I missed last week’s newsletter due to a life event.
Let’s dive in.
Value Capture in AI
The levels of the AI stack where value capture may happen:
Chips - Nvidia is killing it!
Foundation Models - Open AI and other open-source models such as LLaMA and LLaMA 2.
Infrastructure Dev Tools
Application
Jensen Huang unveiled Nvidia Blackwell at the NVIDIA conference for developers. Huang also revealed that GPT-4 contains 1.8T parameters and that 2000 Blackwell chips could finish training the model in 90 days! These chips will cost $30-$40K per unit.
If you are an AI startup wanting to compete with the who’s who of the world's foundation models, you need to invest $80M to train a new model! Unless a marquee investor backs you or you have all the money at your disposal, this is a distant dream.
The first two levels of the stack above will be led by existing players with a head start. There could be opportunities to create a niche and capture a small market segment, but competing with Nvidia and OpenAI will be difficult. The R&D investment required in this space is a huge deterrent for startups.
At level 3 of the stack, you will find players who will help AI companies build, train, and deploy machine learning models at scale. Here, as well, there are existing players who will dominate. AWS provides a machine learning service called Sagemaker, SAP's Leonardo Machine Learning Foundation central hub for machine learning within SAP's Cloud Platform. There are other players, such as Datarobot and Skymind. At this stack, you require cloud computing, and existing cloud providers like AWS, GCP and Azure will find it easy to provide these services compared to a new entrant.
Level 4 of this stack is an open field. Here you will find hundreds of Startups that innovate and create products that solve niche problems. We are very early in this space and are still exploring the use cases where AI will meaningfully solve business problems. The AI apps you see today only scratch the surface of what’s possible.
The next few years will be exciting in the AI product space!
Source - You can listen to David Sacks talking about this AI framework here.
Where you go to college matters, but not for reasons you think
In India, if you are from IIT and/or IIM (Ivy leagues), more opportunities open up relative to others. The label will get you inside, but your work ethic and talent will help you grow. These Ivy Leagues, anywhere in the world, attract top talent. The network you build at these places is intangible. The side projects you build, the hobbies you practice, and the friendships you make will help you multi-fold in the future. This seemingly simple insight is why there is so much competition to get into these Ivy leagues. If you are not from the Ivy League, like me, to build such a network takes a long time, and sometimes you cannot do it. It’s not impossible but the path is long drawn and difficult.
In this episode of the NextLevel podcast titled “All things fintech”, Harshil Mathur (Co-founder of Razorpay) and Mridul Arora (Partner at Elevation Capital) share their journeys of going to IIT for engineering and then landing jobs in Mechanical and Chemical industries based on their streams. They quickly realised they would not like doing jobs in their fields and then relied on their network to start a company and turn to higher education to change careers respectively.
The risk appetite of people from the Ivy Leagues is higher, and they can switch careers easily, as there are opportunities to fall back on if things don’t work out. The same cannot be said for people from non-Ivy Leagues.
I remember when I was in my B-school asking the Alumni about the difference in the calibre of people from IIMs compared to people from B-schools such as ours. They shared that once you are in the system, the labels don’t matter. It is true to an extent, but the labels never stop mattering, especially when the world is driven by it.
IPL - The pressure of justifying the price tag
Indian Premier League, the T20 cricket league in India, began on March 22. It will last two months, and the fans nationwide will support their teams to win the trophy.
Sam Curran, an English all-rounder, was bought by Punjab Kings for 18.5 Crores, which is approximately $2M pounds in 2023. In the last season, he took 10 wickets and scored 276 runs. It was a below-average season, and the pundits claimed the price tag put Sam under duress, leading to underperformance.
The auction has a funny way of rewarding people regardless of their past performance. Sam had put his base price at 2 crore, but due to the demand and the battle between the teams, his price leapfrogged to 18.5 crore! He did not ask for that kind of money, but he wouldn't say no to it either.
It may have happened that he feels that he needs to justify the price tag and has to do exceedingly well in all fourteen games in the tournament. When you try too hard, things don’t work out well. Sam has the 2023 season behind him and has started 2024 on a great note scoring 63 runs in the first match. Will he be able to justify the price tag? Will he be able to keep the price out of his mind and perform? It’s easier said than done. I hope he does well and stops thinking about the price altogether, but that won’t happen, right?
Tasty Morsels from Groovy Hubs
What slows company down is horizontal friction
The danger of a yes-culture
The universe is expanding at a differential pace
That’s it for this week. See you in the inbox next week.