I worked at Google from before the GFC till 2010.
I still regret not keeping those RSUs (or GSUs as they call them). It’s been a 10x since then.
Yet, yesterday, the stock plunged 7% after Eddy Cue, Apple’s services chief, said AI search engines will replace traditional search engines (ie. Google).
The problem with Google isn’t that its other businesses—YouTube, Cloud, Quantum, Gemini to name a few—aren’t good. On the contrary, each of those could be world-class companies on their own1. The issue here is that Google Search is just too darn good—as a massive cash-generating machine with a operating margins at scale that I have yet to see on a S&P 500 company.
Google has been trying for years to diversify and reduce its dependence on Search ads. Lately, they’ve had some success, but the company still fundamentally relies on Search as the bedrock of its profitability.
Let’s put some numbers to it: 1) In 2024, Google generated $350 billion in revenue. 2) Of that, $305 billion came from Google Services (which includes Search Ads, YouTube, Google Network, subscriptions, and other products). 3) The other major contributor was Google Cloud, with $43 billion.
Now, look at operating margins: 1) Google Services ran at ~40% operating margin. 2) Google Cloud’s margin was ~14%.
While Google never discloses specific Search Ads margins, it’s kind of known in the industry that they have operating margins of +75% (unheard of for an S&P500 company).
Google does report revenue for individual subsegments though—Search, YouTube, and Google Network—and there, you’ll see that Search alone accounts for ~65% of the Google Services revenue.
If Services as a whole has a 40% operating margin and Search represents 65% of that, running at ~75% margin, then it’s clear that YouTube, Google Network, and Subscriptions are way less profitable.
I can confirm this from the advertiser side. Take travel insurance, financial services, or some verticals on e-commerce—advertisers can pay $10+ per click on some keywords. The cost of serving that ad? Practically zero.
By contrast, YouTube is a great platform, but CPCs and CPVs are nowhere near Search profits, and YouTube has to split revenue with content creators. Same with Google Network.
Could Cloud make up the difference?
Doubtful.
The margins are far lower, and Google would need to triple AWS's revenues just to match Google Services in size. That seems unrealistic.
So, the hope shifts to Gemini. Again, not even close to search, why? 1) There are many strong contenders in the LLM race—Google doesn’t have a monopoly here like it did in Search for many, many years. 2) The cost of serving a single LLM chat response is at least an order of magnitude higher than a traditional search result2. And 3) It’s not clear whether or how Gemini will even be monetized through search ads, if at all.
That’s why I don’t think Gemini can “replace” Search. More competition, higher costs, and unclear monetization.
Which brings us back: Google’s issue isn’t that its other products are bad. It’s that Search is superlatively good at generating cash. And now, it’s under attack—by well-funded, hungry competitors going straight for its core.
One final argument I often hear in the media is that "valuable" searches—travel, finance, e-commerce—still happen on Google, not on LLM interfaces. That may be true for now. But I think we’re about to see a “gradually, then suddenly” moment.
Why? Because we haven’t yet seen mass adoption of web search inside Grok, GPT, or Perplexity.
I honestly can’t remember the last time I used Google. For planning my last trip I used Perplexity and for buying my last tablet I used Grok. Those would have been pretty valuable clicks that didn’t go to Google (btw the experience was much better that current cluttered search pages). Multiply that by millions and you get the demise of Google Search’s moat.
It really comes down to two questions: 1) How fast will adoption grow for these new interfaces? 2) How long can Google “extend and pretend” that Search is still unshaken?
From experience, I can tell you: Google has many levers to pull to squeeze more out of Search. If you check their 10-K, you’ll see two key metrics for Search Ads: Paid Clicks and Cost-Per-Click. They can influence both—by increasing ad density on SERPs, tweaking the Quality Score algorithm, making ads blend more with organic results, etc.
So I expect a few more solid quarters of Search traction—unless a sharp recession provides cover for a quicker drop—before we start seeing stabilization, then a mild decline on search ad revenue. And once that happens, earnings call questions will get harder to dodge.
And that’s when the narrative turns.
Disclaimer
This newsletter (the “Publication”) is provided solely for informational and educational purposes and does not constitute an offer, solicitation, or recommendation to buy, sell, or hold any security or other financial instrument, nor should it be interpreted as legal, tax, accounting, or investment advice. Readers should perform their own independent research and consult with qualified professionals before making any financial decisions. The information herein is derived from sources believed to be reliable but is not guaranteed to be accurate, complete, or current, and it may be subject to change without notice. Any forward-looking statements or projections are inherently uncertain and may differ materially from actual results due to various risks and uncertainties. Investing involves significant risk, including the potential loss of principal, and past performance is not indicative of future results. The author(s) may or may not hold positions in the securities discussed. Neither the author(s) nor the publisher, affiliates, directors, officers, employees, or agents shall be liable for any direct, indirect, incidental, consequential, or punitive damages arising from the use of, or reliance on, this Publication.
Well, at least YouTube and Cloud for sure as they are actually generating profits.
According to an analysis by Azeem Azhar, each interaction with ChatGPT can consume up to 2.9 watt-hours (Wh) of energy, which is nearly ten times the 0.3 Wh energy cost of a standard Google search. SemiAnalysis further estimates that if Google replaced its current search algorithms with LLMs, each search request could cost up to 8.9 Wh-potentially 30 times more than current search costs. (SemiAnalysis. (2023). The inference cost of search disruption – Large language model cost analysis. https://www.semianalysis.com/p/the-inference-cost-of-search-disruption)
A 2025 AI industry cost guide from Businessware Technologies explains that while LLM costs have dropped with new models like GPT-4o, the expense per request remains significant, especially at scale. Even with cost-optimized models, the total cost of deploying LLM-powered systems-including API usage, infrastructure, and document intelligence-remains much higher than traditional search, especially for high-volume applications.
(Businessware Technologies. (2025, April 21). What does it cost to build an AI system in 2025? A practical look at LLM pricing. https://www.businesswaretech.com/blog/what-does-it-cost-to-build-an-ai-system-in-2025-a-practical-look-at-llm-pricing)