All In - Ep. 167 Notes

Nvidia had record earnings in Q4 2023 with $22B of revenue. Massive growth over the last few years. Stock popped and the market cap gained $270B in one day.

• Chamath: When companies are too profitable, the margins eventually get competed away. This is going to spur other companies to jump into the fray.
• Friedberg: Chamath is right so the question is: how long is Nvidia’s lead going to last? We‘ve seen this type of monopoly before with companies like Cisco and Oracle but eventually competitors came in wiped out their margins.
• Sacks: Nvidia is different then Cisco because chips are so much more complex and difficult to manufactured.
• Chamath: the revenue scale will continue for 2-3 years for Nvidia then the other players will catch up
• Chamath: Nvidia is an enabler of a new ‘economy’ like Intel was for computing and Microsoft was for software. Nvidia will remain a big player but it will enable something far bigger with lots of new players.
• Sacks: in 1999, you couldn’t even upload a profile photo to an online account because the bandwidth was so low. By 2003, the underlying infrastructure enabled it and social media followed soon thereafter. The evolution of better chips are going to continue to open doors in terms of computing capabilities.

This was a big week for Groq.

• Chamath: Invested in Groq since 2016. You got to just keep grinding through the lows. Research came out that Groq outperforms other LLMs and thousands of new customers are flocking to it.
• Friedberg: This is an overnight success that took 8 years. “Deep tech” is like that. They are very difficult businesses where you need to execute everything right for a very long to have A viable product. when they work out, (like Tesla, OpenAI, SpaceX, etc) they unlock massive value in the market with a deep moat
• Chamath: you have to keep going even when things are looking bad. The current AI apps are too expensive and not good enough to really be value creators but Groq changes that
• Chamath: prefers investments that are technical challenges, not physical challenges like fusion

Gemini went off the rails this week and refused to generate images of white people.

• Sacks: It’s a joke
• Friedberg: It’s ironic that Google went from being slow to launch to prematurely launching this product and falling flat on it’s face
• Chamath: the base order principle for AI should be that it is accurate and right.
• Sacks: Gemini reflects the biases of the people who created it
• Friedberg: Google is going to have a hard time going from a data retrieval company to being a data interpretation service. rather than just showing you data created by other people, AI models need to interpret that data and communicate it in some way to the end user
• Chamath: an element of making the model good is reinforcement training which means that Gemini was trained by people who taught it not to tell the truth
• Sacks: there is an objective truth on things like ‘was George Washington white’ and AI needs to get it right
• Friedberg: think about search in 1996 vs 2000 vs now. We are in the 1996 phase of AI. It’s going to get significantly better.

Updates on the war:

• sacks: Russia just took a city in Ukraine. Russia is in the process of annexing a part of Moldova. The war could expand if The West interprets it as an escalation of the war.