Hithink RoyalFlush: Advertising Supplied 78% of 2025 Incremental Revenue, and CNY230.26 Already Sits at the CNY228 Base-Case Value

The H1 2026 pre-announcement strengthens the cyclical interpretation rather than disproving it. Q1 revenue was CNY1.053bn, up 40.8%; attributable profit was CNY255.9m, up 112.6%; operating cash flow reached CNY838.8m, up 167.8%. Contract liabilities rose to CNY2.656bn from CNY1.769bn at year-end, a strong forward-demand signal. Yet sales expense also climbed 77.7% to CNY294.7m as marketing, advertising and sales compensation rose. The company explicitly described active capital markets and greater demand for financial information as important Q1 and H1 drivers.

At the midpoint of the H1 pre-announcement, H1 attributable profit is about CNY928m. Subtracting Q1 implies roughly CNY672m of Q2 profit, compared with approximately CNY381m in Q2 2025. On a pro-forma trailing-twelve-month basis, replacing H1 2025 profit with the midpoint of H1 2026 produces about CNY3.63bn of attributable profit. At the August 14 market capitalization, that is about 47.7 times pro-forma trailing earnings, somewhat below the 51.9 times TTM multiple displayed by a market-data provider that has not incorporated the unreported H1 result. These are my calculations from the filings and the current market capitalization.

The company’s roots explain why it has been able to monetize these cycles so well. The listed company’s legal predecessor dates to 2001, but the operating team’s history reaches further back: current chairman and controller Yi Zheng has been chairman/general manager of Hangzhou Hexin Software Technology since 1994. Its precursor, Shanghai Hexin Software Technology, was established in August 2001; in 2007 it acquired Hangzhou Hexin, moved its center of gravity to Hangzhou, changed its name and converted into a joint-stock company. By then, the product architecture was already centered on securities quotes, online trading systems and investor decision tools.

The IPO preserved that identity. Hithink offered 16.8m shares at CNY52.80 in 2009, with post-issue capital of 67.2m shares, gross proceeds of CNY887.0m and net proceeds of roughly CNY842.5m. The issue P/E was 92.57 times. I use the verified Shenzhen listing date of 2009-12-25; one later company ESG document describes the listing as December 24, but the IPO/listing materials and exchange record support December 25. The IPO story was already a mixture of online securities trading systems, Level-2 market information, premium decision tools, institutional terminals and mobile financial information rather than a single software product.

Viewed vertically, five stages matter more than individual product launches.

From 2001 to the 2009 IPO, Hithink solved a distribution problem created by China’s rapidly expanding retail securities market: ordinary investors and brokerages needed standardized desktop market data, charting and online-trading infrastructure. Its early advantage came from software engineering, securities-market specialization and distribution across brokers rather than ownership of a brokerage balance sheet. The post-IPO capital structure was already founder-controlled, and the company entered the public market at a high growth multiple.

The 2010s brought platformization. Desktop trading migrated toward mobile; free market data became abundant; the economics shifted from selling basic access toward monetizing a much larger installed base through premium decision tools, advertising, fund distribution and higher-value data. Hithink survived the commoditization of basic quotes because its destination accumulated workflows, watchlists, screening, community, alerts and broker connectivity that users visited even when the underlying trade could be executed elsewhere. The strategic residue of this period is today's retail traffic pool.

The next stage was business-model crystallization. Hithink became an unusually asset-light financial-information platform rather than following East Money all the way into brokerage. That choice left commission and margin-financing income on the table, but it also kept balance-sheet requirements low and permitted extraordinarily high gross margins. By 2020–21, a favorable securities-market cycle made that structure visible in the accounts: 2020 revenue rose 63.2%, profit rose 92.1%, and ROE was about 38%; 2021 profit reached CNY1.91bn and ROE remained above 30%.

The 2022–23 period was the most useful stress test in the recent record. Hithink did not break, lose liquidity or require retrenchment. It kept R&D high even as market-sensitive revenues slowed. The cost was falling earnings and ROE: attributable profit dropped from CNY1.91bn in 2021 to CNY1.40bn in 2023. This period proves both sides of the investment case. The franchise remained highly profitable through a poor activity environment, which supports the quality argument. The same record proves that earnings are materially cyclical despite recurring subscriptions and institutional contracts.

The 2024–26 stage combines a market-activity recovery with the commercialization of several years of AI investment. In 2024 the company described upgraded HithinkGPT, a more conversational and multimodal iWenCai, and an institutional intelligent platform through which more than 100 financial institutions had built more than 5,000 agents, with daily calls above three million. By 2025, the company said its institutional services covered more than 90% of domestic securities firms, R&D staff reached 3,141, and the product architecture increasingly embedded large models, RAG, agents and financial knowledge bases. These are substantial product facts. They still sit next to a more mundane financial one: advertising supplied most incremental 2025 revenue.

The qualitative portrait is “re-rating” with a strong cyclical layer. Hithink has proven that it can remain profitable and cash-generative across a full securities-market cycle; it has not yet proven that AI has removed that cycle from its earnings. The market is now being asked to decide how much of the post-2024 profit step-up is a new structural floor and how much is activity-sensitive upside.

Financial Vertical Review, Business Model and Moat

The long financial record is better than the headline volatility suggests. Revenue has more than doubled since 2020, the company remained profitable through every recent market environment, operating cash flow has consistently exceeded or approximated accounting profit, and it carries no meaningful bank-debt burden. The quality screen that surfaced the stock was picking up something real: low capital intensity, high margins, high cash conversion and unusually high returns on equity. What that screen cannot tell an investor is whether CNY3–4bn is normalized earnings or a cyclical high.

Sources: company annual reports and annual-report summaries, including 《2022 年年度报告摘要》 (“2022 Annual Report Summary”), 《2023 年年度报告摘要》 (“2023 Annual Report Summary”), and the 2025 filing.

The earnings arc shows a much stronger business than a conventional brokerage but a less defensive one than a mature subscription-data vendor. In the 2021-to-2023 downcycle, revenue lost momentum first, then profit fell much faster because technology and R&D spending continued. During the 2024–25 upswing the same fixed-cost base worked in reverse. Hithink resembles a digital toll road whose traffic and monetization rate both rise with investor excitement, rather than a seat-based enterprise-data vendor where most annual revenue is contractually recurring.

The 2025 segment mix makes the distinction visible.

The 2023 figures are reconstructed from disclosed segment percentages; 2024–25 figures and 2025 gross margins come from the annual-report disclosures.

This table is the report’s most important business-model evidence. The supposedly enterprise-like software operation was just 6.6% of 2025 revenue. Advertising was 57.4%, and value-added telecom, which includes paid retail information services, another 32.4%. Even before counting fund fees, more than four-fifths of revenue was directly tied to monetizing investor attention and willingness to pay. In 2024, advertising alone added roughly CNY666m while the other three categories collectively shrank; in 2025 advertising contributed roughly 78% of total incremental revenue. Those calculations make the cycle much harder to explain away as “AI transformation.”

AI can still make this cyclical machine structurally better. Better queries and generated research can increase time spent in the app, and better user segmentation can improve advertising yields. Better screening and portfolio tools can raise premium conversion, while better institutional agents can increase terminal value and customer retention. The company explicitly says it uses AI to improve advertising effectiveness, iWenCai, iFinD, research, compliance checking, financial questioning and investment-decision support. The distinction that matters: this is currently an enhancement to existing monetization, not a separately disclosed new revenue pool.

Cost structure explains the extraordinary margins. The 2025 segment cost base was concentrated in information/hosting services and employee compensation: information and hosting accounted for about 51.5% of reported operating cost and payroll around 41.9%. Direct product delivery costs are low relative to revenue. R&D, sales and marketing sit below gross profit and behave more like semi-fixed investment. Once a free digital product has acquired the user and the servers/data infrastructure are in place, another advertising impression or premium subscription can carry a very high contribution margin.

R&D was CNY1.145bn in 2025, down 4.0% from CNY1.193bn in 2024, and 19.0% of revenue against 28.5% in 2024 and about one-third in 2023. The lower percentage mostly reflects the revenue surge; the absolute research budget remained above CNY1.1bn. The full report records R&D expense at CNY1.145bn and I found no separately disclosed material development-cost capitalization that would make the reported expense artificially low. I therefore treat R&D as substantially expensed, while recognizing that “zero capitalization” is an inference from the disclosure rather than an explicitly stated accounting ratio.

Cash conversion is unusually good. For 2021–25, aggregate operating cash flow was about 1.15 times aggregate attributable profit by my calculation; each individual year in that period had an OCF/net-profit ratio above 1.0. In 2025 operating cash flow was CNY3.774bn against CNY3.205bn of attributable profit. This partly reflects subscriptions and contract liabilities being collected before revenue recognition, so cash conversion can become especially strong in an upcycle. It remains a genuine quality advantage because Hithink does not need receivables or inventories to absorb growth.

Capital expenditure is small. Consolidated purchases of fixed assets, intangibles and other long-lived assets were CNY142m in 2025, up from CNY104m in 2024. Subtracting all disclosed 2025 capex, rather than only maintenance capex, from operating cash flow gives roughly CNY3.63bn of conservative free cash flow. The company does not disclose maintenance and growth capex separately. For owner-earnings analysis I assume approximately one-half to two-thirds of this CNY142m is maintenance; even treating the entire amount as maintenance changes cash earnings only modestly.

The balance sheet is strong but requires one adjustment in interpretation. Monetary funds were CNY14.04bn at end-2025 and CNY16.03bn at March 2026, while the company had no meaningful bank borrowing. Yet not all headline cash is freely distributable surplus: consolidated cash equivalents at end-2025 were CNY10.04bn, and the group’s fund-distribution operation creates agency-related working-capital liabilities; at March 2026 the balance sheet also included the CNY2.742bn annual dividend payable before its April payment. For valuation I avoid blindly subtracting every yuan of reported cash from enterprise value.

Dividend behavior has become materially more shareholder-friendly. The 2025 annual distribution was CNY51 per ten pre-transfer shares, totaling CNY2.742bn, alongside a four-for-ten capital-reserve share transfer. Adding the smaller 2025 interim dividend takes distributions related to 2025 earnings to roughly 87% of attributable profit by my calculation. The annual payout was completed on 2026-04-10, expanding shares outstanding from 537.6m to 752.64m. A CNY0.20-per-current-share H1 2026 distribution was subsequently proposed alongside the earnings pre-announcement, although that proposal was not an audited interim-report distribution at the research cut-off.

The capital-transfer shares do not create economic value; they only change the denominator. All per-share valuation in this report uses the current 752.64m-share basis. That point matters when comparing old prices, dividends or EPS with current numbers.

Three moats survive serious scrutiny.

The strongest moat is distribution and habitual workflow, not the large language model itself. Hithink sits on a long-established retail-investor destination while supplying systems or services to more than 90% of domestic securities firms. A new model can imitate financial summarization; recreating the installed user workflow, market-data normalization, alerts, saved screens, brokerage relationships, brand recognition and years of user behavior is much harder.

The second is data plus domain engineering. The company describes databases spanning market data, news, macroeconomics, industries, companies, filings, research reports and other structured and unstructured sources; these datasets feed financial search and model training. By end-2025 it reported 587 software copyrights and 136 granted invention patents, including 31 U.S. patents. The moat is the cleaned, permissioned, continuously updated data and financial application layer. Foundation models themselves are becoming cheaper and more interchangeable.

The third is the economics of an independent platform. A broker’s free app is optimized to retain brokerage assets and transactions. Hithink can sit across many brokerage relationships, giving it an incentive to maximize information consumption and analytical workflow rather than commission capture at one broker. Paid products can then sell depth beyond basic free execution: richer data, Level-2 functions, screening, alerts, specialized indicators, strategy tools, natural-language queries and increasingly AI-assisted analysis. The weakness of this moat is visible too. A retail user who needs only quotes, news and order entry can obtain adequate functionality free from a brokerage app, East Money or a large internet platform. Switching costs are much lower than in institutional core software.

Governance is founder-controlled and economically aligned, but minority shareholders have limited influence. At end-2025 Yi Zheng owned 36.13%, Ye Qiongjiu 11.11%, Hangzhou Kaishun Technology 8.83% and Yu Haomiao 4.95%. Ye Qiongjiu and Wang Jin are also shareholders of Kaishun. Yi remains controlling shareholder and actual controller. There are no dual-class shares. This concentration has accompanied a long period of stable management and conservative financing, but it means investors rely heavily on the founder group’s capital-allocation discipline.

That record is favorable so far. The company reported no material shareholder fund occupation, unlawful guarantees, major litigation, annual-report penalty or material operating related-party transaction in 2025; a small related financial-product transaction generated fees of only about CNY674,000. Tianjian has audited the company for 19 years and issued a standard unqualified opinion. No share repurchase program was active in 2025, and the report does not show a material broad share-based compensation burden. A director/shareholder holding at least 5% did file a share-reduction pre-disclosure on 2026-05-15, which deserves monitoring as a supply and alignment signal without implying a motive.

Industry, Regulation and Horizontal Competition

Hithink operates at the intersection of software, market data, digital advertising, wealth-management distribution and securities-market infrastructure. That makes a single “industry TAM” more misleading than useful. The company cites China’s entire software and information-technology service industry at CNY15.48tn of 2025 revenue, up 13.2%, but that is far broader than the monetizable market for Chinese retail financial information. No primary filing gives a credible stand-alone TAM for paid retail financial-information subscriptions. The relevant economic variables are the number and engagement of investors, securities turnover, willingness to pay for decision tools, financial advertising budgets, fund activity and financial institutions’ technology spending.

The profit pool is split according to licensing and customer ownership. Brokerages capture trading commissions, margin-financing interest and account assets; fund distributors capture sales/service fees. Information destinations take advertising and subscriptions, and enterprise technology vendors take software licenses, maintenance and implementation. Hithink participates heavily in information, only modestly in fund distribution and institution software, and does not have East Money’s brokerage earnings engine. That explains both its extraordinarily high gross margin and why its revenue is far smaller than East Money’s despite comparable mass-market visibility.

The principal cycle is the securities-market activity and sentiment cycle, with an advertising cycle layered on top. In a rising market, investors open the app more often, premium tools appear more valuable, advertisers pay to reach an engaged audience and financial products see more attention. The same underlying variable moves several revenue lines at once. In a weak market, fixed R&D and engineering costs remain while those monetization rates soften. Hithink’s 2022 filing explicitly linked lower value-added telecom, software and fund revenue to securities-market volatility; Compass independently describes its own financial-information revenue as positively correlated, with a lag, to securities-market turnover.

Technology iteration is the second cycle. The company began investing in machine learning, NLP, speech and AI well before generative AI became a stock-market theme. Its 2022 filing already described more than 40 AI products/services; the 2023 filing described HithinkGPT and a financial pre-training corpus; 2024 brought large-scale financial agents; and 2025 embedded RAG, multimodal models and agents deeper into retail and institutional products. The durable advantage comes if these tools materially improve conversion, retention or institutional pricing. A model benchmark or product demonstration without incremental revenue is not enough.

Regulation is a first-order operating constraint because the product sits close to the boundary between information and financial advice. The company carefully describes many retail functions as financial information, analysis tools, strategy support, risk alerts and decision support. As AI agents become more personalized and action-oriented, the economic value rises at the same time that the compliance boundary becomes more important. I would therefore treat the absence of separately disclosed “AI advisory revenue” as prudence rather than evidence that regulation does not matter.

The enforcement history is concrete. On 2023-06-30, the Zhejiang CSRC issued 《关于对浙江同花顺基金销售有限公司采取出具警示函措施的决定》 (“Decision to Issue a Warning Letter to Zhejiang Tonghuashun Fund Sales Co., Ltd.”). The regulator found that some products promoted expected returns in violation of fund-sales rules, issued a warning letter and entered the measure in the securities-and-futures integrity record. This was directed at the fund-sales subsidiary rather than an accounting problem at the listed company, but it demonstrates that marketing language around investment products receives direct regulatory scrutiny.

Data compliance is equally central. The 2025 annual report itself identifies the Cybersecurity Law, Data Security Law and Personal Information Protection Law as part of an increasingly demanding regime governing collection, storage, processing, sharing, cross-border transfer and user authorization. Hithink’s advantage depends on large quantities of financial and behavioral data, so compliance spending is economically similar to maintenance capex: it is a recurring cost of preserving the franchise.

AI increases that burden for institutional customers. In June 2026, China’s National Financial Regulatory Administration issued guidance for safe AI development and use by banks and insurers, requiring lifecycle governance, model-risk controls, approval of high-risk applications, human intervention and stronger data/privacy safeguards; externally introduced generative models must satisfy relevant filing requirements. The rule applies directly to bank and insurance institutions rather than automatically to Hithink’s retail app, but it raises the procurement and validation bar for vendors selling AI technology into those customers.

The competitive set is broader than “financial-data companies.”

East Money, 300059.SHE, is the most important economic comparator because it chose the path Hithink did not. Its original financial portal grew into a vertically integrated retail financial-services ecosystem with East Money Securities and Tiantian Fund. In 2025 it reported CNY16.07bn of total operating revenue and CNY12.09bn attributable profit; its securities subsidiary handled CNY38.46tn of stock-and-fund transactions, while Tiantian Fund reported CNY2.606tn of fund sales. Q1 2026 total operating revenue rose 44.3% to CNY5.03bn as brokerage commissions and margin-financing interest increased.

East Money became a retail-finance balance-sheet platform. Its free information portal acquires users; brokerage accounts capture trading; margin finance captures interest; fund distribution captures assets. The result is deeper monetization per financially active user and a natural reason to keep basic data free. Hithink’s counter-position is independence, higher information-service margins and less capital consumption. In a trading boom East Money captures more layers of the economics; in a regulatory or credit event Hithink carries much less financial-balance-sheet risk.

Compass, 300803.SHE, is a closer comparator for paid retail software. It sells tiered securities-analysis products directly to individual investors and then acquired Maigao Securities, using the paid-information relationship to feed a brokerage. Its own filings say financial-information revenue is highly related to securities-market activity and that premium products are sold in concentrated campaigns rather than evenly through the year. In 2025 revenue rose 40.4% to CNY2.146bn, with financial-information revenue of CNY1.509bn; attributable profit rose 118.7% to CNY228m. Its brokerage subsidiary earned CNY165m, meaning much of group profit is already coming from the newly integrated securities operation. Q1 2026 attributable profit fell to CNY111m from CNY139m a year earlier, illustrating its much more uneven sales pattern.

Compass became a high-touch, high-ARPU retail-software sales organization attached to a brokerage. Its advantage is monetizing a smaller set of users at much higher ticket sizes and then moving them into securities accounts. Its weakness is a far heavier selling burden: the 2025 parent-company statement alone shows CNY1.184bn of sales expense on CNY1.625bn of parent revenue. Hithink has a much larger free traffic funnel and more advertising monetization, which makes its economics less dependent on concentrated premium-product campaigns.

Hundsun Technologies, 600570.SHG, occupies the opposite end of the spectrum. It became an enterprise financial-technology infrastructure vendor: broker core trading systems, asset-management technology, transfer-agent systems and other mission-critical institutional software. Its 2025 R&D expense was CNY2.18bn, 37.7% of revenue, implying about CNY5.78bn of revenue, and it employed nearly 7,000 product/technology staff. Q1 2026 revenue fell 11.8% to CNY905m, showing that institutional IT budgets have their own cycles, but its switching costs are fundamentally higher because the systems sit in regulated customer workflows rather than an optional retail app.

Dazhihui, 601519.SHG, remains the legacy direct retail-software reference: another well-known market-data and securities-information brand that competes for investor desktop/mobile attention. Its weaker long-term profitability and corporate changes make it a less useful valuation anchor than a competitive reminder. Hithink’s survival and monetization advantage over legacy retail software competitors is evidence that brand alone is insufficient; product iteration and traffic monetization mattered.

A narrow numerical comparison shows why peer P/E alone is dangerous.

† The cited Hundsun extract provides parent-company rather than like-for-like attributable profit, so I do not force a comparison. ‡ Compass’s quarterly revenue distribution is particularly lumpy by its own disclosure.

Sources: company filings and market data as of the latest available disclosures; Hithink and East Money P/E data are current to 2026-08-14.

East Money’s roughly 23.5 times TTM multiple makes Hithink’s roughly 52 times vendor-reported TTM multiple look expensive, but the denominator and business risks differ. East Money requires much more financial capital and is economically a broker/wealth platform; Hithink is an asset-light information platform with 90%-plus gross margin and little credit risk. A premium is rational. A premium of more than two times East Money’s earnings multiple already assumes that Hithink preserves exceptional margins and turns some AI/product investment into structurally higher normalized earnings.

Hithink’s ecological niche is distinctive: it is the independent mass-market financial-information layer between exchanges, brokers, financial-product providers, advertisers and individual investors. East Money is most capable of taking its profit pool from below by making more analytics free as part of brokerage customer acquisition. Brokerage apps can do the same with basic quotes and execution. Hundsun competes from above for institutional software budgets. General internet platforms compete for advertising attention. Hithink’s defense is that one independent destination can aggregate more market workflows and user intent than any single broker is willing or able to provide.

Current Fundamentals and Capital-Market Narrative

The last four reported quarters before the research date show increasingly strong operating leverage, but there is an important reporting gap: only Q1 2026 has a full financial statement, while H1 is a profit pre-announcement. The company has told investors how much H1 profit it expects without yet disclosing H1 revenue, segment mix, gross margin, sales spending, cash flow or institutional-versus-retail contribution. That information asymmetry is precisely why extrapolating the 75%–95% profit increase is hazardous.

The 2025 quarterly progression was unusually back-end loaded. Revenue rose from CNY748m in Q1 to CNY1.031bn in Q2, CNY1.482bn in Q3 and CNY2.768bn in Q4. Attributable profit went from CNY120m to CNY381m, CNY704m and CNY1.999bn respectively. Q4 alone supplied roughly 62% of full-year profit. Some seasonality is normal, but the scale of the Q4 contribution reinforces the need to distinguish full-year run rate from calendar-quarter annualization.

Q1 2026 then started from a much stronger base: CNY1.053bn revenue, CNY255.9m attributable profit and CNY838.8m operating cash flow. Revenue growth of 40.8% was impressive; profit growth of 112.6% showed the fixed-cost leverage. Contract liabilities rising 50% from year-end to CNY2.656bn indicate substantial prepaid demand, while the 77.7% increase in sales expense shows the company is spending aggressively to convert the active environment rather than simply harvesting it.

At the midpoint of H1 guidance, the implied Q2 attributable profit of roughly CNY672m would be nearly 76% above Q2 2025. The whole H1 midpoint of roughly CNY928m is about 85% above H1 2025. Those calculations are large enough that market activity plainly matters, because the company’s only separately reported institution-oriented line was just CNY399m of revenue in all of 2025. Even an extraordinary acceleration in institutional AI software could not readily explain the group-level profit change without material contributions from retail subscriptions, advertising and traffic monetization.

Both that earnings acceleration and the AI narrative are in the price. The latest verified close is CNY230.26 on 2026-08-14, for CNY173.3bn of equity value. The post-H1-preannouncement pro-forma P/E is about 47.7 times at the guidance midpoint, while the conventional market-data TTM multiple is 51.9 times. A separate data source reports a median P/E around 53.4 times for the 2022–25 fiscal-year period, suggesting the current conventional multiple is not extreme relative to Hithink’s recent own trading history even though it is absolutely high. Historical percentile estimates vary significantly with earnings cycle, stock-adjustment methodology and look-back window, so I would not pretend to know a precise “77th percentile” or similar number.

That historical comparison can still mislead. A 50 times P/E during the 2023 profit trough capitalized CNY1.4bn of depressed earnings; a 48–52 times multiple today capitalizes a trailing earnings denominator already above CNY3.6bn after incorporating H1 guidance. The same multiple applied to 2.5 times as much profit implies a much larger absolute market expectation. P/E percentile and earnings-cycle percentile have to be read together.

The stock’s recent re-rating began before the current earnings figures were visible. From the late-September 2024 policy-driven A-share recovery into early 2025, a cited market comparison put Hithink’s gain above 200%, ahead of East Money, Compass and Dazhihui. That move coincided with a sharp improvement in securities-market sentiment and an AI re-rating of financial-technology names. Subsequent 2025 results validated a large part of the earnings recovery, but the sequence matters: multiple expansion anticipated the profits rather than merely following them.

The bull case rests on a structural-floor argument. It says the 2022–23 investment period built a materially better product; iWenCai and AI assistants increase retention and paid conversion; institutional agents and iFinD can grow faster than traditional software; the enormous retail traffic pool gives Hithink a low-cost distribution channel for new products; and active capital markets may persist as Chinese household financial assets shift toward securities and funds. Evidence for this view includes 2025’s 20.7% growth in paid value-added telecom revenue, Q1’s 50% rise in contract liabilities and sustained institutional AI deployment.

The bear case rests on denominator normalization. Advertising supplied 57% of 2025 revenue and 78% of its incremental revenue. Institutional software was only 6.6%. The last downturn already showed that value-added telecom, software and fund fees can fall together while R&D remains high. If securities activity normalizes, Hithink may retain better products but still earn materially less than the current trailing run rate. The stock then faces both an earnings decline and a multiple reset.

I place more weight on the bear side of that specific disagreement about the composition of current growth. The filings prove structural product improvement; they do not prove structural replacement of cyclicality. A different conclusion would require the full H1 report to show that institutional software/AI is accelerating far faster than 2025, or that paid-subscription growth remains strong even after controlling for market turnover. Neither data point was available as of 2026-08-15.

The expected next information point is the 2026 interim report. Public pre-disclosure calendars indicate an expected 2026-08-22 release date; this is a scheduled date rather than a guarantee and should be rechecked against the final CNINFO filing. The market should care less about whether profit lands at CNY900m or CNY950m than about H1 revenue, the four-line revenue mix, contract liabilities, sales expense, institutional software growth and operating cash conversion.

Valuation, Risks, Catalysts and Tracking

Valuation has to begin with cash passthrough rather than EPS. Over 2021–25, operating cash flow was about 115% of attributable net profit in aggregate. The ratio ranged from roughly 1.03 times to 1.27 times annually and was 1.18 times in 2025. That removes one common high-multiple risk: accounting profit has not been persistently running ahead of cash.

The capex adjustment does not change the picture much. Total disclosed 2025 capex for fixed assets, intangibles and other long-lived assets was only CNY142m. Using the strictest cash treatment and deducting all of it from CNY3.774bn OCF gives CNY3.632bn of free cash flow, a 2.10% FCF yield on the current CNY173.3bn market capitalization, or about 47.7 times FCF. If only half of capex is maintenance, owner cash earnings rise to roughly CNY3.70bn and the multiple falls to about 46.8 times. The gap from the headline 51.9 times P/E is around 10%. A difference that small does not justify abandoning earnings-based valuation, so I keep reported earnings as the primary anchor and treat the cash figures as a cross-check.

That 2.1% FCF yield is the most useful absolute starting point. It is only modestly above the 1.70% Chinese 10-year government-bond yield on 2026-08-14, despite equity risk, earnings cyclicality and regulatory exposure. The justification for paying this price has to be future growth and franchise durability rather than current cash yield.

Peer valuation provides only a sanity check. East Money traded at about 23.5 times TTM earnings on August 14, versus Hithink’s displayed 51.9 times. Hithink deserves a premium for its asset-light structure and absence of brokerage credit/balance-sheet exposure. Yet a greater-than-two-times P/E premium is a demanding starting point when East Money has much greater absolute earnings, a brokerage license, fund distribution and substantial exposure to the same securities-market upcycle.

My absolute valuation deliberately normalizes earnings rather than annualizing the H1 pre-announcement. The horizon is approximately three years, allowing the current cycle to pass through and AI/institutional products to reveal whether they create a higher earnings floor. “Owner earnings” below means sustainable cash earnings after maintenance investment, not the maximum accounting profit produced in an unusually active quarter.

† Before dividends. Adding plausible cash distributions raises returns modestly but does not alter the relative conclusion.

These are research-framework valuation scenarios, not investment advice. The underlying financial starting points come from company filings and the August 14 market price.

The conservative case is not a disaster scenario. CNY3.5bn of normalized owner earnings is still about 2.5 times 2023 attributable profit. A 34 times multiple remains generous relative to ordinary Chinese software companies because Hithink has high margins, no material financial debt, excellent cash conversion and a durable retail franchise. The conservatism lies in assuming that today's market-activity windfall fades while some post-AI/product gains remain permanent.

The base case requires more than another strong trading year. CNY4.4bn of normalized owner earnings means the business has established a profit floor materially above 2025’s CNY3.21bn accounting profit even after the cycle cools. For that to happen, subscription conversion and institutional software must improve enough that advertising is no longer the overwhelming source of incremental growth. A 39 times multiple then recognizes the company's unusual capital efficiency while still allowing some compression from current headline valuation.

The optimistic case requires structural and cyclical conditions to cooperate: institutional AI adoption must become financially visible, paid retail products must keep gaining monetization, and Chinese securities activity must stay favorable enough that advertising does not retrace sharply. CNY5.4bn of owner earnings at 44 times is already an aggressive outcome; the resulting CNY316 fair value is a ceiling for fundamental optimism rather than a central target.

The market’s largest expectation gap is likely to appear in segment composition rather than consolidated profit. A full H1 report showing software sales and maintenance accelerating to 25%–30% while value-added telecom remains above 20% would materially strengthen the structural case. Conversely, another period where advertising supplies most incremental revenue would make the 75%–95% headline profit growth more obviously cyclical. The next major datapoint should be read line by line rather than against the midpoint of profit guidance.

The margin-of-safety check is harsher. Current price is about 46% above the CNY158 conservative fair value. On that definition the present margin of safety is zero. The most fragile base-case assumption is CNY4.4bn of normalized owner earnings. Cutting it to 70%, while leaving the 39 times multiple unchanged, produces roughly CNY160 per share, almost exactly the conservative valuation. The base valuation is highly sensitive to whether current earnings become a structural floor.

A second check reaches the same result. The 2025 annual and interim distributions, adjusted onto today's post-transfer share count, correspond to roughly a 1.6% cash yield at CNY230.26. If earnings stay flat for three years and the valuation multiple does not expand, the shareholder’s return is approximately that cash yield before reinvestment. The Chinese ten-year government bond yielded 1.70% on August 14. On this flat-earnings test, there is no margin of safety at this buy price.

Margin-of-safety sufficiency verdict: none.

The permanent-loss risks cluster around five variables rather than balance-sheet leverage.

The highest-probability risk is market-activity normalization; probability medium-to-high, impact high. Observable indicators: falling A-share turnover, slower app engagement, contraction in advertising growth, declining contract liabilities and weaker value-added telecom revenue. The transmission is unusually direct: lower investor attention reduces premium conversion and advertising demand at the same time, while R&D and technology spending cannot be cut proportionately. The 2022–23 financial record already shows the mechanism.

The second is an advertising concentration reversal; probability medium, impact high. Advertising went from about 38% of revenue in 2023 to 48% in 2024 and 57% in 2025. At 96% gross margin it has been the most powerful source of incremental profit. If advertisers price a normalization of financial traffic before paid subscription revenue can replace it, both consolidated growth and incremental margin fall sharply. This is the risk most likely to surprise investors who still mentally model Hithink as a subscription-software company.

The third is free-product erosion of retail willingness to pay; probability medium, impact medium-to-high. East Money, broker apps and large internet platforms can subsidize basic financial information because they monetize elsewhere. Generative AI also lowers the cost of producing summaries and simple screening interfaces. Hithink must keep premium value several steps ahead of free quotes and generic AI. A falling value-added-telecom growth rate despite active markets would be the clearest warning.

The fourth is regulatory boundary risk; probability medium, impact potentially high. Personalized AI output, product recommendation, fund marketing and extensive behavioral data all sit in areas of increasing supervisory attention. The 2023 warning letter to the fund-sales subsidiary gives this risk a concrete company precedent, while the 2026 banking/insurance AI guidance shows regulators requiring human oversight, traceability, data controls and stricter treatment of high-risk financial AI. A material restriction on personalized recommendation could reduce product differentiation and force greater compliance spending.

The fifth is valuation compression; probability medium-to-high, impact high even with healthy operations. A stock at roughly 48 times pro-forma TTM profit does not need an earnings collapse to produce a poor outcome. If normalized profit settles at CNY3.5bn and the market eventually values it at 30 times rather than 40–50 times, equity value would fall toward CNY105bn, roughly CNY140 per share before recognizing excess corporate cash. The risk is multiple times earnings normalization, not just earnings normalization.

Positive catalysts are correspondingly concrete: H1/H2 software revenue growth above 20%–25%; contract liabilities remaining elevated after market activity stabilizes; value-added telecom growing independently of advertising; disclosed paid AI products or institutional contracts becoming financially material; continued OCF/net-income conversion above 1.0; and evidence that 2026 profit remains durable after the most active trading periods pass.

Negative catalysts are a sequential drop in contract liabilities, advertising growth turning negative, a sharp rise in sales expense without corresponding revenue conversion, another regulatory measure involving product recommendation or fund marketing, major founder-group selling, or an H1 report showing that essentially all incremental revenue still comes from advertising.

The “normal zones” are my analytical monitoring thresholds rather than management guidance. The financial baselines use Q1 2026 and 2025 filings; the expected interim date comes from the current public pre-disclosure calendar and remains subject to change.

Four research uncertainties remain material. First, the unpublished H1 report means there is no segment decomposition of the 75%–95% H1 profit growth. Second, Hithink does not disclose AI-specific revenue, paid AI users, ARR or unit economics, so structural AI monetization cannot be measured independently. Third, current filings do not provide a clean maintenance-versus-growth capex split; the valuation sensitivity is small because total capex is low, but the split remains estimated. Fourth, retail active-user and paid-conversion disclosure is less complete than an analyst would ideally want for an attention platform, making revenue lines more informative than user KPIs.

Cross-Synthesis Summary

Looking vertically across twenty-five years, Hithink’s proven capability is adaptation without balance-sheet damage. The company began as a securities-software supplier, survived the commoditization of basic market data, moved with users from desktop to mobile, converted a free audience into paid information and advertising revenue, entered fund distribution, built an institution-facing data/software business and invested heavily enough in AI before the generative-AI boom to have real products rather than a newly applied label. Through all of that it retained founder control, remained highly profitable and avoided material financial leverage.

Its success has nevertheless depended on an era-specific tailwind: China created one of the world’s largest, most active retail-investor populations, and those investors needed an independent information interface. Hithink captured that interface before broker apps, internet portals and smartphones made basic quotes a commodity. Management then made the right economic choice for its own strengths: instead of trying to charge everybody for basic information, it preserved a broad free funnel and monetized a small proportion through premium services while selling the rest of the attention to advertisers and financial-product providers. That is why 90%-plus gross margin can coexist with enormous free usage.

The same history explains the cyclicality. When investor attention is scarce, advertising buyers pull back, paid decision tools feel less urgent and fund activity weakens. The revenue lines look diversified in an accounting note but share a common demand factor. This is why 2022–23 matters more for valuation than a ten-year average ROE: it showed that the machine remains very profitable in a downturn, but it also showed that profit can fall roughly one-quarter from a cycle high even while revenue barely declines.

The management team’s strategic record deserves credit. It kept R&D around CNY1.1–1.2bn through the weak period rather than protecting short-term margins, then entered the recovery with HithinkGPT, iWenCai upgrades, iFinD AI functions and an agent platform already deployed inside financial institutions. The result is a better franchise today than in 2021. R&D did not suddenly appear after AI stocks became fashionable.

What remains unproven is whether this investment created a second profit engine. The 2025 accounts answer that question unfavorably so far: software sales and maintenance grew CNY43m; advertising grew CNY1.44bn. AI may have helped advertising yield and retail engagement, but investors cannot attribute that entire CNY1.44bn increase to structural AI without ignoring the concurrent securities-market recovery. The full H1 2026 report is especially important because the pre-announcement provides no segment data with which to update that decomposition.

Horizontally, Hithink’s real advantage is clearer. East Money became the brokerage-and-wealth super-app. Compass became a direct-sales premium-tool funnel attached to a brokerage. Hundsun became enterprise financial infrastructure. Hithink remained the independent consumer information and analytics destination, using advertising to subsidize reach while selling data and software into institutions. Each business model captures a different part of the same investor and financial-institution wallet.

That position is valuable because it avoids the balance-sheet intensity of brokerage and the implementation burden of pure enterprise software. It is also less protected. East Money can afford to make increasingly sophisticated tools free if they attract brokerage assets. A brokerage app can subsidize market data from commission and interest revenue. Large-model costs are falling. Hithink has to keep the difference between “information available anywhere” and “workflow worth paying for here” wide enough to protect subscription revenue. Its data history, product habit and cross-industry distribution give it a head start, not immunity.

The institution business is strategically more valuable than its 6.6% revenue share suggests because it validates technology with demanding customers and could reduce cyclicality if it grows. Yet its present size prevents it from serving as the principal valuation anchor. A 20% growth rate on CNY399m creates roughly CNY80m of incremental revenue; a 20% swing in CNY3.46bn of advertising creates roughly CNY690m. That arithmetic is why the segment mix deserves more attention than AI product announcements over the next several reports.

The strongest financial evidence is cash. Hithink does not need large working capital, inventories, factories or acquisitions to expand. Five-year OCF has exceeded five-year profit, 2025 capex was only CNY142m, and the company returned most of 2025 earnings to shareholders. A cyclical earnings mistake is unlikely to turn into a solvency problem. The permanent-loss pathway runs through paying too high a multiple for a temporary earnings level, not through refinancing distress.

That distinction is crucial at CNY230.26. The business can be very good while the stock offers little margin for an ordinary outcome. My normalized base value is essentially the current price. The conservative value is far below it, and the optimistic value gives meaningful upside only after assuming substantially higher normalized earnings and retaining a 44 times multiple. The current stock already spends a meaningful part of the future productivity gain that AI is supposed to create.

What the market is most likely misjudging is neither “AI is fake” nor “the cycle will immediately collapse.” The more subtle risk is that AI makes Hithink a better company without making today's earnings a stable base. Conversion can improve structurally while advertising still falls 30% in the next weak securities market. Institutional AI can grow 25% while being too small to offset that decline. Both statements can coexist. The market can correctly identify a technology winner and still overpay for a cyclical profit peak.

Over the next twelve months, the critical variables are H1/H2 segment mix, contract liabilities, securities-market activity and sales efficiency. The important test is whether value-added telecom and institutional software continue to grow when advertising growth slows. A persistently high OCF conversion ratio would support earnings quality, but it cannot by itself distinguish cyclical from structural demand.

Over three years, the question changes to earnings floor. I would want to see normalized owner earnings of at least CNY4bn through a quieter market environment, institutional software growing materially faster than its 2025 12.1% rate, and evidence that paid AI features command either higher pricing or higher conversion. If those conditions emerge, today's apparently high multiple could prove to be a reasonable price for a structurally stronger business.

Over five years, regulation and platform relevance dominate. Hithink must stay a destination even as brokers, East Money and general AI interfaces make basic financial information nearly free. It must also ensure that increasingly personalized AI stays within an acceptable regulatory perimeter. A platform with trusted financial data and properly governed agents could become more valuable as generic AI proliferates because accuracy, traceability and permissions matter more in finance than in ordinary web search. The opposite outcome is equally possible: generic models commoditize much of the front end while regulation prevents aggressive personalization.

The stock would become materially more attractive under either of two conditions. The first is price: a decline toward CNY115–125 would provide at least a 20% discount to my conservative CNY158 value, offering compensation for cycle uncertainty. The second is evidence: if the company demonstrates that normalized owner earnings are moving decisively above CNY4.4bn because institutional and subscription economics have structurally improved, my conservative and base earnings assumptions would need to rise. Paying more can be rational after uncertainty falls; paying CNY230 today assumes part of that proof in advance.

The thesis should be overturned positively if institutional software/AI becomes a much larger revenue contributor, value-added telecom grows robustly through a market downturn, and the company can maintain CNY4.5bn-plus owner earnings without relying on advertising growth. It should be overturned negatively if paid retail revenue falls even while markets remain active, advertising share continues rising toward two-thirds of revenue, regulation materially restricts AI investment-support functions, or founder-group selling becomes large enough to change alignment.

10.1 Bull and bear reasons:

Bull reasons:

• Hithink remained profitable and cash-generative through the 2022–23 trough, then restored 38.5% ROE in 2025 without financial leverage.
• Five-year aggregate OCF is about 115% of attributable profit, while 2025 total capex was only CNY142m, giving the franchise unusually strong cash economics.
• Value-added telecom revenue grew 20.7% in 2025 and Q1 2026 contract liabilities increased 50% from year-end, evidence that paid demand strengthened alongside advertising.
• AI investment predates the current market narrative; Hithink disclosed large-model work in 2022–23 and had thousands of institution-side agents deployed by 2024.

Bear reasons:

• Advertising reached 57.4% of 2025 revenue and generated about 78% of incremental 2025 revenue, making the current growth mix much more cyclical than a software label implies.
• The institution-oriented software line was only 6.6% of 2025 revenue and grew 12.1%, leaving no disclosed financial evidence yet for a large second AI growth curve.
• The 2022–23 downturn showed that nearly flat revenue can coexist with a roughly 26% decline in profit from 2021 to 2023 because R&D and technology costs remain.
• At CNY230.26, the stock is roughly 48 times pro-forma TTM profit and about 48 times 2025 conservative free cash flow, leaving little valuation protection against normalization.
• The fund-sales subsidiary has already received a regulatory warning over expected-return promotion, and more personalized AI raises rather than lowers the importance of the advisory, marketing and data-compliance boundary.

10.2 Pre-mortem: where might I be wrong?

The first failure script is a classic two-variable compression. During 2027, A-share trading activity normalizes after the 2025–26 burst. East Money and major brokerage apps respond by giving away more AI screening and research functions to win brokerage assets. Hithink’s advertising revenue falls 30%–35%, value-added telecom falls 15%–20%, while institutional software still grows 15%. R&D remains above CNY1.1bn and selling costs cannot fall quickly enough. Attributable profit settles near CNY2.1bn. If the market then values the stock at 35 times rather than almost 50 times, implied equity value is around CNY98 per share, roughly 57% below the current price. The business would still be healthy; the investment would have failed because both earnings and the multiple were bought too high.

The second script is regulatory/product rather than macro. During 2027–28, increasingly personalized financial agents come under stricter suitability, explainability and human-review requirements. Hithink must limit automated recommendation functions and increase compliance costs just as generic financial models make simple research and screening cheaper. Premium conversion stops improving; software growth remains around 10%; advertising continues to provide more than half of revenue. Normalized profit stalls near CNY2.5bn, and the market decides the appropriate multiple is 30–32 times rather than an AI-platform multiple above 40 times. The resulting CNY100–106 per-share valuation again implies a loss of more than 50%. The 2023 fund-sales warning and 2026 financial-AI regulatory direction make the pathway plausible, although such a rule change is not my base case.

10.3 Final research conclusion:

Hithink RoyalFlush is a rare combination: a genuinely high-quality, asset-light franchise whose earnings are nevertheless strongly cyclical. Its long history, high cash conversion, lack of financial leverage, founder continuity, data assets and retail distribution deserve a premium valuation. The recent AI work is substantive and began before the theme became fashionable. These strengths explain why I would not value the company on a trough-market P/E or compare it mechanically with a brokerage.

The current price already recognizes much of that quality. My base normalized value of about CNY228 is almost exactly the CNY230.26 market price. The conservative value is CNY158. The 2025–H1 2026 growth mix is still dominated by advertising and market activity, while the directly identified institutional software business remains small. A buyer at the current price receives a strong company but little protection from the possibility that CNY3.6bn-plus trailing profit proves cyclical. Existing ownership can be justified by franchise quality and upside from a higher structural earnings floor; fresh capital has a materially better risk/reward below the present quotation.

My final judgment is that Hithink is a high-quality cyclical platform at roughly base-case fair value, with no current margin of safety.

【Company-profile scores】

• Fundamental quality: high
• Growth: medium
• Moat: strong
• Financial soundness: strong
• Management credibility: high
• Valuation attractiveness: low
• Risk level: medium
• Suitable investor type: long-term growth / cyclical; valuation-disciplined investors able to tolerate securities-market earnings sensitivity

【Investment rating】

• Rating: Hold
• One-line thesis: High-quality cash economics and real AI assets are already priced against an earnings base still dominated by cyclical advertising and retail activity.
• Ideal buy price: see standalone line below.
• Acceptable hold price: CNY200–250, corresponding to roughly ±12% around the CNY228 base valuation.
• Clearly overvalued price: CNY350–380, beginning more than 10% above the CNY316 optimistic valuation.
• Current-price classification: acceptable hold.
• Whether to wait for a better price: yes. For new money, the preferred trigger is CNY125 or below under unchanged fundamentals, or a higher price only after evidence raises conservative normalized earnings materially above CNY3.5bn. The opportunity cost is missing an AI-driven structural re-rating if Hithink proves CNY4.5–5.0bn of normalized owner earnings before the share price corrects.
• Target holding horizon: 3–5 years.
• Expected annualized return: conservative about -10% including modest dividends; base about 1%; optimistic about 12%–13%, using a roughly three-year realization horizon.
• Max-loss risk: roughly 50%–60% in the principal pre-mortem, triggered by market normalization that takes profit toward CNY2.0–2.5bn while the earnings multiple compresses to 30–35 times.
• Reassessment-trigger signals: institutional software growth above 25% for two reporting periods; value-added telecom growth above 15% through a weak trading environment; TTM OCF/net income below 0.9; advertising declining more than 20% with no offset from subscriptions; or a formal regulatory measure materially restricting AI-based financial decision-support functions.

【Ideal Buy Price】115–125 CNY Basis: the conservative scenario produces about CNY158 of fair value; CNY125 is approximately 21% below that value, satisfying the required ≥20% margin of safety, while CNY115 provides additional protection against cycle normalization.

【Valuation Range】

• current: 230.26 (close as of 2026-08-14)
• bear (conservative · ideal buy zone): [115, 125]
• base (fair · acceptable hold zone): [200, 250]
• bull (optimistic · above the clearly-overvalued line): [350, 380]

The bands are deliberately discontinuous. CNY126–199 is below base fair value but does not meet my required conservative-case margin of safety; CNY251–349 is above the acceptable-hold range but not yet 10% above the optimistic CNY316 scenario. The valuation discipline is intended to prevent the bear, base and bull cases from being blurred into a single target.

Selected source ledger: the core primary materials were the company’s 《2025 年年度报告摘要》 (“2025 Annual Report Summary”) and full 《2025 年年度报告》 (“2025 Annual Report”), the 《2026 年第一季度报告》 (“Q1 2026 Report”), the 《2026 年半年度业绩预告》 (“H1 2026 Earnings Pre-announcement”), the 2022 and 2023 annual-report summaries, the IPO/listing materials, East Money’s 《2025 年年度报告摘要》 (“2025 Annual Report Summary”) and Q1 2026 report, Compass’s 2025 annual report/summary and Q1 2026 report, Hundsun’s 2025 annual report and Q1 2026 report, the Zhejiang CSRC’s 2023 warning-letter decision concerning Tonghuashun Fund Sales, and the National Financial Regulatory Administration’s June 2026 AI-safety guidance. Current price and valuation were cross-checked to 2026-08-14 market data; the ten-year government-bond comparison uses the ChinaBond Ministry of Finance curve for the same date.

Other tickers mentioned

• 300059.SHE — East Money, the principal mass-market comparator whose information portal is vertically integrated with securities brokerage and large-scale fund distribution.
• 300803.SHE — Beijing Compass Technology Development, a closer paid-retail-financial-software peer that has added Maigao Securities and uses information products as a brokerage acquisition funnel.
• 600570.SHG — Hundsun Technologies, the institutional financial-software comparator with much higher enterprise-system switching costs and a substantially larger R&D organization.
• 601519.SHG — Shanghai DZH, the legacy retail financial-information and securities-software competitor used as a reference for the durability of Hithink’s consumer franchise.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.