Rating: Cautiously Bullish | Target Price: HKD 108–126 (fair value range, base-case scenario) | Current Price: HKD 75.95 (close, 2026-10-23) | Margin of Safety: +42% (vs. lower bound of base-case range) | Time Horizon: 12 months
At HKD 75.95, the market cap is approximately HKD 206.3 billion (approximately RMB 175.5 billion, at 1 HKD = 0.8504 RMB), already below the company-defined net cash of approximately RMB 179.2 billion (about HKD 77.6 per share); even on the more conservative narrow-scope net cash of RMB 136.8 billion (about HKD 59.2 per share), the implied enterprise value the market pays for all operating assets — search advertising, AI Cloud, iQIYI, Apollo Go combined — is only about RMB 38.7 billion, roughly 2.65x TTM Non-GAAP net profit attributable to shareholders (RMB 14.58 billion), with a P/B of 0.66x — the lowest decile in 10 years. The other side of the deep discount is genuine bleeding: online marketing revenue growth deteriorated from -3.3% YoY (FY2024) to -19% (2026Q2), and AI capex has driven free cash flow deeply negative (2026H1: -RMB 11.2 billion). We believe the market has priced in an extreme assumption of "earnings shrinking by three-quarters and net cash never returning to shareholders," while the first-ever dividend (management guidance: by end-2026), buybacks, southbound flows, and +50% AI Cloud growth constitute catalysts for mean reversion — but whether the advertising decline accelerates a second time and whether net cash consumption can converge will determine whether this is deep value or a value trap. Cautiously bullish, confidence 0.55.
On a frozen price-anchor basis: market cap of RMB 175.47 billion − narrow-scope net cash of RMB 136.8 billion = core business EV of RMB 38.67 billion (about HKD 45.5 billion, roughly HKD 16.7 per share), or 2.65x TTM Non-GAAP earnings and about 0.30x FY2026E revenue; PE (TTM, Non-GAAP) of about 12.0x. Two constraints must be emphasized: first, about RMB 123 billion of cash sits in domestic subsidiaries and VIEs (~74% of cash-type assets), subject to profit repatriation frictions; applying a 20% availability discount, the per-share cash floor is about HKD 48.6. Second, the cash layer is being consumed by AI capex at a slope of roughly RMB 11 billion per half-year (total cash and investments fell from RMB 294.1 billion to 283.1 billion), and Fitch downgraded the company to A- in August 2026. In other words, the deep discount is fact; the "hard net cash floor" depends on capital-allocation discipline being delivered — which is precisely why this report lists dividend declaration as a key verification condition.
Online marketing YoY: FY2024 -3.3% → FY2025 -13.7% → 2026Q2 -19%, accelerating year by year. Management attributed this at the Q2 earnings call to "AI search transformation + deliberate deferral of AI search monetization + macro factors," and explicitly stated advertising would remain under pressure in H2 2026; AI-native marketing revenue growth plunged from +301% in FY2025 to approximately 0% in 2026Q2, as replacement momentum fades. Advertising is the largest profit source with ~50% marginal margins, so each point of decline erodes operating profit far more than the revenue line suggests. Notably, MAU's -5.2% half-year decline cannot fully explain the -19% revenue drop; the gap mainly comes from monetization-rate contraction and deliberate concession to AI transformation — hence the core evidence for "structural decline" is "the company's persistent divergence from the industry's +7.1%," not a single traffic metric. Apple Intelligence's mainland launch in December 2026 and base effects offer a mechanical path to a narrowing decline, but there is no data evidence yet of an organic bottom.
The growth numbers are real and accelerating, but three points must be seen together: AI new businesses overall fell -8% QoQ in Q2 (13.6 → 12.5 billion); quarterly FCF deteriorated from -3.25 billion in Q1 to -7.95 billion in Q2 (Q2 standalone capex of 11.39 billion nearly doubled QoQ); annualized capex of ~34.6 billion now exceeds annualized intelligent cloud infrastructure revenue (~29.2 billion), and that revenue is low-margin infrastructure — during the depreciation ramp-up, AI Cloud is simultaneously a growth engine and a cash-flow bleed. Until unit economics converge (GPU utilization, customer mix, depreciation cadence), it should not be assigned a standalone growth-stock premium. AI Cloud's overall share is only 6.1%, ranking fifth (Omdia 1H25; Alibaba Cloud 35.8%, Volcano Engine 14.8%) — the company is a chaser trading growth for share.
The triple-buying structure (dividend + buyback + southbound) constitutes "conditional support" at HKD 75.95, not a validated floor: the dividend is soft guidance of "possibly by year-end," amount undetermined, with no timetable reiterated in the latest quarterly report; buybacks totaled only USD 259 million year-to-August 18, accelerating only in September; southbound net buying coexists with falling prices (the stock fell from the September buyback average of HKD 88.48 to 75.95). Citi's USD 6 billion estimate is a sell-side promotional figure; the original report is unpublished. Verification conditions are clear: formal dividend declaration (amount + repatriation arrangements), buybacks raised above HKD 100 million per day, and southbound ownership percentage climbing for consecutive weeks — until all three materialize, this thesis should be treated as "improving, unconfirmed."
Baidu holds ~57.7% of Kunlunxin. At a conservative IPO valuation of USD 20 billion, the equity stake would be worth about HKD 88.7 billion (~43% of the current market cap); but the "USD 50 billion target valuation" is a single-source The Information report and a rumored placement-style offering — the market gave back all of its 8% gain after the 2026-06-29 news day, showing the market has seen this catalyst and refuses to price rumors. Apollo Go delivered ~1 million fully driverless orders in 2026Q2, down about 70% QoQ (regulatory adjustments), recovering from August; its option value is real (first-tier global fleet and license head start), but milestones may be delayed by safety incidents and regulatory reversals. Both options should be treated as low-probability/high-payoff and do not form the basis of a core position.
| Metric (RMB 100M, unless noted) | FY2023 | FY2024 | FY2025 | 2026Q2 (latest) |
|---|---|---|---|---|
| Revenue | 1,345.98 | 1,331.25 (-1.1%) | 1,290.79 (-3.0%) | 313.25 (-4.2%) |
| Net profit attributable to shareholders | 203.15 | 237.60 (+17.0%) | 55.89 (-76.5%) | 23.19 (-68.3%) |
| Non-GAAP net profit attributable | — (not available) | 270.02 | 189.41 (-29.9%) | 25.73 (-46.3%) |
| Gross margin (est.) | 51.7% | 50.3% | 43.9% | ~39% |
| Net margin (est.) | 15.1% | 17.8% | 4.3% | 7.4% |
| Operating cash flow | 366.15 | 212.34 | -30.13 | 34.36 (H1: 61.06) |
| Free cash flow (company basis) | 254.25 | 131.00 | -150.86 | -79.54 (H1: -112.0) |
| Cash + cash-like (narrow scope, period-end) | — | — | 2,393.54 | 2,407.32 (2026-06-30) |
| Interest-bearing debt (period-end) | — | — | 895.12 | 1,038.60 (2026-06-30) |
| Liabilities/assets ratio (company basis) | — | — | 24% | 27% (2026-06-30) |
| Online marketing revenue | 812.03 | 785.63 (-3.3%) | 678.37 (-13.7%) | 131.00 (-19%) |
| Baidu Core AI new business revenue | — | 270 | 400 (+48%) | 125 (+25%) |
| Baidu App MAU (millions) | — | 679 (2024-12) | 679 (2025-12) | 644 (2026-06) |
| Apollo Go quarterly fully driverless orders (millions) | — | 1.1 (2024Q4) | 3.4 (2025Q4) | ~1.0 |
Note: The company is in a net cash position (narrow-scope net cash of ~RMB 136.8 billion), so net debt/EBITDA is not applicable; recurring net profit (our estimate, excluding fair-value net gains on long-term investments of ~RMB 3.99 billion and disposal gains of 0.43 billion, adding back long-term investment impairment of 0.28 billion) was approximately RMB 15.26 billion in FY2025 and approximately RMB 21.65 billion in FY2024 on the same basis.
Drivers of metric changes (items with YoY ≥ ±20%): ① FY2025 net profit attributable -76.5%: a one-time impairment of long-term assets of RMB 16.2 billion in 2025Q3 (core asset-group recoverability test failed; third-party DCF discount rate ~13%, revenue CAGR assumptions 8%–18%, attributed to "the rapid iteration of AI technology" and internal operational changes); the company disclosed net profit attributable of RMB 19.4 billion excluding the impairment; ② FY2025 operating cash flow turned negative (+21.2 → -3.0 billion): the company did not itemize reasons; the CFO only stated H2 2025 turned positive with a combined RMB 3.9 billion; ③ 2026H1 capex +158% (6.7 → 17.3 billion): purchases of servers/network equipment, expansion of cloud data centers, and support for generative AI R&D, plus RMB 21.5 billion in contracted capital commitments; ④ FY2025 cost of sales +10%: higher bandwidth, depreciation, and server rack costs, mainly tied to cloud services development; ⑤ 2025 goodwill +14.2 billion: completed the YY Live acquisition in February 2025 for ~USD 2.1 billion; ⑥ 2026H1 net profit attributable -62%: high base from prior-year fair-value gains on long-term investments (other income, net: 9.35 → 0.81 billion) + net FX losses of RMB 2.15 billion.
Q2 revenue was RMB 31.325 billion (-4.2% YoY, -2.3% QoQ), net profit attributable RMB 2.319 billion (-68.3%), Non-GAAP net profit RMB 2.573 billion (-46.3%, net margin ~8%, a multi-year low). The structure showed "one down, one up": online marketing of RMB 13.1 billion (-19%, wider than Q1) and iQIYI of RMB 6.287 billion (-5%) continued to shrink, while intelligent cloud infrastructure of RMB 7.3 billion (+50%) and GPU Cloud +283% accelerated; AI new businesses totaled RMB 12.5 billion, half of Baidu's general business revenue. Cash flows showed a scissors gap: operating cash flow of RMB 3.436 billion (positive for four consecutive quarters, cash conversion recovering), but quarterly capex of RMB 11.39 billion (+199.7%) drove FCF deeply negative at -7.954 billion. The market reacted sharply: US shares fell 12.73% the next day; Jefferies cut its target from HKD 170 to 163, Deutsche Bank from 186 to 162 (both maintaining Buy) — the divide lies between "ad stalling + AI investment squeezing profits" and "AI Cloud scaling + net cash support." Against sell-side consensus, both Q2 revenue and profit missed (FY2026 Non-GAAP consensus range of RMB 11.0–15.6 billion is extremely wide), but AI cloud infrastructure growth, buybacks, and listing progress beat expectations.
Model sketch: Baidu is transitioning from an "asset-light traffic tax" (search performance advertising, 52.6% of revenue, historical gross margin of ~50%+) toward a hybrid "asset-heavy compute and operations" model (GPU cloud infrastructure, Robotaxi fleet, front-installed complete vehicles). Advertising revenue is non-subscription and squeezed by both budget cycles and entry-point substitution, with no pricing power (declining volume and price); cloud infrastructure revenue enjoys pricing power during the compute-scarcity period but carries gross margins significantly below advertising; iQIYI is subscription + advertising but with rigid content costs; Apollo Go is still in an investment phase and has not yet generated revenue at scale.
Cash content of earnings: OCF/attributable net profit of 1.80 (FY2023) → 0.89 (FY2024) → -0.54 (FY2025) → 1.06 (2026H1); FCF/attributable net profit of 1.25 → 0.55 → -2.7 → -1.94. FY2025's "book profit of RMB 5.59 billion but operating cash flow of -RMB 3.0 billion" is a clear earnings-quality warning; 2026H1 OCF recovered to RMB 6.11 billion (four consecutive positive quarters), but was consumed by RMB 17.3 billion of CapEx, leaving distributable cash to shareholders negative—if the long-term FCF/net profit ratio below 0.5 persists for 2–3 years, the "net cash floor" will be self-consumed (see valuation chapter sensitivity analysis for details).
Non-GAAP looseness: In the FY2025 Non-GAAP reconciliation, SBC of RMB 3.605 billion was added back in full (64.5% of GAAP attributable net profit), and impairments of RMB 16.2 billion and severance of RMB 708 million were also added back—Non-GAAP net profit was 3.4× GAAP. The definition is loose, and the 2026 new equity incentive plan reserves approximately 273 million shares (~10% of existing share capital), a dilution not reflected in the add-back definition. The declining absolute SBC (RMB 6.345 → 4.784 → 3.605 billion) is a positive signal. On recurring earnings: FY2025, excluding fair value and disposal gains/losses, was approximately RMB 15.26 billion—a large gap versus attributable net profit of RMB 5.59 billion. The true trend in core profitability is "-29% rather than -76%", though it is by no means the cliff the reported attributable figure suggests.
Return on capital: This report estimates FY2025 ROIC at approximately 5.3% (impairment-excluded operating profit of RMB 10.4 billion × (1-15%) ÷ operating invested capital of ~RMB 164.9 billion), below the 8% warning line and well below the ~9.5% cost of equity—the capital efficiency of the current AI reinvestment has yet to prove itself; this is the substance of the "pseudo-growth" risk. The presence of ~RMB 136.9 billion in book net cash means capital returns depend mainly on reinvestment efficiency rather than the stock.
CapEx quality: FY2023–FY2025 CapEx/depreciation was only 0.52/0.40/0.57; 2026H1 CapEx of RMB 17.3 billion already reaches 1.4× full-year FY2025—investment intensity has surged, and the company does not break out maintenance vs. growth CapEx. This report's EPV estimate assumes maintenance CapEx of ~RMB 9.5 billion/year (GPU 3–5-year refresh cycle + routine maintenance), with the remainder as growth investment. If compute customers and utilization fall short of expectations, these assets carry precedent risk of repeating 2025's impairment.
Red flags: Accounts receivable FY2025 +28.4% vs revenue -3% (loosened collection terms, credit impairments provisioned year after year); goodwill of RMB 36.8 billion (13.5% of net assets); shrinking core KPI disclosure (ERNIE API call volumes ceased disclosure from 2025, MAU YoY figures no longer given); zero dividends since listing (dividend policy only first adopted in 2026).
Consistency of words and deeds: High delivery on capital-markets commitments—the dual primary listing promised at the Q2 earnings call as "effective within this year" was delivered on 2026-09-01, and September Stock Connect inclusion also came on schedule; the buyback authorization (USD 5 billion) is being executed. Low delivery on operational quantitative commitments—in February 2025, Robin Li said "AI-related investments will deliver greater results in 2025"; actual FY2025 revenue fell -3% and attributable net profit -76.5% (-29% even excluding impairment). The dividend "may be paid by the end of 2026" remains soft guidance to date. The company rarely provides quantitative revenue/profit guidance, relying mainly on qualitative statements like "strong momentum," and when asked about margins only says "there is still room for improvement." Overall judgment: prefers qualitative statements, pragmatic on capital-markets actions, operational delivery yet to be verified.
Shareholder friendliness: Zero dividends since listing (dividend policy first adopted 2026-02, first payout possibly by end-2026); buybacks totaled ~USD 1.7 billion cumulatively in 2023, over USD 1 billion in 2024, and ~USD 375 million from early 2026 through September, accelerating from September to ~HKD 43–50 million per day; total shares fell from 2.790 billion weighted in 2024 to 2.717 billion (cancellation-based buybacks), with no record of share issuance or placements. Judgment: shifting from neutral to friendly; dividend delivery is the watershed.
Risk signals: Robin Li holds 60.0% of voting rights via Class B shares (10 votes/share), so capital allocation is effectively decided by one person—there is no check on the trade-off between "AI capex prioritized over shareholder returns" and "buybacks + dividends prioritized." Ma Dongmin has established a 10b5-1 prearranged trading plan for share sales (waiver granted by the Stock Exchange on 2026-08-26; she sold shares multiple times in 2023–2024); YY Live was acquired at a reduced price of USD 2.1 billion (originally USD 3.6 billion), adding RMB 14.2 billion in goodwill, with the company itself stating it "cannot guarantee expected benefits" and has no substantive recourse against JOYY; the timing of the RMB 16.2 billion impairment purge concentrated in 2025Q3 smacks of flattering next year's base.
| Segment | FY2025 Revenue (RMB bn) | Share | YoY | 2026Q2 YoY | One-line business logic |
|---|---|---|---|---|---|
| Online marketing services | 67.837 | 52.6% | -13.7% | -19% | P4P performance ads; volume and price both shrinking under AI entry-point substitution; largest profit source continuously bleeding |
| Intelligent Cloud infrastructure (incl. GPU cloud) | 19.8 | 15.3% | +34% | +50% (GPU cloud +283%) | Asset-heavy expansion during compute scarcity; only double-digit growth driver |
| AI applications (Wenku/Netdisk/agents) | 10.2 | 7.9% | +5% | +3% | Subscription + usage-based fees; sticky but small |
| AI-native marketing services | 9.8 | 7.6% | +301% | ~0% | AI-transformed ad formats; stalled after high growth |
| Others (intelligent driving/Xiaodu, etc.) | — | ~9% in 2026Q2 (RMB 2.3 bn) | — | -15% | Apollo Go has not generated revenue at scale; investment phase |
| iQIYI | 27.290 | 21.1% | -7% | -5% (segment turned loss-making in H1) | Membership + ads; rigid content costs; 2026H1 segment operating loss of RMB 0.33 bn |
Profit engine: By the "revenue share × marginal margin" logic, online marketing (52.6% × ~50% marginal margin) remains the absolute profit mainstay—which is exactly why the structural decline in C2C is so fatal; Intelligent Cloud infrastructure's revenue share has reached 15.3% with the fastest growth, but it is a low-gross-margin infrastructure line whose gross profit contribution does not match its revenue share; iQIYI accounts for 21.1% of revenue but segment costs approximately equal revenue (FY2025 segment costs RMB 27.061 bn vs revenue RMB 27.290 bn), contributing almost no profit and turning loss-making in 2026H1. Gross margin structural differences: the gross margin gap between advertising (historically ~55%+) and cloud infrastructure (significantly below the group average) far exceeds 10 percentage points—their business essence is the difference between "collecting a traffic tax" and "selling compute as a heavy asset." The group gross margin decline from 51.7% to ~39% is a mirror of the revenue mix migrating from the former to the latter, not deterioration of a single business.
Accounting red flags: ① Large one-time impairment (medium severity): 2025Q3 alone provisioned RMB 16.2 billion in long-lived asset impairments ("due to external environment fluctuations (particularly the rapid iteration of AI technology) and internal operational changes"), clearing the decks at the earnings trough and sharply lowering the profit base from 2026; ② Loose Non-GAAP definition (medium): FY2025 added back RMB 3.605 bn SBC + RMB 16.2 bn impairment + RMB 0.708 bn severance, and deducted RMB 3.989 bn fair value gains, yielding Non-GAAP of RMB 18.941 bn—3.4× GAAP; ③ Accounts receivable growth divergence (medium): FY2025 net receivables +28.4% vs revenue -3%; a one-time credit loss provision of RMB 561 million was made in 2024Q4; ④ Fair value/disposal gains noise (medium): FY2025 other income, net, of ~RMB 12.5 bn vs impairment-excluded operating profit of ~RMB 10.4 bn—non-operating items exceed operating profit; ⑤ Low-price acquisition with high goodwill (low): YY Live, even at the reduced price, added RMB 14.2 billion in goodwill.
Cross-period consistency: ① OCF/attributable net profit deteriorated year by year 1.80 → 0.89 → -0.54 without complete explanation (only noting a positive RMB 3.9 bn in 2025H2)—inconsistent; ② Attribution for the online marketing decline shifted progressively across three reporting periods (2024 annual report: "weakness in offline industries" → 2025 annual report: "AI transformation impact on monetization + macro" → 2026 transcript: "proactively deferring AI search monetization")—inconsistent attribution, with the trend increasingly acknowledging structural causes; ③ Explanations for CapEx growth and FCF deterioration (servers/data centers/AI R&D) were stable across all three periods—consistent; ④ Shrinking core KPI disclosure (ERNIE API call volume, MAU YoY, peak weekly orders successively discontinued or downplayed)—unexplained by the company; ⑤ Under the same shock environment, RMB 16.2 billion was impaired on long-lived assets while RMB 36.8 billion of goodwill saw zero impairment, and the annual report itself acknowledges "circumstances that led to recent long-lived asset impairments could likewise lead to future goodwill impairment"—inconsistent.
Current market data and multiples: Share price HKD 75.95 (2026-10-23 close), market cap HKD 206.3 bn ≈ RMB 175.47 bn (HKD 1 = RMB 0.8504, 2026-10-23/25 CFETS), total shares 2.717 billion. PE(TTM, GAAP) is negative (distorted by the RMB 16.2 bn impairment; not meaningful); PE (static FY2025 GAAP) 33.4× (66th percentile over the past 5 years, equally distorted)—percentiles are for reference only; this name's earnings are in a transition period of "advertising recession × AI depreciation ramp," so mechanical percentiles are not taken as a cheapness/expensiveness signal. Representative measures: PE(TTM, Non-GAAP) 12.0×, PE (FY2026E Non-GAAP midpoint RMB 13.1 bn) 13.4×, PB 0.66× (this report's calculation: 0th percentile over the past 1/3/5/10 years—a ten-year low), PS(TTM) 1.38×; core business measure: EV RMB 38.67 bn ÷ TTM Non-GAAP RMB 14.58 bn = 2.65×, ÷ FY2026E revenue ~0.30×—this is this report's most important cheapness/expensiveness reading.
| Company | PE(TTM) | ROE | Growth profile | Key difference vs Baidu |
|---|---|---|---|---|
| Tencent (0700) | 14.8× | 21.1% (FY2025) | FY2025 revenue +13.9% | Cash cow funding AI; ads not cannibalized by its own AI |
| Alibaba (9988) | 24.8× | 9.9% (FY2026) | Cloud segment +45% (June 2026 quarter) | Cloud ~6× Baidu's size; AI capex similarly suppressing profit |
| Kuaishou (1024) | 7.6× | 26.3% (FY2025) | FY2025 revenue +12.5% | Net winner of ad budgets migrating to short video |
| Alphabet (GOOGL) | 15.9× | 35.7% (FY2025) | 2026Q2 search ads +17% | Search ads still growing after full AI Overviews rollout—AI does not necessarily kill search |
| Baidu (09888) | 12.0× (Non-GAAP) | ~2.1% (FY2025, depressed by impairment) | 2026Q2 revenue -4.2%, AI cloud infrastructure +50% | Most direct casualty of AI-displaced search + AI compute chaser |
Market-implied expectations: The current price of HKD 75.95 = net cash (narrow definition, HKD 59.2/share) + HKD 16.7/share for the core business. Solving backward at a 9.5% cost of equity: the core business need only permanently contribute ~RMB 3.7 billion/year of FCFE to be worth the current price—equivalent to assuming core earnings permanently stagnate at zero growth after shrinking ~75% from the current RMB 14.58 billion. In one sentence: the current price requires core earnings to shrink to a quarter of actual levels, while reality shows AI new businesses growing +25%~+48% and TTM Non-GAAP still at RMB 14.58 billion. This pessimistic pricing is partially justifiable on two counts: net cash is being consumed at ~RMB 22.4 billion annualized (2026H1 FCF -RMB 11.2 bn), and management previously made no large return commitments—the market prices "cash returned to shareholders" at zero; if the Q4 dividend lands, that assumption will be directly falsified.
Three-tier value (EPV): Asset value (floor) = narrow-definition net cash of ~HKD 59.2/share (mainland cash of RMB 123 bn faces repatriation friction; at a 20% accessibility discount, ~HKD 48.6); zero-growth EPV = normalized FCFE of HKD 4.76/share (FY2026E Non-GAAP midpoint RMB 13.1 bn → bridged to FCFE of ~RMB 11 bn, after deducting maintenance CapEx of ~RMB 9.5 bn) ÷ 9.5% cost of equity + net cash 59.21 ≈ HKD 109.3/share (range 102–116); the growth option tier is negative (~-44% of current price): the current price is at a ~30% discount to zero-growth EPV—the market is not only paying nothing for growth from AI Cloud, Apollo Go, Kunlunxin, or the Apple partnership, but has even marked down the zero-growth value to 70 cents on the dollar. Under the three-tier framework, the current price's support structure is: net cash 78% + core earnings ~22%; skepticism about the cash tier's accessibility and consumption slope is the valuation's first sensitivity variable. Asset value is an independent floor reference and is not added to EPV.
Three scenarios and odds (end-2026–2027, HKD):
| Scenario | Probability | Fair range | vs current price | Key swing factors |
|---|---|---|---|---|
| Bear | 35% | HKD 55–75 | -27.6% ~ -1.3% | Ads accelerate again to -25%+, AI cloud stalls below +15%, net cash falls to RMB 90–110 bn with RMB 15 bn annual burn; core priced at 6–7× trough earnings; external reference: Morgan Stanley downgraded to Underweight on 2026-10-23, target USD 70/ADS (≈HKD 68.6) |
| Base | 45% | HKD 108–126 | +42.2% ~ +65.9% | Ad decline narrows to -5%~-10% (low base + Apple entry point + AI monetization pilots), AI new businesses +25%~+30%, 2027 Non-GAAP recovers to RMB 15.5–17.5 bn, net cash holds at RMB 115–120 bn; core at 9–10× (anchored to Kuaishou 7.6×/Tencent 14.8×/industry median 10–12×, taking the low-to-middle band for Baidu's "declining ads + growing AI" mix); US consensus mean of USD 137.47/ADS ≈ HKD 134.8 equivalent sits above this range's cap; the range excludes any Kunlunxin IPO uplift |
| Bull | 20% | HKD 180–210 | +137%~+176% | Apple partnership scales + AI cloud +45%~55% + Kunlunxin IPO at USD 20–30 bn (57.7% stake worth RMB 77–115.4 bn) + Apollo Go achieves positive unit economics and expands; for reference, Jefferies HKD 163/Deutsche Bank HKD 162 (options possibly partially priced in); this scenario requires full delivery |
The current price of HKD 75.95 sits in the upper half of the bear range—distribution skewed low, odds skewed right: downside to bear-case midpoint ~-14%, upside to the base range +42%~+66%. Probability-weighted fair value ~HKD 114.4 (expected return ~+51%). The high elasticity of this expected value comes from the deep discount itself; delivery timing depends on two verification conditions—dividend and advertising inflection—and should not be understood as a linear extrapolation.
Our forecasts vs third parties:
| Definition | FY2026 Revenue | FY2026 Non-GAAP attributable net profit |
|---|---|---|
| This report | RMB 126–129 bn (-2%~0%; implied H2 RMB 62.6–65.6 bn, -2%~+3%) | RMB 12.2–14.0 bn (midpoint RMB 13.1 bn, ~-31%) |
| Management guidance | No quantitative guidance (qualitative: ad pressure in H2, AI cloud strong or accelerating, ROIC discipline upheld) | — |
| Sell-side consensus | Extremely wide range | RMB 11.0–15.6 bn (Yahoo $5.66/ADS ~ SeekingAlpha $8.02/ADS) |
FY2027: This report forecasts revenue of RMB 129–138 bn (+2%~+7%) and Non-GAAP of RMB 15.0–17.5 bn (recovery)—drivers: ad decline narrowing to -5%~-10%, AI new businesses +25%~+30%, with a high depreciation base suppressing GAAP elasticity.
Conclusion: Undervalued. Target price HKD 108–126 (base case), safety margin +42% (on the base-case lower bound). Quality and price considered separately: quality is mediocre-to-weak (ROIC 5.3%, structural advertising decline, negative FCF), while price is deeply cheap (PB at a ten-year low, core earnings at 2.65×, implied assumption of a 75% earnings shrinkage). The base-case 9–10× exit multiple for the core business is anchored to the peer median (within the Kuaishou 7.6×/Tencent 14.8× band, low-to-middle tier), not the company's current multiple; the current 12.0× (including the net cash dilution effect) is for reference only.
China internet advertising: 2025 market size RMB 793.08 bn, +4.6% YoY (QuestMobile); 2026H1 reached RMB 385.34 bn, +7.1% YoY—a stable single-digit growth market in mild recovery. The broad definition (iResearch, including e-commerce ads) already exceeded RMB 1.1 trillion in 2023; the two definitions must not be mixed. Implied steady-state growth of ~5%–7% over the next 3 years (trend extrapolation; no front-line institution's public 5-year CAGR forecast available). China AI cloud: The "AI cloud market" as first defined by IDC reached RMB 68.4 bn in 2025; MaaS (Model-as-a-Service) revenue in 2025H1 was RMB 1.29 bn, +421% YoY; industry-wide token volume in 2025 was 1,994 trillion, +16× YoY, with inference demand already surpassing training as dominant. IDC projects China's total AI investment to exceed USD 100 bn by 2028 (5-year CAGR 35.2%), with intelligent computing capacity growing from 1,037 EFLOPS in 2025 to 2,782 EFLOPS in 2028. Robotaxi: Frost & Sullivan projects the China market to grow from ~USD 200 million in 2025 to USD 39 bn by 2030 (implied CAGR ~187%), and the global market to ~USD 66.6 bn by 2030 (RMB 834.9 bn); global 2024–2030 CAGR ~239%. A commercialization-launch market with an extremely low base.
Baidu sits at the intersection of a negative and a positive structural demand curve. Negative side (entry point substitution): AI assistant users reached 1.157 billion in September 2026, +44% YoY (QuestMobile); AI-native apps reached 546 million MAU (+82.1%); meanwhile Baidu App MAU fell from 679 million (Dec 2025) to 644 million (Jun 2026). The query entry point is migrating from the "search box" to the "dialog box," rewriting the compute consumption and monetization chain per interaction—search is being downgraded from "the entry point" to "one of the AI's answer sources." Positive side (compute explosion): a single AI Q&A consumes far more inference compute than keyword retrieval, and industry daily token consumption has climbed several-fold within a year from roughly 1.6 trillion; Baidu AI Cloud infrastructure revenue grew +79% in 2026Q1 and +50% in Q2 (GPU cloud accelerating quarterly from +184% to +283%), which is the P&L reflection of this curve. Both curves stem from generative AI, which means they are not an independent hedge: GPU depreciation and expenses are devouring the advertising cash cow on the same income statement—if advertising drops another notch, AI cloud will simultaneously lose its internal source of capital.
Advertising chain (content ecosystem/traffic entry—agencies—advertisers): Baidu owns its own entry point with low upstream dependence, and historically margins are highly retained at the platform. AI chain (chips—cloud infrastructure—large models—MaaS/API—applications): the upstream is AI chips; the反复 U.S. H20 export restrictions (ban in Apr 2025 → conditional release in Jul → H200 restored in Dec with a 25% revenue share) have substantially weakened procurement bargaining power, while domestic substitution (Ascend/T-Head/Kunlunxin) accelerates—Baidu is one of the few full-stack in-house vendors in China spanning "chips (Kunlunxin)—framework (PaddlePaddle)—model (ERNIE)—platform (Qianfan)"; Kunlunxin shipped 116,000 units in 2025, ranking third among Chinese vendors, and won 70–100% share of the CUDA-like segment in China Mobile's centralized procurement. Robotaxi chain (front-mounted vehicle RT6—L4 autonomous driving stack—self-operated/Uber and other channels): Baidu is fully vertically integrated. Direction of value distribution: the gross profit pool is being redistributed from the "advertising platform tax" to "computing infrastructure," which is an order of magnitude more capital-intensive—the group gross margin falling from 51.7% to about 39% is the mirror image.
Advertising: concentration among top media is extremely high (Douyin/WeChat/Taobao dominate traffic distribution), and search's share is passively shrinking; the substitution threat is precisely AI assistants themselves (Doubao MAU approximately 340–620 million—Month Fox and QuestMobile report 620 million/382 million respectively; the large gap between methodologies, both shown for reference; DeepSeek's share at the API layer is 86.8%, Month Fox, Jul 2026). AI Cloud: a highly concentrated "one super, one strong" landscape—per Omdia's "China AI Cloud Market 1H25" (market size RMB 22.3 billion): Alibaba Cloud first at 35.8%, Volcengine second at 14.8%, Baidu AI Cloud fifth at 6.1%; on the large-model public cloud call volume metric (IDC 1H 2025), Volcengine leads at 49.2% and Baidu has dropped out of the top tier. Baidu has slipped from being "jointly No. 1 in AI public cloud" in 2024 (IDC methodology, 24.6%) to a chaser currently trading GPU compute growth for share. The price war has shifted from the extreme subsidies of 2024 to over 70% of vendors raising prices in 2025, with the pricing anchor moving toward "business outcomes." Robotaxi: China's first tier consists of Apollo Go, Pony.ai, and WeRide (plus Didi Autonomous Driving), with CR3 holding absolute dominance; Pony targets a fleet of 3,000+ vehicles in 2026 with an overseas pipeline of 4,000+, while WeRide's global fleet of roughly 1,300 vehicles with a 35% gross margin validates a path to profitability for L4 services per se; Baidu leads in fleet mileage and regulatory access with 28 cities and 240 million cumulative fully driverless kilometers. On the compute supply side, "not a single GPU is idle" (Cailian Press, May 2026), and Tencent + Alibaba's combined quarterly capex exceeded RMB 120 billion in 2026Q2—the arms race has raised the cost of following.
Advertising is in a mature steady state; AI cloud/large models are in a high-growth phase at the mid-stage of an upward capex cycle (global hyperscalers' 2026 capex is about USD 602 billion, with 75% going to AI); Robotaxi is transitioning from the introduction phase to the growth phase. Leading indicators: for advertising, quarterly market growth and Baidu App MAU; for AI cloud, token call volume, GPU cloud growth, and hyperscaler capex; for Robotaxi, the pace of license issuance and daily orders per vehicle. Regulation is a double-edged sword: the generative AI "dual filing" system raises barriers and favors compliant full-stack large vendors, and the "AI+" initiative was included in the Government Work Report; but after the Apollo Go Wuhan suspension incident in Apr 2026, regulators tightened access (Reuters reported a nationwide pause on new license issuance, while People's Daily Online denied the "pause"—both reports shown; the tightening is a fact), directly hitting its home-market expansion; chip restrictions have spurred the formation of a domestic compute chain while introducing supply uncertainty.
Search: StatCounter share across all platforms in China stands at 46.85%, still No. 1 (Sep 2025–Sep 2026), but on desktop it has been overtaken by Bing (Bing 37.15% vs Baidu 19.89%, host methodology, Aug 2025–Sep 2026)—both metrics shown; share ≠ monetization; with online marketing revenue declining continuously, the positioning is "defensive leader, with both share and revenue falling." AI Cloud: a fifth-ranked chaser with the fastest marginal improvement (No. 1 in contract wins + triple-digit GPU cloud growth). In the "AI entry point" competition it is clearly the laggard: Robin Li said at the Q2 earnings call that ERNIE must "return to the first tier of foundation models"—an official admission that iteration has fallen behind DeepSeek/Doubao/Qwen. Robotaxi: global first tier, with moats from safety mileage data + front-mounted mass-produced vehicle (RT6) + first-mover licenses in multiple countries. Overall positioning: a transformation case of "the old entry point being replaced by AI, while new compute + mobility positions await realization"—the stock is down -34.8% over the past year, -47.2% year-to-date, and -52.7% from its 52-week high of HKD 160.5 (as of 2026-10-23), which is precisely the market's pricing of this repricing.
Overall assessment: cautiously bullish, confidence 0.55, time horizon 12 months. The investment essence of Baidu is "buying, at about 2.65x core earnings, a search cash cow being repriced by AI, a +50%-growth compute chaser, two unpriced options, and a pile of net cash constrained by repatriation." The deep discount already prices in extreme pessimism, and mean-reversion triggers are densely lined up over the next two quarters (Q3 low-base results, Baidu World Conference, launch of Apple Intelligence in China, the first dividend); but the -19% slope in advertising has not yet bottomed, and net cash is being consumed at a pace of tens of billions per half-year, making the duration of value realization uncertain. Strategically, it is advisable to build positions in batches and add based on verification conditions: formal declaration of the dividend and advertising decline narrowing to within -10% are the two hardest confirmation signals; if advertising falls below -25% or full-year FCF is worse than -RMB 15 billion, the deep-value thesis should give way to the declining-asset thesis, and one should exit with a stop-loss.
Risk warnings: second acceleration of the structural decline in advertising; runaway AI capex (burning RMB 15 billion+ annually for three years would erode a third of the cash floor); linked impairment of the RMB 36.8 billion goodwill; renewed regulatory tightening on Robotaxi; failure of the Kunlunxin IPO or pricing far below rumors; friction on repatriating onshore cash; single-decision-maker risk in capital allocation under Robin Li's 60% voting power; execution of Melissa Ma's 10b5-1 selling plan.
This report is based on public information and the report's own estimates; prices are frozen at the 2026-10-23 close; this does not constitute investment advice.