Baidu's 1%: A Growth Story Standing on One Doubling Business
The headline number is a rounding error, and that is precisely the story. Baidu General Business revenue rose 1% year over year to RMB 25.0 billion — the first positive print after a long stretch of investors watching it shrink. “In Q3, Baidu General Business returned to year-over-year revenue growth, increasing 1% to RMB 25.0 billion” — Yanhong Li, Co-Founder and CEO · 2026-11-16. Every bit of that 1%, and more, came from one line: AI cloud, where Infra revenue grew 94% year over year, accelerating sharply from 50% in the prior quarter.
The acceleration is not a rounding artifact. GPU Cloud revenue grew 191% year over year and about 43% sequentially; Qianfan's model services saw external-customer token revenue grow more than 12-fold year over year. Dou Shen, who runs the cloud group, framed it as durability rather than a spike: “This quarter, AI Cloud Infra revenue grew 94% year-over-year, once again outpacing the market.” — Dou Shen, Management · 2026-11-16 Notably, this is one of the rare cases where a management promise got kept — in August he told investors he was “confident that AI Cloud Infra can maintain strong growth in the second half with the potential for further acceleration” — Dou Shen, AI Cloud Infra Executive · 2026-08-18, and the number then accelerated by roughly 44 points.
The counterweight is unmissable. Total revenue was still RMB 30.8 billion, down 1% year over year. iQIYI fell 10%. Henry He, the CFO, did not dress it up: “our advertising business will remain under considerable pressure from macro conditions, intense competition and our ongoing AI search transformation” — Haijian He, Management · 2026-11-16. So the company has effectively swapped one growth engine for another in public, and the second engine is now large enough to mask, but not yet to overwhelm, the decline of the first. AI-powered business reached RMB 13.2 billion, or 53% of General Business revenue — the AI Power business milestone that management had flagged for quarters now sits above half and is no longer the talking point it once was.
The capex inversion
Here is the genuinely unusual detail, and it cuts against the global hyperscaler script. Baidu's capital spending fell roughly RMB 4 billion sequentially while cloud growth nearly doubled: “the CapEx in Q3 was RMB 4 billion lower compared with Q2, while our cloud business still maintained very strong momentum” — Haijian He, Management · 2026-11-16. Management's explanation is that it pre-built compute capacity and is now sweating it harder — plus, increasingly, that customers pay up front. New GPU Cloud contracts are carrying more prepayments, and the three largest GPU Cloud customers by contract value all renewed into multiyear terms, one of them at six times its previous term length.
If that holds, it is a structurally different funding model from the debt-and-depreciation treadmill the US cloud complex runs on: shorter payback periods, faster cash recycling, less balance-sheet strain. The evidence is directionally supportive — operating cash flow of RMB 2.8 billion marked a fifth consecutive positive quarter, and the company sits on RMB 276.6 billion of cash and investments, with roughly RMB 133 billion of that net cash. But cash flow is not the same as earnings: cost of revenues rose 9% year over year to RMB 19.9 billion on cloud and traffic-acquisition costs, leaving operating income of just RMB 626 million and a 2% operating margin.
The margin low point that keeps receding
Which brings us to the sharpest tension in the transcript. One year ago this month, Henry He told investors the trough had been reached: “we believe Q3 represents a low point for margins” — Haijian He, Unknown · 2025-11-18. A year later, the message is that the trough is still ahead:
Our profit and margin will remain under pressure in the near term in the midst of continued softness in advertising. We will keep investing in AI, including cloud infrastructure, in-house chips and foundation models. These investments are critical to our long-term competitiveness, but they will limit our near-term profit improvement.
The mechanism for eventual improvement is coherent — mix shift toward higher-margin GPU Cloud, Kunlunxin and MaaS, better utilization, inference optimization, and the full-stack cost advantage — and AI Cloud Infra did grow both profit and margin year over year. But the investor should notice that the trough keeps being re-dated one quarter forward. That is what happens when your fastest-growing segment is also your most capital-hungry, and when the declining segment is the one that used to fund everything.
What stopped being said
The most instructive part of this call is the silence. A year ago Julius Luo walked investors through AI search commercialization with actual numbers: roughly 70% of mobile search result pages carrying AI-generated or multimodal content, an e-commerce MCP module peaking near RMB 6 million in daily GMV over Double 11, advertiser agents generating over RMB 25 million in daily revenue. On this call, that thread is essentially gone. The AI transformation of search has slid out of the keyword set, and what remains is aspirational: monetization "at an early stage," user experience prioritized over near-term revenue. When the hard datapoints vanish while the strategic commitment stays, the honest read is that the monetization curve is not bending yet.
Kunlunxin is the second quiet spot. For a third consecutive quarter the answer to the IPO question is the same sentence in different clothes: “We continue to move forward with Kunlunxin's listing process, and we'll update the market as appropriate.” — Yanhong Li, Co-Founder and CEO · 2026-11-16 Compare the May version — “The listing process for Kunlunxin is still ongoing, and we will update the market as soon as we have more to share” — Dou Shen, Executive or Senior Management · 2026-05-18 — and the pattern is a process that is perpetually "progressing." The product roadmap is real (P800 shipping, M100 for large-scale inference nearing volume production, M300 for ultra-large multimodal models next year), but the value-unlock event keeps slipping, and the Kunlunxin narrative has now absorbed three quarters of hope without a date.
Apollo Go shows the same pattern of deliberate narrowing. In February Robin Li said: “We will add more cars in our existing cities. We will expand to more cities.” — Yanhong Li, Co-Founder and CEO · 2026-02-26 Now the language is consolidation — “we are entering a new phase of development with our efforts concentrated on cities with the strongest demand and commercial potential” — Yanhong Li, Co-Founder and CEO · 2026-11-16. The footprint is 30 cities, Shenzhen fully driverless rides rose about 170% sequentially and the Dubai fleet doubled, and driverless testing continues in London and on Hong Kong's Airport Island. But cumulative ride counts and unit-economics-breakeven claims — staples of earlier calls — were absent. Depth over breadth is defensible; it is also a softer promise than the one made nine months ago.
Bottom line
Baidu has produced a genuine inflection in one business and a genuine turn in its cash discipline, and the market's tape is not paying for it. The global AI cloud services cluster — IREN, CRWV, APLD, and Baidu itself — has been a broad decliner across 30-, 90- and 360-day windows, meaning the accelerating fundamentals are landing inside a de-rating complex. That mismatch is the setup worth watching. The bull case needs three things to hold: cloud growth that stays north of 50%, capex that stays flat as prepayments rise, and an advertising decline that stops widening. The first has been delivered twice in a row; the second is one quarter old; the third is still the open wound. And after three years of promises about margin troughs and listing processes, credibility on timing — not on technology — is what Baidu is actually trading on now.