Specialist
Former executive at Codasip sro
Agenda
- Evolution of customer spend on AI hardware, pricing and TCO (total cost of ownership) trends
- Competitive dynamics vs Nvidia (NASDAQ: NVDA), Intel (NASDAQ: INTC), Tenstorrent, Ampere, Blaize, Cerebras, Groq and SambaNova
- Strengths and weaknesses of hardware approaches vs IP approaches such as Imagination Technologies and Codasip
Questions
1.
I think it’s safe to say Graphcore has had quite a difficult time over the past 12-24 months or so judging by its most recent reporting. What would you identify as the key industry headwinds that are impacting the company?
2.
It seems you think there is a chance for Graphcore to succeed despite the difficulties it has experienced more recently. Although Nvidia has been valued at USD 1tn, I think Graphcore has around USD 730m in investment behind it. We’ve discussed why the AI gains have mostly been concentrated on legacy players such as Nvidia and Baidu, but could you elaborate on why more of the gains haven’t happened with Graphcore? You said the company has a very good team – one of the best architects in the world is part of the team, and with a lot of these start-ups it is all about team. Do we just credit the team with managing to keep Graphcore alive despite other companies folding?
3.
What is it about the wafer-on-wafer approach to the Bow IPU [infrastructure processing unit] processor that makes you think that this could be a huge opportunity? I believe there is a wafer-on-wafer deal with TSMC, but why is this new approach to Graphcore’s hardware such an opportunity, especially knowing the company has historically been excellent at marketing?
4.
Could you quantify the performance and power advantages that you see compared to where Graphcore was in 2021, when you would have been a lot less positive about the company? How much closer would you say it is to where you think it needs to be in terms of being able to offer customers a compelling alternative to Nvidia? What’s the progress since 2021? How does the wafer-on-wafer aspect help?
5.
Is there any kind of moat around Graphcore’s wafer-on-wafer technology?
6.
You mentioned that Graphcore had not been as good in the field of software as other players that had developed much more robust ecosystems. How does RISC-V change any of the dynamics with numerous players having different ecosystems instead of there being one ecosystem that everyone could use? Does that have advantages or disadvantages to companies such as Graphcore?
7.
What happens to CPUs [central processing units] with RISC-V? Could anything happen to the GPU [graphics processing unit] ecosystem, which is an equivalent to RISC-V? If that happened, would that have any positive or negative impact on Graphcore?
8.
On the cloud side, one of the more significant things that happened recently was the Microsoft deal with Nvidia. Microsoft isn’t using Graphcore chips when it comes to its data centres – it’s developing its own AI chips. Obviously, there are a few different approaches to achieving the outcome of an AI chip needed for AI processes – companies are developing their own AI chips and then licensing technology from people, while companies such as Nvidia are licensing IP and then producing their own chips. There are also the AI chip hardware companies that do everything, such as Graphcore. Why does this latter approach seem to be losing?
9.
What would make a company such as Microsoft decide not to go with Graphcore’s approach? If there was a loose deal or agreement with Graphcore, which could have been pursued, why switch to a strategy of developing its own chips for its data centres?
10.
Are you saying that if Graphcore can survive for the next 10 years, we can expect a massive upside because of secular trends?
11.
Graphcore has 633 employees around the world and you just mentioned that talent is obviously scarce in this particular industry. People might have been surprised about the company’s lay-off announcement, although it hasn’t been that specific. Is that wise, in your opinion?
12.
Where do you think the cuts should be made? What percentage of the 633 employees do you think should be cut?
13.
Of the exit routes we were discussing, the first one is business as usual – carry on trying to win business and succeed as a company. Graphcore CEO Nigel Toon talks about the potential in China or the Asia market, where Nvidia might find it more difficult to win business because of US-China geopolitical tensions. Is it reasonable to think that these geopolitical tensions are so much worse than those between the UK and China, and would give Graphcore more of an opportunity vs Nvidia in that market?
14.
Could you elaborate on closer-to-home opportunities, specifically the EU? Obviously, in the UK, the Chancellor has recently announced a new GBP 900m exascale super computer – Nigel Toon wrote to the UK Prime Minister and the Chancellor about using Graphcore systems for that super computer. What’s your assessment of the opportunities in the UK and the EU? What’s the likelihood of Graphcore winning that business over Nvidia?
15.
When you look at the size of the opportunity in Europe – including the UK and the EU – how confident are you that Graphcore will be able to continue in its current form with the funding it has, maybe being able to raise more funding when things improve? Are you confident that the company’s financial situation will improve in the next 24-36 months?
16.
What would be the other exit routes if Graphcore weren’t able to survive?
17.
Is there anything else to mention about Graphcore as a conclusion?