scripod.com

Dylan Patel - Inside the Trillion-Dollar AI Buildout - [Invest Like the Best, EP.442]

Shownote

My guest today is Dylan Patel. Dylan is the founder and CEO of SemiAnalysis. At SemiAnalysis Dylan tracks the semiconductor supply chain and AI infrastructure buildout with unmatched granularity—literally watching data centers get built through satellite i...

Highlights

The rapid expansion of artificial intelligence is not just a technological shift—it's an industrial and geopolitical transformation unfolding in real time. At the center of this revolution is the massive infrastructure required to power ever-larger AI models, from data centers to semiconductors, all underpinned by global competition for talent, compute, and energy.
07:23
OpenAI is in a race against trillion-dollar companies for compute dominance.
10:36
Larger models like GPT-4.5 were smarter but couldn't be served cost-effectively
13:52
An infinite number of senior engineers with low operational costs could replace $2 trillion in global software wages.
14:12
NVIDIA captures ~$35B of $50B data center CapEx with 75% gross margin through OpenAI deal
20:10
OpenAI kept GPT-5 the same size as GPT-4.0 to ensure it could serve all users.
24:38
Reinforcement learning via custom environments is key to advancing AI beyond internet data.
40:09
Childhood memories can be fake, being a series of remembered and invented pictures
40:36
DeepResearch demonstrates breakthroughs in information compression and recall for long-context modeling
48:03
People are paid $100 million to run experiments on $100 billion-worth chips, with significant wasted compute
52:49
ML research and semiconductor manufacturing both involve massive search spaces requiring expert intuition.
1:00:53
NVIDIA benefits most when venture-funded startups spend heavily on compute.
1:17:55
OpenAI's two-gigawatt data center marks a turning point in AI infrastructure scale
1:20:56
Electrician shortage is a major bottleneck for data center construction
1:23:09
Without the AI boom, the U.S. might lose its global hegemony by the end of the decade.
1:35:35
A more efficient battery chemistry could bring great economic value and advance wearable technology.
1:46:06
xAI may have a large data center but lacks compute compared to big companies.
1:57:31
The kindest thing anyone's done for me was my brother always pulling me back on track
1:58:13
I'm bad at task-orientation and sometimes inconsiderate, which makes me feel like an asshole

Chapters

Welcome to Invest Like the Best
00:00
The AI Infrastructure Buildout
05:12
Scaling AI Models and Compute Needs
08:25
Reinforcement Learning and AI Training
11:44
The Future of AI and Compute
14:07
AI in Practical Applications
17:47
The Importance of Data and Environments in AI Training
22:29
Human Analogies in AI Development
29:45
The Challenge of Infinite Context in AI Models
40:34
The Bullish and Bearish Perspectives on AI
44:08
The Talent Wars in AI Research
48:25
The Power Dynamics in AI and Tech
56:54
The Future of AI and Its Economic Impact
1:13:29
The Gigawatt Data Center Boom
1:18:55
Supply Chain and Workforce Dynamics
1:21:12
US vs. China: AI and Power Dynamics
1:24:23
AI Startups and Innovations
1:37:16
The Changing Economics of Software
1:52:44
The Kindest Thing
1:58:12

Transcript

Patrick O'Shaughnessy: The best operators have a relentless focus on leverage, finding ways to multiply their impact rather than just working harder. But here's what I see happening in finance teams everywhere. Brilliant people getting buried in expense ma...