Baidu's Q1 2026 Is Two Companies in One Filing

Baidu's Q1 2026 Is Two Companies in One Filing

Robin Li's own framing in Baidu's Q1 2026 earnings release, issued to the market on 20 May, was that "AI-powered business" had crossed a majority of the company's core-business revenue for the first time. In the 6-K filing Baidu handed the SEC the same day, the number underneath the framing was blunter. Online advertising revenue fell to RMB 12.6 billion from RMB 16 billion a year earlier, a drop of roughly 21%. AI-powered business revenue crossed RMB 13.6 billion, up 49%. For the first time in the company's history the AI line came in bigger than the ad line, and the ad line was not merely slowing. It was contracting.

Two businesses, going in opposite directions, inside the same filing. That is the sentence any Chinese ad-tech investor should have taped to their monitor this quarter.

What is being cannibalised

The convenient reading is that Baidu is a company-specific story. Weak execution. Late to short video. Losing share to Douyin and Kuaishou. Management is watching Ernie and Kunlunxin generate a new revenue base to offset a legacy ad line that was always destined to shrink under macro pressure. That reading is not wrong. Baidu's search share has been softening for years, and China's economy has stayed sluggish through 2026, pressing ad demand across the board. It does not survive the second look.

Sit with the industry data. China's digital advertising market is still growing. The Business Research Company forecasts expansion from about US$163 billion in 2025 to US$266.6 billion by 2029, a 17.8% compound annual rate. Weibo booked advertising revenue of US$369.8 million in Q1 2026, up 9% year over year, and management attributed the recovery in part to AI-powered ad-tool efficiency and stronger performance in autos, local services and consumer internet. Douyin now commands about a quarter of China's digital ad market, roughly 25.9% by industry-tracker estimates, up from around 16% two years earlier. Take that tracker estimate with due caution: ByteDance does not itemise Douyin ad revenue in any regulator-facing filing, and market-share splits here are triangulated rather than audited.

The tap is not drying up. The tap is being re-plumbed. Baidu's advertising business is not shrinking because Chinese advertisers stopped spending. It is shrinking because their RMB is moving to surfaces that do not need Baidu's index-driven summary to broker a click, and because Baidu's own AI answers are increasingly resolving user queries before an ad slot ever loads. Baidu is the same company on both sides of that transaction. The Chinese ad-tech industry is not.

Weibo's Q1 story is worth pausing on for a second, because it is the counterpoint to Baidu inside the same regulatory perimeter. Weibo did not have a search-driven revenue base to defend, and its Q1 recovery leaned on AI-powered targeting and creative-optimisation tools rather than on rescuing a legacy channel from the AI-answer layer. Management's own commentary read like a company retooling around a new advertising surface, not a company mourning an old one. That is the split. Ad platforms that never depended on the referral funnel are absorbing the AI shift as a productivity gain. Ad businesses that lived off the funnel are absorbing it as a permanent haircut.

Xiaohongshu is getting out ahead

The obvious analog to what US observers have watched Reddit do with AI licensing is Xiaohongshu, the recommendation-heavy content platform sometimes translated as "Little Red Book." Bloomberg reported on 23 June that Xiaohongshu is preparing a confidential Hong Kong IPO filing by month-end. The equity story reads as e-commerce and lifestyle. The more interesting story is what a Xiaohongshu float will make visible about the market value of a content corpus with the properties AI grounding actually wants: high-context, recommendation-shaped, user-authored, and not already ingested by every foundation-model lab in the field.

Xiaohongshu is not selling training rights the way Reddit did in its Google and OpenAI deals. The Chinese regulatory geometry does not favour that route, and the platform's own leadership has been publicly reticent about opening its corpus to outside labs. What it is doing — building an ad business on top of a corpus that recommendation-heavy AI answers cannot easily commoditise — reads as the same idea, arrived at from a different direction. The lever the company is pulling is the corpus itself, and the ad business sitting on top of it, not the licensing contract that would sit alongside. That is a slower monetisation path than Reddit's flat-fee-then-dynamic-pricing route, and it may prove more durable for anyone who thinks the AI-labs-as-permanent-payers story looks shakier over a five-year horizon than a five-quarter one. A Hong Kong listing prices the corpus once, at valuation, and lets the operating business grow into it. A licensing deal prices the corpus every quarter, at whatever the labs' negotiating leverage happens to be.

The regulatory geometry is not the American one

I want to be careful here, because the temptation from Amsterdam is to read the Chinese picture through US lenses. There is no Chegg-style antitrust route open in China. There is no analog of New York Times v. OpenAI winding through a Chinese court. The Cyberspace Administration of China's generative-AI Measures, and the content-labeling rules in force since 1 September 2025, address training-data authenticity, IP compliance and generated-content labeling. They do not address compensation frameworks for publishers whose corpora are ingested. A Chinese publisher cannot sue its way to a licensing revenue line.

That absence changes the shape of the response. In the United States the publisher plays for court leverage or a coalition-driven statutory regime, on a timetable measured in years. In China the publisher pivots the business model directly — into recommendation, into short video, into e-commerce, into paid membership — because the courtroom-lever option is not on the shelf. Weibo's ad-tech reinvestment reads as that pivot in visible form; so does Xiaohongshu's IPO preparation. Neither is waiting on a legislator or a judge. Both are trying to build a value proposition that survives the AI-answer layer on their own timetable, and both look, to me, like better bets on a five-year horizon than the US publisher coalition currently waiting for either a Chegg victory or a statutory-licensing bill to move.

Where I part company with the Baidu bull case

The bull case on Baidu runs, in the version most sell-side analysts have published, roughly: AI Cloud compounds; the Ernie 5 series monetises the AI-search transition; the ad line stabilises around a smaller base once the mix shift works through; the stock is undervalued relative to the AI franchise being built inside it. On the numbers that case has some traction — AI Cloud revenue grew 79% in Q1, per the same Baidu earnings release, and management is guiding to further mix shift as Ernie 5 monetisation ramps.

Where I part company is with the stabilisation assumption. Barchart's read of the quarter, which I take to be the honest one, puts the position plainly: this is advertising revenue that is not coming back to trend. Search-driven ad monetisation lives on a referral funnel. Once the AI answer resolves the query at the top of the results page, the ad slot is no longer competing with a link for the user's attention. It is competing with a summary that has already answered the user. That change is not a cyclical bottom. It is a permanent recalibration of what a search-driven ad slot can charge for each impression it does still serve, because the marginal impression is now shown to a user who has already been told what they asked to know. CPMs, in that world, do not rebound to where they were before the answer was written. They reset lower, and the base they reset from is the new ceiling for the business.

That is my stake for this post. The AI Cloud line becomes a real business for Baidu. The ad line does not stabilise at a lower base; it structurally undershoots its pre-transition trend for years. The Baidu that emerges is a smaller, more concentrated, faster-growing AI-infrastructure company. It does not turn back into the general-search advertising franchise of the early 2010s in a new suit. Investors who buy the stock for the AI-Cloud franchise are buying something real. Investors who buy it in the expectation that the ad line finds a floor and recovers are buying a story the top-line is refusing to tell.

A last image

Xujiahui, the old commercial district in Shanghai I try to walk through on any China trip. The elevated pedestrian rings around Xujiahui Metro station used to be lined with printed advertising — property listings, mobile-phone tariffs, cram-school entrance-exam league tables. Ten years ago you could look up on almost any afternoon and count two dozen printed banners layered onto the concrete pillars.

Last spring, walking through the same rings, most of the banners were gone. Where they had been, the pillars carried QR codes and short handles pointing at recommendation-feed accounts. The physical infrastructure of the Chinese ad-supported publisher used to be visible from a walkway. Now it lives on someone else's feed, brokered by someone else's algorithm, and priced against someone else's engagement number. Baidu's Q1 filing is that walkway, drawn as a P&L.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan, an Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.