The Last Shovel

In the second quarter of 2026, Baidu’s online marketing revenue was 13.1 billion yuan, down 19% from a year earlier. That was the fourth straight quarter of double-digit decline. For the first time, the business brought in less than half of Baidu’s core revenue. In the same earnings report, AI revenue accounted for 50% of the total, and GPU cloud revenue grew 283%.

The two numbers sit in the same report without touching.

One quarter earlier, in Q1 2026, search and feed advertising revenue fell 29%. Baidu App’s monthly active users dropped from 735 million a year earlier to 644 million in June 2026. That is roughly 90 million fewer users on a comparable basis. StatCounter puts Baidu’s share of China’s search market at about 44.6%, down from a peak of 87%. On PC, Bing passed it with more than 50%.

Fitch downgraded Baidu in August 2026 and used the phrase “structural decline.” Its reason: AI competition “is eroding the monetization capacity of Baidu’s search business.” The phrase matters. This is not one bad quarter for ad budgets. The place where people get information has moved.

What did DeepSeek do to that curve?

By user count alone, DeepSeek did not knock out Baidu. QuestMobile put DeepSeek at about 163 million monthly users in June 2026. Baidu App had 644 million. Add Doubao, Tongyi Qianwen and every other AI native app, and the total is still under 500 million. That is not enough to empty Baidu one user at a time. If impact means “uninstall Baidu, install DeepSeek,” the impact is small.

Users do not change behavior by moving house.

A commuter asks DeepSeek which subway exit to take. Gets an answer. Closes the chat. At the office, the same person opens Baidu to find an official website link. Both habits survive. But every time the first one happens, the second becomes less necessary. The loss is not replacement. It is high-frequency, low-effort queries being ladled out of the search traffic pool.

DeepSeek also does not need to win as a standalone app. It is being tested inside WeChat Search. Users in the grayscale test can call DeepSeek-R1 from the search box at the top of a WeChat chat window without switching apps. DeepSeek stops depending on people remembering to open an AI app. It sits inside a place they already open all day. Once search finishes inside WeChat, Baidu is no longer competing with one startup. It is competing with a super-app ecosystem.

If the damage stopped at “people search less,” Baidu could raise ad prices. AI search hits something lower: the click.

Search auction ads run on clicks. A user types a keyword, sees links, clicks one, and an advertiser pays. Baidu’s revenue rises when users leave the search page. More clicks mean more inventory and hotter bidding.

AI search runs the other way. When an AI summary takes the first screen, the user gets an answer, not a list of links to test. More than 70% of searches now produce no click. On mobile, the zero-click rate is above 75%. For informational searches, it runs between 77% and 85%. When an AI summary appears, paid search click-through falls from 21.27% to 9.87%, a 53.6% drop. Fitch says AI summaries “may continue to erode click-through rates.”

The ad slots on the first screen of Baidu results are becoming a road sign people drive past.

Baidu has to embed AI summaries to keep users. The better those summaries work, the less reason users have to click. Fewer clicks mean less ad inventory and weaker pricing. If Baidu holds back on AI, users leave for other AI entry points. If it pushes AI, its own answers eat its ad clicks. One analysis described the result: Baidu was “timid” in its redesigns, and the product ended up “neither fish nor fowl.”

So did Baidu play itself to death?

Inside the company, the judgment holds. Baidu’s core revenue depends on people clicking ads. Its users want answers without clicking ads. That tension was always going to break somewhere. AI search made it break faster. In the old page, ads could hide among links because users still had to sort results. Now the answer arrives first, and the sorting action that ad slots depended on disappears. Baidu saw the shift. It also depended on the cash cow. It could not decide whether to weaken its most profitable product. It did not fall behind only in the AI era. It has been unable to answer that question for years.

Outside pressure still matters. DeepSeek and other AI search tools gave dissatisfied users somewhere to go. WeChat made that exit easier to reach. Without AI search, the dissatisfaction would still be there, but the substitute would be worse. Migration would cost more effort. DeepSeek lowered the effort. Tencent Yuanbao and similar products are turning “ask AI inside a super app” into a habit. Every habit formed there makes Baidu as a separate search destination less necessary.

“Played itself to death” is right but incomplete. Baidu dug the grave. DeepSeek handed over the shovel.

Measured by “did DeepSeek users come straight from Baidu,” the impact is small. Google offers a comparison. ChatGPT overlaps with 95% of Google’s users, and most large language model users have not left Google. They use AI as a supplement. AI search has not killed traditional search in any major market.

Not killing is not the same as not hurting. The hurt is in the split. Light, frequent queries go to chat boxes. Heavy, complex queries stay with search. Baidu gets squeezed at both ends. DeepSeek and WeChat AI search take the light end. Xiaohongshu, Douyin in-app search and vertical platforms take the heavy end. What remains for Baidu is a narrower middle: people still need search, but they no longer need Baidu as the default place to start.

Baidu’s Q2 report already shows the company has stopped treating traditional search ads as its growth story. AI revenue reached 50%. GPU cloud grew 283%. Online marketing fell 19%. The old engine is stalling. The new one has not fully caught. DeepSeek cannot steer that picture alone. It has changed what “search” means inside it.

Baidu faces a migration that must be endured. It cannot win back the old race. The size of the hit depends on whether the new revenue base can carry the company before the old one loosens further. As of autumn 2026, the answer sits between two numbers: 50% on one side, 19% on the other.