FSD: Are Semis Back?

In this FSD episode, titled AI Would Have Rejected Elon's Business Plan, Ram makes the case that AI is structurally incapable of giving you an edge. Multiply enough matrices and the output lands in the center of the distribution, which means AI returns the consensus view by construction. The best opportunities are non-consensus. Elon Musk is the proof: reusable rockets, driverless cars, brutal capital intensity, competing with the government. No model would have approved that plan.

The positioning is constructive into September as midterm risk gets priced in. Semiconductors have had a clean positioning reset, and he bought Micron as the American led memory name alongside interest in SK Hynix and Applied Materials. Independent power producers have started to rally on real earnings growth. China is the trade nobody mentions, where he chose Tencent over Alibaba on reward versus risk. Defense tech is the pre position, with Ukraine as Europe's first line of defense.

He also walks through the agent build out at Lumida, roughly a dozen now and a hundred expected within three months, covering podcast outreach, newsletter drafting, and pre meeting introductions. The constraint is imagination about which workflows to delegate, not capability. The back half turns philosophical: Adam Smith as moral philosopher, Schumpeter's creative destruction as the right frame for this cycle, and leadership as the one skill AI cannot supply.

(00:00) Grokbot and the Anthropic S-1 as next week's catalysts

(00:36) Semis are back: Micron, SK Hynix, Applied Materials

(01:57) Power, earnings growth, and the GRID ETF

(02:19) China is cheap: why Tencent over Alibaba

(02:53) Ten agents in three days, and a lumber mill running on AI

(03:45) Why non-farm payroll reactions reverse within days

(05:05) Adoption is further along than people think

(05:45) OpenAI's Astra model and Anthropic's next drop

(06:53) The outreach agent that does everything but the idea