I was editing a meta description for a website I’d just deployed. It looked terrible on Google — not explanatory at all — and I got into a back-and-forth with my Openclaw agent about whether to request a recrawl through Search Console or just wait for the automatic one. We decided to wait.
But the conversation got me thinking. How much does it actually cost Google to do that? To crawl the entire internet. To read every website, parse every page, keep the index fresh — even for small sites like mine with single-digit visitors.
The honest answer is that nobody outside Google really knows. Alphabet spent over $52 billion on infrastructure in 2024, but that covers everything — Cloud, YouTube, AI training, not just search. What we do know is what search makes them.
$198 billion. In 2024. Just from search ads. Over half of Google’s total revenue — and that total includes Cloud, Android, YouTube, hardware, everything else. I knew search was profitable. I didn’t expect it to be this profitable.
But how do they even earn from search? If I look up a Python recursion error, it probably sends me to an AI Overview or Stack Overflow. Does Google make money on that?
No. Most searches earn nothing.
The money comes from the other kind of query. “Best laptop 2026.” “Plumber near me.” “Cheap flights to Tokyo.” Those are searches with buying intent, and advertisers bid on them in real-time auctions. Google gets paid when you click. Insurance keywords go for $50–100 per click. A single credit card signup might be worth $200 to a bank, so paying $50 for that click is rational.
Most searches earn zero. But enough of them are “I want to buy something” that the fraction that monetises adds up to $540 million a day.
Google isn’t selling search. It’s selling the gap between your question and the answer — that moment where you have intent but haven’t acted yet. That gap is the most valuable real estate on the internet.
So what happens when AI closes the gap?
This is where it got interesting for me.
A traditional Google search costs about $0.001 in compute. Index lookup, rank, ten blue links. Cheap. An AI-generated answer — the kind that now sits at the top of your results — costs roughly ten times more per query.
And AI is changing how we look for information. You read the generated answer and leave. No click. No ad impression. You got what you wanted faster — which is genuinely better for you, and genuinely worse for Google. Personally, I’ve been using Google a lot less since ChatGPT launched in late 2022, and most people I talk to say the same.
Google is stuck improving AI Overview — a product that undermines its own revenue model. And it can’t stop — because the alternative is watching you leave for ChatGPT or Perplexity.
But here’s what tripped me up
If AI search is supposedly killing Google, why does the revenue keep going up?
| Year | Search Revenue | Total Revenue | YoY Growth |
|---|---|---|---|
| 2020 | ~$104B | ~$183B | — |
| 2021 | ~$149B | ~$258B | +43% |
| 2022 | ~$162B | ~$283B | +9% |
| 2023 | ~$175B | ~$307B | +8% |
| 2024 | ~$198B | ~$350B | +13% |
Search revenue nearly doubled in four years. That doesn’t look like a business under threat.
It’s because multiple things are pushing revenue up at the same time AI is pulling it down. Advertisers keep bidding higher on the same keywords — as long as the return is positive, the auction ratchets up. Google keeps opening new ad surfaces beyond the search box — Shopping, Maps, Discover, Gmail, and now even ads inside AI Overview. Southeast Asia, India, Africa are still adding internet users — more phones, more searches, more inventory. And the high-value keywords — insurance, credit cards, lawyers — are essentially immune, because those industries will pay for leads regardless of how the user found them.
The growth is masking the erosion.
Nobody else has figured it out either
This is the part that really stuck with me. Perplexity — the AI search startup everyone talks about — made roughly $100 million in 2025. Google Search makes that in about four hours.
The problem is almost spatial. A list of ten blue links has room for ads. You can slot them between results and most people won’t mind. But a single AI-generated answer doesn’t have that room. Put an ad inside the answer and you destroy trust. Put it after and nobody sees it. Put it before — like the sponsored results at the top of a search page — and you’re training people to skip past it.
OpenAI tried. They started running ads in ChatGPT in Feb 2026 — it’s still too early to say whether that works at scale, but the fundamental tension is the same. Perplexity is experimenting with “sponsored follow-up questions.” Neither comes close to the economics that made Google the most profitable information business in history.
Where this leaves things
If there’s a collapse coming, it won’t look like a cliff. It’ll look like growth slowing — 13% to 8% to 4% to flat — two forces, one pushing up, one pulling down, and somewhere out there they cross.
Google isn’t dying. $198 billion buys a lot of runway. But every improvement to AI search makes the product better and the business model worse.
The gap between your question and the answer is where Google built its empire. AI is closing that gap. And nobody — not Google, not Perplexity, not OpenAI — has figured out what replaces it.