Are LLMs Hitting A Wall, Microsoft & Alphabet Save The Market, TikTok 'Ban'

Episode Summary

Executive Summary: The episode centers on a major earnings week in which Meta, Microsoft, and Alphabet reshaped market sentiment while revealing a tension in AI: model capability is advancing fast, but costs for data, compute, and energy are soaring. The hosts debate whether LLMs are nearing a scaling wall, whether investors will tolerate years of heavy AI spending, and whether Google’s dividend and strong results signal a new phase for the company. They also examine the TikTok divest-or-ban law, arguing that the real battle now shifts to Beijing, the courts, advertisers, and users.

Main Topics: LLM scaling limits and rising AI costs (Priority: 5/5): The conversation opens with concerns that large language models are becoming prohibitively expensive to train and run, requiring vastly more data, compute, and energy. The speakers debate whether the industry is approaching a wall or whether efficiency gains in chips, training, synthetic data, and power infrastructure can keep progress going. Meta’s earnings, AI spending, and investor reaction (Priority: 5/5): Meta’s strong revenue growth was overshadowed by Mark Zuckerberg’s warning that AI investment will remain heavy and that returns will take years. The hosts discuss why the stock sold off sharply despite strong fundamentals and frame Meta as a founder-led company willing to spend aggressively for long-term AI leadership. Microsoft’s AI monetization signals (Priority: 4/5): Microsoft’s earnings and Azure growth suggested that AI is beginning to contribute meaningfully, but the reported numbers still felt vague. The hosts note that investors seem willing to trust Microsoft’s positioning and OpenAI partnership even though the revenue impact of Copilot and AI services remains hard to isolate. Alphabet’s rebound, dividend, and search resilience (Priority: 5/5): Alphabet’s strong quarter and new dividend helped revive confidence in Google. The discussion highlights that search has not yet been materially disrupted by AI chatbots, that AI Overviews may be increasing search usage, and that the company is presenting itself as both a growth and shareholder-return story. Market context: rates, stagflation fears, and safety trades (Priority: 4/5): The episode repeatedly ties tech valuation to macro conditions, especially higher rates and fears of stagflation after softer-than-expected GDP. The speakers argue that investors may favor nearer-term earnings, dividends, gold, and cash-rich megacaps until the payoff from AI becomes clearer. TikTok divestment law and the limits of U.S. action (Priority: 4/5): The hosts discuss the rapid passage of the TikTok bill and conclude that the decisive obstacle is likely Beijing’s refusal to allow transfer of the recommendation algorithm/source code. They argue that the law may hurt TikTok by creating uncertainty, but users may keep using it regardless.

Key Arguments: LLMs are becoming more expensive to build, not cheaper; the industry may need major new breakthroughs in efficiency, energy supply, and chip performance to sustain progress. Meta’s strong ad business can support AI investment, but investors punished the stock because the returns are framed as years away and rates make near-term earnings more attractive. Microsoft’s AI story is real but still hard to quantify; 7 percentage points of Azure growth from AI services sounds meaningful but remains opaque. Alphabet’s dividend signals maturity and shareholder friendliness, suggesting the company is entering a new phase where AI and search can coexist. Google search has not yet been cannibalized by chatbots; AI Overviews may actually increase search usage rather than reduce it. TikTok’s fate is now less about Congress and more about whether China will permit a sale of the core algorithm; without that, a real divestiture may be impossible. Even if TikTok faces legal and regulatory uncertainty, user loyalty and creator behavior may keep the app alive until the final moment. Meta and Google’s scale advantages may become more important because they control distribution, cloud infrastructure, and capital spending at a level competitors cannot easily match.

Data Points: Lama 2 to Lama 3 data increase: 10x more data - Meta generative AI head Ahmed Al-Dahle said the jump from Llama 2 to Llama 3 required far more training data. Lama 2 to Lama 3 compute increase: 100x more compute - Ahmed Al-Dahle said Meta used dramatically more compute to improve its latest model. Projected LLM build cost by end of year: ~$1 billion - Dario Amodei said building a large language model could cost about a billion dollars by the end of this year. Projected LLM build cost by 2025-2026: $5-10 billion - Amodei warned model-building costs could rise into the multi-billion-dollar range next year or soon after. Meta quarterly revenue: $36 billion - The hosts cite Meta revenue rising 27% to $36 billion for the quarter. Meta revenue growth: 27% - Year-over-year revenue growth discussed as evidence of Meta’s continued ad strength. Meta prior-year quarterly revenue: $28 billion - Referenced as the comparable quarter last year. Meta stock move after hours: -18% - The market punished Meta after Zuckerberg emphasized heavy AI spending and delayed returns. Google EPS beat: $1.89 vs. $1.51 expected - Alphabet beat earnings per share estimates strongly. Azure quarterly growth: 31% - Microsoft said Azure rose 31% during the quarter. AI contribution to Azure growth: 7 percentage points - Microsoft said AI services contributed 7 points to Azure growth, up from 6 points last quarter. Alphabet dividend: $0.20 per share - Alphabet announced its first dividend as part of a more shareholder-friendly capital allocation approach. YouTube + cloud run rate: Over $100 billion annual run rate by year-end - Pichai said YouTube and cloud would exit the year above this combined run-rate level. TikTok notification campaign: Millions of users - The hosts note TikTok pushed notifications to users, prompting calls to lawmakers, though the impact is debated.

Pivotal Quotes: "the biggest constraint is going to be energy and you might need a nuclear level plant in order to keep running these models" — Alex: Discussion of the infrastructure needed to keep scaling LLMs "we're in this era where we're going to have to spend a lot on AI. It's not going to get more efficient anytime soon" — Zuckerberg (as paraphrased by hosts): Meta earnings call explanation for why the company would keep investing heavily "this lies with Beijing" — Deirdre Bosa: Assessment of the TikTok divest-or-ban bill and the practical obstacle to any forced sale

Implications: AI leaders may need far more capital, power, and patience than investors want. Expect continued market volatility, scrutiny of monetization, and more value placed on scale, dividends, and infrastructure. TikTok’s future now hinges on geopolitics as much as U.S. law.