Google's real search threat is agentic AI, not chatbots

1. Search still funds everything — ▶ 4:39

   Google Search and other generates $243 billion in revenue, growing 17% year over year and making up 55% of total revenue. The presenter estimates, without disclosed support, that it produces 70-80% of overall profit. Total company revenue is $445 billion, growing 20%. Cloud, other bets, and AI infrastructure spending all sit on top of one legacy advertising business. That concentration is why any structural change to search economics matters more for Google than AI disruption does for other big tech firms.

2. Network business as canary — ▶ 7:20

   Google Network, which places ads on sites Google does not own, contracted 2% over the last 12 months. The presenter attributes this to zero-click searches, which he says run around 65% by some estimates, driven by AI Overviews and LLM usage. Google only earns when someone clicks through to a site. Because it must pay publishers for that inventory, this is a lower-margin business and matters little to profit. He treats it as a signal of structural pressure on the open web.

3. Platform layer creates stickiness — ▶ 12:40

   Cloud generated $78 billion over 12 months, grew over 80% last quarter, and has a $100 billion ARR and a mid-30s operating margin. The presenter argues the durable value lies in the platform layer, such as BigQuery and developer tools integrated into customers' code, which raises switching costs and supports pricing power. Neoclouds like CoreWeave, which grew out of GPU-first crypto mining, mostly sell raw compute. Google says most new contracts are for AI intelligence, not raw compute, which is a better sign for retention.

4. Distribution, not just product — ▶ 23:20

   Google beat earlier search engines largely through distribution. It gave away Chrome with Google Search as the default, gave away Android on the condition that its apps were preloaded, and paid Apple a 36% revenue share of search revenue, a figure that surfaced in the antitrust trial. The presenter stresses that this playbook matters now. Google is again pushing AI Mode, Gemini, and Chrome features across every surface it owns, to retain the consumer interface before a rival takes it.

5. Brand bidding as a toll — ▶ 30:12

   Advertisers must bid on their own brand names, such as Nike, or competitors like Adidas win a meaningful share of those clicks. Citing a conversation with ad-tech expert Eric Seufert, the presenter notes that lift studies justify the spend but never compare against a world with no sponsored ads. Studies he has seen put such brand-driven spend at roughly 20-35% of Google revenue, though the true share is unknown. If interfaces change and users go direct, this toll-like revenue may be less defensible.

6. Informational queries left first — ▶ 33:33

   Despite fewer clicks, paid clicks rose 6% in 2025 and 13% last quarter, with price per click up 3%. The presenter's explanation is that the searches leaving for ChatGPT, Claude, Grok, or Gemini are mostly informational, while commercial queries still start at Google. He stresses this holds only "as of now." Google reports over 1 billion people used AI Mode in the past month, 2.5 billion used AI Overviews, and over 950 million are on Gemini, suggesting it is cannibalizing itself before others can.

7. AI ads likely earn less — ▶ 38:20

   The presenter thinks Google can insert ads into AI conversations, noting OpenAI's free tier already carries them, but doubts they will match search ads. Time in AI Mode accumulates compute cost, whereas a search ad is simply a link someone clicks, which he calls one of the highest-margin businesses in technology. His conclusion is that even a successful transition probably leaves a lower-margin business, a structural headwind rather than a collapse. This is his judgment, not a disclosed margin.

8. Agents threaten monetizable searches — ▶ 39:40

   The presenter calls agentic AI the underappreciated risk because it targets commercial searches directly. Meta said 25% of Muse users shopped, spending an average of $200, and users are also booking travel. A hotel search that used to surface sponsored links from Booking, Expedia, and Marriott disappears if an agent books the trip. Muse has only about 10 million users, and Google Spark, Google's agent, has no disclosed usage and little visibility. His case rests on inference and an expectation that user trust will build.

9. Capital cycles produce air pockets — ▶ 45:50

   Historically, without exception per the presenter, new-technology capital cycles end with supply overshooting demand and then price pressure. He expects an "air pocket" in AI compute: not an end to AI, but a pacing mismatch where a couple hundred billion dollars of excess data center capacity takes years to fill, causing price cuts and supply-chain contractions before a new equilibrium. He considers it unlikely that Google's roughly $200 billion in capex, and more next year, will be perfectly sized. He concedes that if any technology were different, AI might be.

10. Why Google is better insulated — ▶ 48:30

    The presenter lists reasons Google would fare better than rivals in an overbuild: TPUs may give a structural inference-cost advantage, it sells AI intelligence rather than substitutable raw compute, about half of a backlog over $500 billion should be realized within 24 months, and it can redeploy capacity to its own products such as YouTube ad targeting. Trailing incremental return on capital is about 19%. The weak spot is customer quality: Anthropic and OpenAI are multi-cloud and could move on price once capacity loosens.

11. Valuation hinges on 2030 earnings — ▶ 57:16

    At $320 per share, Google has a $3.9 trillion fully diluted market cap, $3.65 trillion in enterprise value, and trades at 31x trailing NOPAT of $117 billion. After backing out other bets losses, a $100 billion Waymo stake, and YouTube at $500 billion, the core services business implies 19-25x. Scenarios give 2030 NOPAT of $175 billion, $230 billion, and $295 billion, or 20x, 16x, and 12x today's EV. A 25x exit in the bull case implies roughly 100% upside; the base case implies about 8-9% a year.

The presenter's central claim is that the threat to Google Search has been mismeasured. Chatbots have taken some queries, and AI Overviews have pushed zero-click searches to around 65% by some estimates. Yet search revenue is growing in the mid-teens. Paid clicks rose 6% in 2025 and 13% last quarter, with average price per click up 3%. His explanation is composition. The queries leaving for ChatGPT, Claude, or Grok are mostly informational, while commercial queries, the ones that carry ads, still start at Google. The company has lost the low-value traffic first and kept the profitable traffic.

He stresses that this protection holds only as of now. Agentic AI has only just reached mass consumers, and his key evidence is Meta's Muse. On Meta's earnings call, 25% of Muse users shopped, spending an average of $200, and users are also booking travel. If an agent books the hotel, the sponsored listings from Booking, Expedia, and Marriott never appear. The same logic applies to insurance comparison. Google's own agent, Spark, has no disclosed usage, and he doubts most people know it exists. The argument is plausible but thinly supported: one product with roughly 10 million users, plus his expectation that consumers will come to trust agents.

Google's response is rooted in how it won the first time. The presenter argues that Google beat earlier engines less on product quality than on distribution. It gave away Chrome with search as the default, gave away Android on the condition that its apps were preloaded, and paid Apple a 36% revenue share. Today it is applying the same logic to AI. Over 1 billion people used AI Mode in the past month, 2.5 billion used AI Overviews, and over 950 million are on Gemini. The strategy is to cannibalize search before someone else does, then introduce ads once Google owns the interface.

The problem is that a successful transition probably still produces a worse business. The presenter believes ads can go into AI conversations, since OpenAI's free tier already has them, and that someone has to pay the compute bills. But each minute in AI Mode consumes compute, while a search ad is a link someone clicks. He calls traditional search ads one of the highest-margin businesses in technology and finds it hard to believe Google emerges with something equally good. This is judgment, not disclosed data. His conclusion is a structural headwind that Google can probably grow through but should not be assumed to.

Part of today's search revenue may also be less defensible than it looks. Advertisers feel forced to bid on their own brand names because a competitor like Adidas would otherwise win clicks on searches for Nike. Drawing on a conversation with Eric Seufert, the presenter notes that lift studies justify this spend but never compare it with a world of no sponsored ads. Research he has seen puts such spend at roughly 20-35% of Google revenue, but the real share is unknown. He compares it to a toll. If interfaces shift and users go direct to sites, that toll weakens.

Cloud is the second bet, and the numbers are strong. Trailing revenue is $78 billion, last quarter's growth topped 80%, ARR is $100 billion, and the operating margin is in the mid-30s, with operating income up 200% year over year. The contract backlog exceeds $500 billion, with about half expected to be realized within 24 months. The presenter's view of quality is more specific than the headline growth. Customers who build on Google's platform layer, such as BigQuery and developer tools, face real switching costs. Customers buying raw GPUs are price sensitive and can leave. Google says most new contracts are for AI intelligence, which favors the stickier kind.

His concern is the capital cycle, not AI itself. Historically, he says without exception, innovation attracts capital, early returns look good, more capital follows, and supply overshoots demand. He calls the likely result an "air pocket": a timing mismatch where a couple hundred billion dollars of excess capacity takes years to fill, prices fall, and capex plans are pulled back across the supply chain. The market has also changed from a duopoly to a field that could include Meta, SpaceX, Oracle, and the neoclouds. His case relies on historical analogy rather than utilization data, and he concedes that if any technology were different, it might be AI.

He also explains why Google keeps building anyway. Underbuilding while AI turns out to be enormous would cost far more than a few years of sagging margins from overbuilding. A reasonable editorial inference is that if every large player reasons this way, the aggregate result is the overshoot he fears. Google is better insulated than most. TPUs could lower its inference costs, it sells AI intelligence rather than raw compute, and it can shift spare capacity to its own products. Trailing incremental return on capital is about 19%. But shifting capacity toward AI Mode runs into his own point that AI Mode earns less than search.

The valuation work shows how much depends on assumptions. At $320 a share, Google carries a $3.9 trillion fully diluted market cap and $3.65 trillion of enterprise value, or 31x trailing NOPAT of $117 billion. The presenter backs out other bets losses, a $100 billion Waymo stake, and YouTube, which he values at $500 billion on estimated revenue of $65 billion and a 30% margin. He values Cloud at $1.3-2.1 trillion. That leaves the core services business implied at 19-25x. He treats the choice within that range as the real debate: a secular grower deserves 25-30x, while a melting ice cube deserves perhaps 15x.

His scenario analysis frames the same debate in 2030 terms. Projected NOPAT of $175 billion, $230 billion, and $295 billion put today's enterprise value at 20x, 16x, and 12x. A 25x exit in the bull case implies roughly 100% upside, while the base case implies about 8-9% a year. The bear case still assumes 25% cloud growth and 5% growth in the rest of Google, with flat services margins. None of the scenarios models a collapse in monetizable search, the risk he rates most important. If that risk is real, the range may be too generous. Whether Google's agent arrives before rivals' agents capture shopping and travel intent is not answered by anything in the model.