Business with Akram’s Razor (Part 2)

Last week was Part 1 of our conversation with Akram’s Razor where he gave his account of how the crypto crash of 2018 impacted Nvidia and AMD. This week, Akram and James go through Nvidia’s data center business in depth and assess the growing competitive landscape for AI accelerators, from cloud and startup companies alike. Akram shares more about how he approached his research for his thesis on Nvidia and we get into a serious discussion about Google’s AI advantage and why he believes they are not making their TPU training chips freely available. We then debate Nvidia’s pricing for their data center boards and talk about their CEO, Jensen Huang, and his refreshing transparency and openness.
Key Points from This Episode:
The substantial market value that Nvidia lost in 2018 after the crypto mining crash
Getting familiar with TPU and other research for his “Nvidia: A Data center AI Bear Thesis”
Talking to people at Google about them using Nvidia’s GPU
Google making TPUs available and competing with Amazon Web Services (AWS)
AI as Google’s biggest advantage and why their price is not coming down
Speculating over why Google isn’t giving away their TPU training chips
The approaching competition for TPU and its impact on Nvidia’s data center business
The large number of startups that are raising capital for deep and machine learning
A theory that training is starting to get saturated at the kernel level capacity
Nvidia’s prices for their data center boards and why Akram believes they were over earning
Why Nvidia still appears to be well-positioned in the market
 
Tweetables:
“Nvidia last year, lost more in market value than any semiconductor company since Intel crashed in 2000.” — @akramsrazor [0:03:21]