Jensen Huang
TPU competition, selling chips to China, & Nvidia's supply chain moat
Listen / read the full episode ↗Predictions (6)
Vera Rubin ships this year, Vera Rubin Ultra next year, Feynman the year after, a new architecture every single year.
No supply-chain bottleneck (CoWoS, logic, memory, EUV) lasts longer than two or three years.
Token cost decreases by roughly an order of magnitude every single year.
TPU and Trainium growth is '100% Anthropic', a unique instance, not a trend.
Logic and EUV capacity can be scaled 2x/year, easy within two or three years once there is a demand signal.
The agentic-security future, one capable AI agent surrounded by thousands of agents keeping it safe, surely is going to happen.
Where they disagreed
Mental models (5)
Input is electrons, output is tokens, and in the middle is Nvidia.
AI is a five-layer cake, and the lowest layer is energy.
Do as much as needed, as little as possible, the reason Nvidia won't become a hyperscaler.
Don't pick winners: either let them all take care of themselves, or take care of all of them.
An abundance of energy makes up for chips; an abundance of chips makes up for energy.
Claims (5)
Blackwell is 35x to 50x more energy efficient than Hopper.
Nvidia gross margin is ~70%; Broadcom-style ASIC margins are ~65%.
Nvidia has ~$100B in purchase commitments; SemiAnalysis reports it trending toward $250B.
Anthropic's 'Mythos Preview' found a 27-year-old zero-day in OpenBSD.
China makes ~60% of the world's mainstream chips and has ~50% of the world's AI researchers.