Specialist
Former Technical Product Manager, Cloud AI & HPC at Nvidia Corp
Agenda
- AI data centre applications for hyperscale and enterprise customers
- Alternative AI inference processors competing with Nvidia (NASDAQ: NVDA) including CPUs (central processing units) and ASICs (application-specific integrated circuits)
- Revenue growth outlook and promising industry verticals for potential expansion such as healthcare
- Enterprise AI platform adoption and processor development roadmap
Questions
1.
Could you discuss Nvidia’s operating environment, highlighting the trends you are monitoring?
2.
What are your adoption expectations for Nvidia’s GPU [graphics processing unit] portfolio for AI and highperformance computing applications across tier 1 and tier 2 cloud providers?
3.
What are your expectations for visibility on hyperscale demand? Do you expect demand in general to become less cyclical? Could the visibility of demand materially change?
4.
How are you thinking about the penetration for Nvidia’s offerings for hyperscalers within tier 1 and tier 2 cloud operators? What percentage of servers could be accelerated for hyperscaler players by Nvidia and how might that change? Is there a ceiling based on some the internal work at some hyperscalers?
5.
Could you outline Nvidia’s growth in the core compute networking segment? What proportion of that is driven by cloud revenue growth vs vertical industry growth and enterprise-type opportunities? Is there are a split that we can work with? What are the potential growth rates for those sub-segments or the customer sets within that particular segment of compute networking?
6.
What are your thoughts on the uptake of the vertical industry and enterprise-type business for Nvidia?
What could be the contribution from this enterprise or industry vertical type business vs the cloud business?
7.
Could you outline the pace of adoption for Ampere Architecture Tensor Core GPUs for internal or external worlds for hyperscalers? You suggested that they will buy as much as they can get their hands on. Do you think that the demand and pace of adoption for this particular generation – considering the hyperscale compute revenue doubled YoY – could be replicated by Nvidia in 2022 as well?
8.
Can you estimate the breakdown for Nvidia data centre revenue by application type? You referenced highperformance compute and machine learning.
9.
What are your growth rate expectations across inference vs training vs machine learning?
10.
What’s driving adoption of GPUs for inference vs CPUs [central processing units], or even particular ASICs
[application-specific integrated circuits] that are coming to market and trying to win in the inference segment?
What are hyperscalers doing in the inference segment and with their own custom chips? Why are GPUs
becoming more widely used for inference?
11.
What are your thoughts on inference chips through custom design, maybe for any specific workloads or particularly for the internal workloads? The thinking must be that for specific applications it makes more sense to use internally developed chips that are designed for specific tasks. Is that a natural ceiling to the TAM for a company such as Nvidia, that those are types of workload that the company’s GPU platforms will really struggle to address? What is the threat hyperscalers’ internal design work?
12.
If GPUs are taking share in AI inference, what niche could the Grace processor – the Arm-based CPU – fill in the portfolio? How might that be positioned when it rolls out in 2023?
13.
What are your thoughts on server GPU and the positioning of AMD vs Nvidia? What threat does AMD pose to Nvidia? Could AMD ever catch up to Nvidia for server GPUs?
14.
What would AMD have to do to indicate that it could pick up ground in Nvidia? Is it mostly about the software development and increasing available developer tools?
15.
Could you outline the pace of adoption for Triton inference server stacks? What are your thoughts on the TensorRT software stack vs TensorFlow or PyTorch? What are the mechanics behind the way in which Nvidia will take its AI inference offer in the market and how might that play out?
16.
Could you elaborate on the industry verticals that showed promise such as the healthcare sector? How quickly are projects being put out into the market or being worked on? How quickly could demand scale up for those? When might revenue from those projects materialise?
17.
Could you elaborate on ramps on projects in the healthcare sector as an example? How long before revenue could scale in a typical drug discovery or medical imaging project? How might revenue per customer or application trend in healthcare? What are your timing and revenue expectations?
18.
What are your thoughts on the Nvidia AI Enterprise stack, highlighting your adoption expectations for that software? To what extent is that correlated or in lockstep with hardware sales? How fleshed out is that particular stack?
19.
What are your expectations for Nvidia’s development strategy around software portfolios and the roadmap to support Enterprise AI adoption? You referenced how wide it is. Are there still gaps in the development roadmap?
20.
A big part of the most recent Nvidia GTC conference was dedicated to Omniverse and the opportunities there. How would you size the opportunity for Omniverse for Nvidia? Could you describe the intersection between Nvidia and the Metaverse? What’s the opportunity for Nvidia there?
21.
What are the key risks for Nvidia over the next few years and why?
