BIDU: Buybacks set the floor, but the turnaround hinges on the 'narrative' ---

$Baidu(BIDU.US) released Q4 results after the HK close on Feb 26, 2026, and overall Q4 was mediocre. Given the prolonged correction, plus buybacks/dividends providing a floor and the Kunlun Chip IPO and potential Stock Connect inclusion as year-long narratives, the market is unlikely to impose sustained punishment.

This quarter formally adopted a new disclosure framework, split into AI and other legacy businesses, with AI covering AI infrastructure, AI apps, and AI-native marketing. As the Street is still modeling under the old framework, Dolphin Research restores the old format for comparative analysis and continues to focus on BIDU, excluding iQIYI.

In detail:

1. AI mix rose to 43%: AI-related revenue reached RMB 11.3bn in Q4, 43% of total, up from 39% in Q3. Breakdown follows.

(1) AI infrastructure (cloud, LLM API, compute leasing) accounted for half, up 35% QoQ, a solid trend. This reflects BIDU's full-stack AI from compute to LLMs offering better fit and cost-efficiency, one of the few highlights this quarter.

(2) AI apps (Baidu Wenku, Baidu Netdisk, digital employees) were flat QoQ. Netdisk likely remained a drag.

(3) AI-native marketing (Agents and digital humans) edged down QoQ. Given industry trends, competition likely weighed, and the call should clarify any seasonal or other objective factors.

2. Legacy ads still struggling: Traditional ads (search, feed ads) fell 26% YoY. That is narrower than last quarter's -30%, but low-base effects are unclear, making the timing of a real recovery to positive growth hard to judge.

Dolphin Research leans that legacy ads remain at the bottom, given macro and competitive pressures in and outside the sector. Mobile Baidu continued to lose users in Q4, with MAU sliding to 679mn. Despite seasonality, two consecutive quarters of decline make it hard to ignore peer competition, especially new-gen AI chatbots.

BIDU's AI chatbot Ernie Bot has a standalone app 'Wenxiaoyan', but penetration is still mainly via the embedded experience in the Baidu app. The company disclosed Ernie Bot now reaches 200mn MAU, about a 30% penetration rate within Mobile Baidu.

3. Autonomous driving and other: This bucket delivered RMB 2.5bn revenue in Q4, up QoQ. In H2 last year, Apollo focused on Intl expansion, piloting via partnerships with Uber and Lyft in the Middle East, the UK, and Korea.

As of now, Apollo covers 26 cities globally.

4. D&A optimization, organizational efficiency: On expenses, severance of c.RMB 700mn and bad-debt provisions weighed, so GAAP operating profit missed. Excluding these, with lower amortization/depreciation (one-off impairments on some legacy equipment were taken last quarter) and workforce optimization driving efficiency (with similar market cap, SBC fell 32% QoQ), Adj. operating profit slightly beat.

5. More 'generous' shareholder returns: Although the core cashflow engine—ads—slumped, cash reserves remain sizable, with RMB 115.3bn cash+ST investments at end-2025. Net of RMB 22.4bn ST borrowings, that leaves roughly RMB 93.0bn, or $13.5bn.

With profit improvement in Q4, quarterly FCF appeared to turn positive after a year. However, capex dropped below RMB 2.0bn this quarter, and Dolphin Research views this as non-recurring, so normalizing growth likely keeps FCF negative.

In early Feb, the company announced a two-year buyback program of up to $5bn. The first dividend plan has also been approved by the BOD, with the amount undisclosed; it is expected to be announced and implemented this year.

Ex-dividends, if buybacks are executed at the cap, overall shareholder return would rise to the 5–6% range (vs. the current $45.6bn market cap). Versus the prior ~2% return, this strengthens downside support during pullbacks.

6. Financials at a glance

Dolphin Research view

Since Q3 last year, BIDU's valuation sentiment has risen notably, with FY26 EV/GAAP EBIT moving from 10x to a peak of 25x, then recently easing to 20x. The early rally was driven by new 'narratives' resetting expectations—domestic compute substitution, spin-offs/IPO of innovation businesses, higher shareholder returns, and dual-primary listing followed by Stock Connect inclusion.

The slide since early this year alongside HK equities also stems from a shift in the AI disruption narrative.

As of now, the above narratives: (1) Domestic compute-substitution trades flare up periodically, the Kunlun Chip spin-off IPO is in process, and the $5bn buyback is announced. Remaining catalysts include Robotaxi (Luobo Kuaipao) potential spin-off, and dual-primary listing followed by Stock Connect inclusion.

In other words, half of the near-term positives are priced in, while the other half has some expectations gap but is partly known, with some optimistic capital likely pricing them. This suggests only under specific conditions (valuation pullback + approaching execution) can the rerating effect recur.

(2) Negative narratives seem persistent, with tighter-liquidity expectations and ByteDance's siphoning effect weighing on near-term sentiment.

At this point, the $45.6bn market cap aligns with Dolphin Research's neutral valuation in 'Kunlun Chip IPO Accelerates: BIDU's Google Moment?' (some institutions are more bullish on BIDU given higher pricing assumptions for Kunlun Chip).

The difference now is higher shareholder returns add more downside support if valuation falls. Upside would require positive fundamental surprises—e.g., cloud revenue meaningfully beating estimates, or legacy ads bottoming and recovering.

For Q4, Dolphin Research believes cloud outperformed, but legacy ads dragged, and other AI businesses did not show aggressive growth. Fundamentals thus lack a strong catalyst to push the multiple higher.

In other words, without a valuation pullback creating better risk-reward, any rebound still relies on near-term 'narratives': new progress on Kunlun Chip IPO timing (mid-year expected), Robotaxi's potential listing/financing, and dual-primary listing plus Stock Connect inclusion (expected in H2) to drive a meaningful repair.

Detailed analysis below

I. Business architecture

BIDU is one of the few internet names to break out results in detail as: 1) Baidu Core: legacy ads (search/feed ads), innovation (Intelligent Cloud/DuerOS Xiaodu speakers/Apollo, etc.), and AI-related revenue including AI infrastructure (cloud, LLM API, compute leasing), AI apps (Baidu Wenku, Baidu Netdisk, digital employees), and AI-native marketing (Agents and digital humans);

2) iQIYI: membership, ads, and licensing/other. The split is clear, and with iQIYI a separate listed company with detailed disclosure, Dolphin Research also breaks the two businesses down.

As there is roughly ~1% (RMB 0.2–0.4bn) of eliminations between the two, Dolphin Research's Baidu Core sub-splits may slightly differ from reported figures. The differences do not impair trend assessment.

II. AI mix rose to 43%

Baidu Core revenue was RMB 26.1bn, down 6% YoY, mainly dragged by search/feed ads. AI-related revenue was in high growth, reaching RMB 11.3bn and lifting mix to 43%, partially offsetting the decline in legacy ads.

Dolphin Research restores the old disclosure framework for a more direct comparison: (1) Ads: total -16.5% YoY; legacy ads -26%, with the recovery inflection still unclear. Mobile Baidu MAU was 679mn, with users continuing seasonal QoQ declines.

(2) Cloud and other revenue: up 13% in total. Last year's base was high (projects bunched), so growth decelerated, in line with market expectations.

III. Revenue under pressure; profit defended by efficiency

The AI-centric reorg, optimizing legacy staff redundancies and coordinating cross-dept resources, helps margins. This alleviates overall earnings pressure from the slump in high-GP ads.

In Nov, BIDU set up two LLM R&D depts, one driving a general LLM and the other fine-tuning domain-specific models. In Jan, alongside team changes, BIDU merged Netdisk and Wenku.

Dolphin Research believes this mainly involves adjustments to Baidu Netdisk, particularly workforce optimization after competitive pressure from Quark Netdisk. Meanwhile, Baidu Wenku performed well with AI enablement, and netdisk storage aligns with the same demand chain, making bundling feasible.

Q4 profitability in detail: (1) Core GPM rebounded QoQ despite ad pressure, primarily because some legacy equipment was impaired one-off last quarter; D&A dropped visibly from RMB 2.2bn in Q3 to RMB 1.7bn in Q4.

(2) Core OP was RMB 1.4bn with a 5.5% OPM, both up QoQ but below Street, mainly due to RMB 700mn severance and some bad-debt provisions. Ex these and SBC/other non-recurring items, Non-GAAP OP was RMB 2.84bn, above consensus (RMB 2.54bn).

IV. Capex likely a temporary contraction

With profit improvement in Q4, quarterly FCF appeared to turn positive after a year. However, capex fell below RMB 2.0bn this quarter, and Dolphin Research believes this is not the norm; normalizing growth likely keeps FCF negative.

Dolphin Research 'BIDU' historical pieces:

Recent earnings season

Nov 18, 2025 Transcript 'BIDU (Transcript): AI-Native Marketing as the Next S-Curve'

Nov 18, 2025 Earnings Take 'Heavy AI Storytelling: Can BIDU Really Be Saved?'

Risk disclosure and statements: Dolphin Research Disclaimer and General Disclosure