Ep. 023 - Everyone Leaves Google, Elon Forecasts 1T ARR, Reflecting On GPT-5, Building Personalized Software (Roundtable) | Jon Y, Doug O'Laughlin, Jordan Nanos

Ep. 023 - Everyone Leaves Google, Elon Forecasts 1T ARR, Reflecting On GPT-5, Building Personalized Software (Roundtable) | Jon Y, Doug O'Laughlin, Jordan Nanos – SemiAnalysis Weekly

Market views

Bullish

NVDA— NVIDIA GPUs are viewed as superior to alternatives, with SpaceX explicitly choosing them as 'the best' and former Google employees raising billions to buy NVIDIA chips rather than using Google's TPUs.
MEMORY— Memory chip supply is severely constrained with Apple unable to negotiate discounts from CXMT, and products sitting idle in Taiwan due to lack of memory. Elon Musk expected to prioritize memory production at TeraFab due to high margins and demand.
AIMODELS— AI coding assistants have reached a capability level where individuals can build complex custom software (video editors, audio processors) for under $1000, demonstrating significant productivity gains and model improvement trajectory from GPT-5 through 5.6.
INNOLIGHT— InnoLight is assessed as technologically superior to Western competitors in 1.6T transceivers, with the entire supply chain already established in China, making them a leader in optical components for data centers.

Bearish

GOOGL— Google losing critical AI talent (Jeff Dean, Sanjay, Oriole, Kwok Lee) while Gemini hasn't been a frontier model. Company characterized as having 'L culture', poor execution, bureaucratic structure, and history of acquiring rather than building innovation internally. Compared to Bell Labs and Kodak as companies that invented transformative technology but failed to capitalize.
GCP— Google Cloud Platform has poor support for TPUs outside of Google's internal use, driving customers and former employees to choose NVIDIA GPUs instead. Lack of software ecosystem and documentation makes TPUs uncompetitive despite hardware advantages.

Not investment advice.

Mentioned

Market views