DeepSeek, AI and the markets

Episode Summary

Executive Summary: The episode examines why China’s DeepSeek AI model rattled markets, especially Nvidia and AI infrastructure stocks. The hosts argue the shock comes from DeepSeek appearing to deliver strong AI performance at far lower cost and with fewer advanced chips, undermining the belief that AI leadership requires massive proprietary compute spending. They debate whether this is a structural challenge to US tech dominance or just a sharp repricing.

Main Topics: DeepSeek’s market shock (Priority: 5/5): DeepSeek’s release and apparent performance triggered a major reassessment in markets, hitting Nvidia and related AI names hard as investors absorbed the implications of cheaper model development. The ‘moat’ in AI (Priority: 5/5): The hosts explain investor language around moat as a barrier to entry, arguing DeepSeek may have punctured the idea that leading AI models are protected by wide, durable moats for incumbent US firms. China’s efficiency-driven innovation (Priority: 4/5): They highlight that US chip restrictions may have forced Chinese developers to innovate under constraints, producing efficient AI systems that challenge the assumption that compute-heavy approaches are inevitable. Repricing of AI infrastructure bets (Priority: 5/5): The conversation notes that utilities, data-center-related stocks, and other infrastructure plays fell even more than Nvidia because cheaper models imply less electricity and hardware demand. Is this a dot-com-style bubble burst? (Priority: 4/5): They compare the selloff to the dot-com era but conclude this is different because the dominant AI firms are highly profitable, established businesses rather than empty speculative shells. Portfolio and macro implications (Priority: 4/5): The hosts discuss whether investors should reduce US exposure or hold more cash amid expensive US equities, uncertain rates, and a potential correction, while rejecting panic as a strategy. Long/short segment (Priority: 1/5): The lighter close covers Brazilian jiu-jitsu as a personal/ cultural trend and expresses disdain for winter and dry January.

Key Arguments: DeepSeek matters because it challenges the market’s assumption that competitive AI requires vast spending on chips, data centers, and energy. The release weakens the belief that AI is a winner-take-all industry dominated by proprietary US models. US export controls on advanced chips may have unintentionally encouraged Chinese efficiency and innovation. Nvidia’s selloff is large but not fatal; it remains an extremely valuable company even after the drop. Utilities and data-center infrastructure stocks may be more exposed than chipmakers if AI compute demand proves less intensive than expected. This is not the same as the dot-com crash because the key AI firms are profitable, high-quality businesses rather than unprofitable hype stocks. Investors may want to think about diversification away from expensive US large-cap stocks, though timing a correction is nearly impossible. The episode sees the market reaction as a challenge to US tech exceptionalism, not necessarily a financial crisis.

Data Points: Nvidia market value loss: about $600 billion - Estimated hit to Nvidia’s market capitalization after DeepSeek’s emergence triggered a selloff. Nvidia stock move: 16% down - Tuesday decline discussed as the initial repricing of Nvidia shares. DeepSeek build cost: $5.6 million - Claimed cost to build DeepSeek, contrasted with far larger US AI spending. OpenAI build cost reference: $110 bazillion dollars - Hyperbolic contrast used by the hosts to describe perceived US AI spending intensity. Data center utility stock performance: up 40% to 200% last year - Mentioned as the prior surge in utilities tied to data-center growth before DeepSeek’s efficiency implications. Magnitude of AI/stock valuations: around 50 times earnings for Costco - Used to illustrate how expensive large US stocks are broadly, not just AI names. Nvidia quarterly earnings reference: $30-odd billion a quarter - Described to emphasize Nvidia’s existing profitability despite the selloff. Tariff scenario example: 25% tariffs on Canadian exports - Used in an example of DeepSeek producing a credible economic answer.

Pivotal Quotes: "there goes the moat" — Rob Armstrong: Investor shorthand for the idea that DeepSeek may have weakened barriers protecting US AI leaders. "legit invigorating" — Sam Altman: How the OpenAI chief described DeepSeek’s emergence, interpreted by the hosts as competitive pressure. "this thing really works" — Katie Martin / discussion: Reaction to the realization that DeepSeek was not just hype but produced a genuinely capable model.

Implications: DeepSeek suggests AI may be cheaper and less centralized than assumed, pressuring US tech, chipmakers, utilities, and data-center bets. It likely accelerates competition and narrows moats, but does not by itself signal a financial crisis.