BG2 with Bill Gurley & Brad Gerstner | NVDA, Chips, AI Compute Build Out, Impact of AI on Big Tech, Earnings & Macro Set-up | E03

Episode Summary

Executive Summary: The conversation argues that AI is a major phase shift across consumer, enterprise, and infrastructure markets, with winners and losers already emerging. The hosts focus on how memory and actions could make AI assistants transformative, why Google’s search monopoly may face an innovator’s dilemma, why Meta and Microsoft look advantaged, and why NVIDIA’s long-term demand may still be underestimated despite near-term uncertainty.

Main Topics: AI as a phase shift and the productivity thesis (Priority: 5/5): The hosts frame AI as a durable platform shift that will drive major productivity gains in coding, customer service, self-driving, content creation, and enterprise workflows. Google’s innovator’s dilemma (Priority: 5/5): They argue Google’s search business faces structural pressure from AI answer engines that raise serving costs, reduce ad monetization, and weaken margins over time. Memory and actions as next AI breakthroughs (Priority: 5/5): The discussion highlights memory (longitudinal context) and actions (doing tasks like booking) as the two features that could make AI assistants truly indispensable. Big tech winner/loser framework (Priority: 4/5): They evaluate Microsoft, Apple, Meta, Amazon, and Google by asking whether AI strengthens core businesses and opens new ones, concluding some are clear winners while others face mixed outcomes. NVIDIA, compute demand, and the AI infrastructure buildout (Priority: 5/5): They debate whether NVIDIA’s growth is being underestimated, emphasizing that AI compute demand, sovereign demand, and power constraints may drive a multi-trillion-dollar infrastructure cycle. Chips, fabs, TSMC, and the difficulty of reshoring (Priority: 4/5): The hosts are skeptical that new entrants, sovereign funds, or U.S. policy can quickly replicate Taiwan’s semiconductor manufacturing advantage or compete with TSMC and NVIDIA. Business models and monetization in AI (Priority: 3/5): They discuss subscriptions vs ads for consumer AI and whether AI services will be priced at cost, with strategic implications for disruption and user adoption.

Key Arguments: AI adoption is already visible in products like Tesla Full Self-Driving, Microsoft Copilot, and Meta’s ad targeting, showing that the ‘juice’ can justify the cost when productivity gains are real. Google’s core search economics are vulnerable because AI answers are far more expensive to serve than blue links and likely reduce ad clicks and margin. Google’s search business is already slowing in engagement growth, and its high-margin monopoly is more exposed to disruption than the market assumes. Memory is essential because a truly useful assistant must remember personal history, preferences, contacts, and prior tasks over time; without it, AI remains limited. Actions are the other key leap: moving from answers to doing things like booking travel or placing orders would massively increase AI utility and switching costs. Apple is positioned to win if it can layer an AI assistant on top of the iPhone using personal context and on-device models, even if it outsources deeper reasoning to other models. Meta has already benefited from AI in core ad targeting and engagement, and its AI-enabled products and consumer hardware could create additional businesses. Amazon’s retail and AWS businesses are helped by AI through better targeting, lower costs, and sticky data gravity, though the company still lacks a clear NVIDIA-like silicon breakthrough. NVIDIA’s revenues may stay strong because the AI compute buildout is likely much larger than consensus and could extend across consumer, enterprise, and sovereign demand. Chip manufacturing is not easily disrupted: advanced fabs require specialized labor, high yields, and deep operational experience, making the idea of quickly building competitive fabs outside Taiwan unrealistic. Sovereign demand from the Middle East and elsewhere may become a major source of AI infrastructure spending, but the economics still depend on energy, TSMC capacity, and long lead times. The market may face a ‘zone of disillusionment’ when supply and demand temporarily mismatch, but the hosts expect that to be a buying opportunity rather than the end of the AI cycle.

Data Points: Google queries per day: 10 billion - Used to illustrate the scale and maturity of Google search. Google query growth: 4% annually - Estimated growth in basic engagement/query volume. Google monetization growth: 13% - Driven by more ads on the page. Combined search KAGR: 17% - Approximate blend of query growth and monetization growth. Cost per traditional search query: About a third of a penny or less - Serving 10 blue links in Google search. Cost per AI answer query: About 4 cents per 750 tokens today - Serving a ChatGPT-like answer is much more expensive. Relative AI answer cost vs blue links: Up to 50x more expensive - High-quality answer generation versus traditional search results. Google’s estimated profit contribution from search and YouTube: Over 80% of profits - Used to argue AI disruption hits the core business. Meta stock decline cited historically: Around $90/share - Referenced as the level when investors thought Facebook was dead before AI-driven recovery. NVIDIA expected growth rates (2023-2025): 42% - Consensus growth expectation cited versus peers. Microsoft expected growth rates (2023-2025): 14% - Compared in valuation framework. Amazon expected growth rates (2023-2025): 12% - Compared in valuation framework. Google expected growth rates (2023-2025): 11% - Lowest among the four large-cap peers discussed. Google valuation: 21x 2024 P/E and 18x 2025 P/E - Used to argue the market already discounts some AI risk. NVIDIA current share of global compute buildout: 55% today - From the chart discussing AI data center buildout. NVIDIA projected share of global compute buildout: 26% by 2028 - Consensus assumption embedded in valuation. Installed base of data centers: About $1 trillion - Jensen Huang quote cited for the existing global data center base. Projected global data center base over next 4-5 years: About $2 trillion - Expected shift toward accelerated compute. Meta annual spend on new AI-related businesses: $20 billion a year - Referenced as investment in AI/other businesses that may generate returns later. Sovereign financing target cited: $7 trillion - Attributed to Sam Altman’s exaggerated but influential framing of AI buildout needs. SoftBank/Masa financing target: $100 billion - Mentioned as a separate bid to fund chip/fab competition. TSMC fab churn in Texas during recession: 12% - Morris Chang’s example of operator turnover in a Texas fab. TSMC fab churn in good times: 25% - Used to explain why U.S. fabs are difficult to operate competitively. Rabbit demo: Large action model - Example of AI performing cursor-based actions like booking Uber.

Pivotal Quotes: "If you don't do macro, macro does you." — Brad: Used to justify discussing inflation, rates, and the Fed even though investors may prefer to focus on stocks. "Profit margins are probably the most mean-reverting series in finance." — Bill/Charlie Grantham quote referenced: Applied to Google’s margin risk and the idea that monopoly profits attract competition. "Would you rather have an assistant with the intelligence of like Einstein, but they have no access to the internet and they don't know anything about your history? Or would you like to have an assistant that's just above average intelligence, knows everything about you, and they can use the Internet? Okay? You would choose that." — Brad: The central analogy for why personal memory plus context may matter more than raw model intelligence.

Implications: AI is moving from novelty to infrastructure. Investors should watch memory, actions, and compute as the next value-creation layers, while expecting volatility, margin pressure, and major shifts in who wins search, devices, cloud, and semis.