Baidu Didn't Collapse. It's Doing Something Worse — Slowly.

On March 16, 2023, Baidu's founder Robin Li stood up to show the world China's answer to ChatGPT — and played a video. ERNIE Bot was real, it would launch to the public that August⁴, but investors had come for a live duel with the American chatbot and got a scripted clip instead. Baidu's Hong Kong shares fell 6.4% that day, wiping out roughly $3 billion in value before lunch.⁴ That moment hardened a story that has followed Baidu ever since: China's Google fumbled the future, lost search to ByteDance, and collapsed. It is a clean, satisfying narrative. It is also wrong about almost every load-bearing fact.

Here is the official obituary, struck through: Baidu collapsed — its search monopoly destroyed by short-video apps and its AI dreams exposed as vaporware. Here is what the filings actually show: full-year 2024 revenue was about $18.24 billion, down a mere 3.8% from 2023; FY2025 revenue declined a further ~3% as legacy ad weakness continued to outpace AI Cloud growth.¹¹⁰ A company losing four cents on the dollar in a bad year is not a company that collapsed. It is a company being quietly taken apart by something slower and harder to see.

A 4% dip is not a collapse — so where did the panic come from?

Trace the gap between the story and the numbers and you find it almost entirely in the stock chart, not the income statement. BIDU returned -31% in 2021, -23% in 2022, then +4% in 2023, then -29% in 2024 — and by mid-2025 it traded roughly 74.7% below its all-time high.⁷ A three-quarter wipeout looks like a collapse if you start the clock at the peak. But that peak was a pandemic tech-bubble high, the moment when every Chinese internet name was priced for permanent escape velocity. Strip out the bubble and what's left is multiple compression — the market deciding a monopoly is worth far less than it used to be — layered on top of a business that is genuinely, but gently, shrinking.

The distinction matters because operating decline and valuation decline have completely different cures. If the business were collapsing, Baidu would need a miracle. If the valuation merely re-rated against the operating reality, then the operating reality is the thing to autopsy — and the operating reality is a search-ad engine starved of demand.

The real killer was the customer, not the competitor

The popular villains are ByteDance and ChatGPT. Neither fits the evidence. In January 2024 Baidu still held 61.11% of all-platform search share in China and 75.93% of mobile search.⁶ By May those had slipped to 52.15% and 67.74% — real erosion, but the math of a leader, not a loser.⁶ And the only challenger taking measurable formal share was Bing, which peaked near 50% of desktop in late 2024.⁶ ByteDance never appears in the search-share tables, because the threat it poses isn't a better search engine. It's the disappearance of the query itself, as discovery migrates into short-video and social feeds.

But the share story, even read generously, is a sideshow. The actual mechanism is dumber and more brutal: who buys Baidu's ads. Baidu's online-marketing revenue is overwhelmingly small and medium businesses bidding for keywords — local clinics, training schools, e-commerce sellers. When China's consumer economy weakened, those SMBs cut their ad budgets first, because performance ads are the most reflexively cuttable line item a struggling business has. Note the tell: full-year 2022 online-marketing revenue fell 6% amid COVID disruption⁹, then recovered +8% in 2023 as conditions eased.³ A business killed by search-quality loss does not bounce back 8% the moment the economy reopens. A business hostage to its customers' budgets does exactly that.

The AI pivot is real. It just doesn't pay the bills yet.

Baidu's defense is to become an AI company before the ad engine fully cools, and the effort is more substantial than the ERNIE demo memory suggests. The non-online-marketing segment — chiefly AI Cloud — surged 12% in Q3 2024 even as online marketing softened, and net income that quarter rose 14% to $1.09 billion.² Apollo Go, the robotaxi arm, gave 3.1 million fully driverless rides in Q3 2025, up 212% year over year, and has passed 17 million cumulative public rides on a fleet that has logged 240 million autonomous kilometers.⁸ These are not vaporware numbers. They are the early innings of two genuine businesses.

The problem is arithmetic, not ambition. Generative-AI revenue in Q1 2024 was about $90 million — roughly 2% of the quarter's total.⁵ The ad engine it must eventually replace is many times larger and shrinking. So Baidu is running a race against itself: the new businesses are compounding fast off a tiny base, while the old one bleeds slowly off a huge one. Whether the lines cross before investors lose patience is the entire question, and right now the cross is years away.

Isn't a shrinking monopoly with a billion-dollar moonshot still a winner?

The honest counter is that this autopsy might be premature, even backwards. Baidu still throws off real profit, sits on a fortress balance sheet, and owns two of the most credible AI-era assets in China: a top-ranked cloud and a robotaxi network with millions of paid driverless rides. Trading at single-digit forward earnings, a patient buyer could argue the market has confused 'derating' with 'dying' and left a transition story on the discount rack. That case is fair, and it may even be right. But notice what it concedes: the bull thesis is no longer about search. Nobody underwrites Baidu today on the strength of its query box. The monopoly that built the company is now treated as a melting ice cube whose only job is to fund the thaw — and a business whose own believers root for its core to be replaced is not a healthy monopoly. It is a hostage being ransomed for cash.

Baidu did not fall off a cliff. It is doing something stranger and, for an incumbent, more unsettling: aging in place. The search monopoly still works, still leads, still prints cash — and is worth a quarter of what it once was because the market no longer believes it is the future, only the funding. The ERNIE demo was never the wound; it was the diagnosis going public. A company can survive a bad demo. What it cannot easily survive is the slow discovery, by everyone including itself, that its best asset is now a battery to be drained into the next one. That is not a collapse. It is the sound a monopoly makes when it stops being the story and becomes the wallet.