Tencent Holdings Limited

The conclusion, first

Pass — on quality; the price offers no margin of safety

Verdict: HOLD / WATCH — the achievable return now sits just below the 15% hurdle. The quality is confirmed: the people clear all five diagnostics, the moat is wide on output evidence, and the business is extraordinary. On the framework-v29 method the central case — 14× terminal, a 10-year explicit horizon to 2035, 9,250m shares — puts the achievable return at HKD 471.80 at 14.23% a year against the 15% required return, about eight-tenths of a point short. Every reasonable variation is lower still: a 20-year horizon gives ~13.40%, a top-of-range 9,520m share count 13.69%, and a same-forecast 9× terminal 11.05%; only the most optimistic durability assumption (a 20× terminal) clears 15%. So this is not a buy: Tencent is fairly-valued-to-slightly-rich against the hurdle, a HOLD / WATCH, with no margin of safety at HKD 471.80 — the price has risen about 6.5% since the prior strike, putting it further from a buy. This is the same investment call as before — the fresher, higher price only reinforces it.

The pillar scorecard, and what each score rests on

The one fact that would most change this — the falsifiable test

WeChat advertising revenue per user growing at less than 10% a year for two consecutive years. That is below even the 2021–2025 compound rate of 10.0%, on a user base that is no longer growing; it would kill the bull case and put the base case in question. The full reasoning, the runner-up metric and the other watch-triggers are in the one metric that matters, below.

What changed in this framework-v29 recast (1 August 2026). The method moved to fair-value, whole-company, return-deduced (rulings #40/#3): the whole business is valued as a fair value — the destination discounted at the long-term risk-free rate (5.2%, the 30-year US Treasury) with a 14× terminal free-cash-flow multiple, summed by parts — and the decision is the achievable return the same forecast delivers at the price, tested against the 15% required return. The method changed; the investment call did not. The prior method (a probability-weighted intrinsic value discounted at the 15% hurdle, resolving on a g-derived ~9× terminal) produced a 13.83% return — a hold. The v29 central case, on the house 14× durability convention over a 10-year horizon, gives a 14.23% achievable return at the refreshed strike of HKD 471.80 — just below the 15% bar, and lower still once the horizon (20-year → 13.40%), the share count (9,520m → 13.69%) or a 9× terminal (11.05%) are varied; only a 20× terminal clears — so Pillar 4 is scored borderline and the conclusion stays a HOLD / WATCH, the same call as before. The whole-company fair value of ~HKD 825 is context (a risk-free-discount artefact), not a margin of safety. The three v27 changes are retained: the body leads with the people (ruling #38, clears on all five diagnostics); the moat is scored on output evidence (amended #25, R&D non-disclosure re-classed to a transparency demerit under #18 / C18); the valuation carries the accurate-not-conservative lens (#39). The old FY2026 operating-margin falsifiable test remains demoted to a bear gauge (see below). Stage-4d capital-allocation retrofit (quality-side only): the People and portfolio sections now carry Tencent’s in-specie distribution mechanism (JD.com and Meituan, ~HK$255bn returned in-kind, 2021–2023), Pillar 3 is split (operating reinvestment weak; strategic investing + capital return a demonstrated strength), and investment income is softened from noise to a strategy-driven, realisable yield. No valuation number and no verdict changed — it stays HOLD / WATCH. The effect is the opposite of a buy signal: the business is even higher-quality than the operating analysis alone showed, which raises the bar for how cheap the price must be before it clears with a margin of safety.

¹ Basis & sourcing: the cash-flow forecasts and the durability assumption are analyst estimates; the 2025 balance-sheet figures (net cash, the value of the stakes) are facts from the annual report; the price and the bond yield are third-party market data. The method follows the house valuation rules (fair value from the sized destination, the 15% required return, opacity classified before it is penalised); every figure is listed in the sources appendix.

Assessed first — ruling #38

The people: a founder-run, integrity-clean, allocator-competent machine — with one concentrated point of failure

The framework leads with the people because they are the first determinant of success and the largest source of market inefficiency, and because management can create, but more importantly DESTROY, moats. The 1,418-million-node network below was not bought; it was built — and the same team that built it decides every year whether to protect it, harvest it, or spend its cash flow defending it. So the people are read first, on five evidenced diagnostics, each scored clears / ambiguous / fails from primary filings rather than from reputation.

The controllers, in the filings — who actually decides

The board is deliberately small: eight directors — one executive (Ma Huateng), two non-executive and five independent (FACT, FY2025 corporate-governance report). The two non-executives are the Prosus/Naspers seats: Jacobus (Koos) Bekker, non-executive chair of Prosus N.V., and Charles Searle, chief executive of Naspers Internet Listed Assets — the through-line to the largest outside owner.

The ownership picture is now on the primary record, not on recollection. Prosus/Naspers holds 2,079,512,000 shares — 22.80% — through MIH Internet Holdings B.V., a wholly-owned subsidiary of Prosus N.V. controlled by Naspers Limited (FACT, FY2025 annual report — this upgrades the earlier “~24% third-party recollection” to a disclosed figure, and softens compromise C11). Ma Huateng holds 8.82%. The total number of issued shares at 31 December 2025 is disclosed as 9,120,235,999. The auditor is PricewaterhouseCoopers.

The combined chairman-and-chief-executive role, weighed honestly. Tencent discloses the deviation from Code provision C.2.1 and explains it — in a “fast and rapid” market the single leader “must be technically sophisticated and sensitive to… market changes” — and lists the mitigations actually used: major decisions taken in consultation with the board and committees, and a FY2025 chairman-with-independent-directors meeting held without other directors present (FACT, FY2025 AR). This is founder concentration disclosed and reasoned, which is the integrity-preserving version of a governance compromise, not the concealed version. It is scored a named compromise, not a fail.

Capital allocation, the allocator’s scorecard — the share count genuinely shrinks

Tencent returns capital three ways — buyback, dividend, and, its differentiator, in-specie distributions of appreciated stakes — and the first is unusually well evidenced. In FY2025 the company repurchased 153,415,000 shares for approximately HKD80.0 billion, all of them cancelled, “for the enhancement of shareholder value in the long term” (FY2025 annual report). The test that matters is not the gross amount repurchased but whether the count actually falls after new shares issued to employees are counted — and it does: the weighted-average share count fell from 9,491.9 million in the first half of 2024 to 9,304.9 million in the first half of 2025, a reduction of 1.97%, derived from the reported earnings-per-share arithmetic in the interim reports. Repurchases run continuously back to at least the first half of 2008, when 2,772,600 shares were bought in. Dividends have been paid every year since FY2003; the FY2025 final dividend is HKD5.30 per share (FY2024: HKD4.50).

The promise audit on this leg, and it is the one management passes cleanly. On 14 May 2024 management guided that it would “repurchase over HKD100 billion of our shares in 2024”. About 307 million shares, roughly HKD112 billion, were actually repurchased in FY2024 (the cash-flow statement shows payments for repurchase of RMB102,331m that year against RMB73,312m in FY2025). A forward commitment made in a numeric form and then over-delivered is the strongest kind of management claim available, and it is why capital-return statements are weighted up in the destination work below while the strategic claims — a separate artificial-intelligence revenue line, Video Accounts as a business — are weighted down: those have produced no disclosed metric in three years.

The third mechanism is the one almost no peer has: distributing appreciated minority stakes to shareholders in-kind, at scale. Twice in eighteen months Tencent handed listed holdings straight to its own shareholders rather than selling them and recycling the cash. JD.com (declared 23 December 2021): 457,326,671 JD shares, some 86.4% of the holding, one JD share for every twenty-one Tencent shares — cutting the stake from roughly 17% to about 2.3% (share count and ratio: tier-1, Tencent HKEX announcement); the block was worth about HK$127.7 billion at the distribution-date market price (a market-price-derived value, ~US$16.3bn per Forbes/CNBC — the announcement itself states no HKD amount). Meituan (declared 16 November 2022, dispatched March 2023): about 958 million of roughly 1,055 million shares, a deemed-disposal gain of RMB106.6 billion, cutting the stake to about 1.7% (tier-1, Tencent HKEX announcement; FY2022 annual report). Together roughly HK$255 billion (a market-price estimate, not an announced figure) was returned in-kind across 2021–2023 — larger than several years of buybacks combined, tax-efficient for holders, and genuinely rare: it is best-in-class capital-return behaviour, and it is the proof that the minority portfolio has a realisation mechanism, not just a mark.

The percentages rest on a market value of HKD4,364.2 billion, which is 9,250 million shares at HKD 471.80, the previous trading-day close as at 1 August 2026 (the yields fall slightly from the prior HKD443.00 strike purely because the price rose). The per-share valuation share count is derived, not the disclosed year-end figure. It comes from dividing reported profit by reported earnings per share in the interim filings (a weighted-average diluted basis); the FY2025 annual report separately discloses 9,120,235,999 issued shares at 31 December 2025, which reconciles with a higher weighted-average-diluted count earlier in the year given the buyback. Three widely-quoted third-party counts (8.90bn, 9.37bn and 9.52bn) were rejected. A 4% error in the share count is a 4% error in every per-share figure on this page, and it is flagged rather than buried (compromise C15). The 1.83% repurchase percentage (buyback value over the higher market value) is independently corroborated by the 1.97% observed net reduction in the share count, which is price-independent.

Before → after: the 2014 share split, and the false alarm it creates

Read the per-share dividend history without adjusting for corporate actions and it looks as though the dividend was cut by half in the middle of the last decade. It was not. In May 2014 Tencent carried out a one-for-five share split — each share became five — so every per-share figure before that date is on a base five times larger than every figure after it. Adjusted for the split, the record is close to monotonic across twenty-two years: the FY2016 figure of HKD0.61 is equivalent to HKD3.05 on the pre-split base, well above FY2013’s HKD1.20. An unadjusted reading manufactures a false integrity flag where there is none.

Reads on the thesis: the people clear all five diagnostics, which is why the constraint is price, not quality — the hold/watch call is a valuation call, not a doubt about the business. The one People-driven way the thesis breaks is a succession or culture change at the founder-run business that built the moat — a first-order event, carried in the risks and monitoring sections below.

The engine

What the business is, and how the money actually arrives

Tencent Holdings Limited listed on the Hong Kong Stock Exchange on 16 June 2004. Underneath everything it reports sits one asset: a messaging and identity network. Weixin in mainland China and WeChat outside it are a single combined service with 1,418 million monthly active users at the end of 2025 (Tencent 2025 Annual Report) — people who message each other two ways, under real identities tied to a payment instrument, inside a graph dense enough that leaving it means leaving your friends, your building’s notices and your payment history behind.

The user pays nothing for that network. The money arrives from three other directions, and the company reports them as three segments.

Value-added services — RMB369,281 million in FY2025, 49% of revenue, up 16% — is discretionary consumer spending: items and content inside games, plus music, video and social subscriptions. This is the profit engine. Domestic games alone were RMB164.2 billion, up 18% (FY2025 annual results announcement, 18 March 2026). The company does not publish a games margin, which is a real limit on what any outsider can say about the franchise; it is recorded here as a gap rather than estimated.

FinTech and business services — RMB229,435 million, 31% of revenue, up 8% — is what merchants and enterprises pay for: the WeChat Pay payment rails and the cloud-computing business. No payment volume, no cloud revenue and no segment margin is disclosed, so this segment is measurable only by its total and its growth rate.

Marketing services — RMB144,973 million, 19% of revenue, up 19% — is advertising: what brands pay to reach the attention the network holds. It is the fastest-growing of the three, and the significance is that it is growing without any more users to grow into; the growth is coming out of intensity of monetisation per person rather than out of reach. It is also, for that reason, where the one metric that matters is drawn from.

Group profitability, FY2025 (all from the 2025 Annual Report). Gross profit RMB422,593 million, a 56.2% gross margin — up 331 basis points (3.31 percentage points) from FY2024’s 52.90%. Operating profit RMB241,562 million, a 32.1% operating margin. Profit attributable to shareholders RMB224,842 million, a 29.9% net margin. Operating cash flow was RMB303,052 million, up 17.2%, against capital expenditure of RMB79,198 million — 10.53% of revenue, down from 11.6% the year before. Net cash (cash of RMB141,041m plus term deposits of RMB353,837m, less borrowings of RMB250,987m and notes payable of RMB136,746m) was RMB107,145 million at 31 December 2025, against RMB76,798 million a year earlier, and rose again to RMB146.9 billion at 31 March 2026 (first-quarter 2026 results, 13 May 2026).

Free cash flow, the definition used throughout this page. Operating cash flow minus capital expenditure: RMB303,052m − RMB79,198m = RMB223,854 million. Both are reported lines, and this construction makes the artificial-intelligence capital-expenditure burden visible — which is why the old accrual owner-earnings construction is recorded as compromise C19. One caveat, named: Tencent’s own free-cash-flow definition subtracts two further recurring cash outflows — payments for media content and lease liabilities (fourth quarter 2025: FCF RMB34.0bn = operating cash flow 66.5 − capex 22.4 − media content 8.1 − leases 2.0). On the issuer’s own definition full-year free cash flow is roughly RMB183 billion, about RMB40 billion a year below the operating-cash-flow-minus-capex figure used here. This is the optimistic edge of the free-cash-flow input: applying Tencent’s own stricter definition lowers the fair value and pulls the achievable return at HKD 471.80 below the 14.23% central case, further under the 15% bar — one more reason the central return sits below the hurdle. It is carried as a sensitivity in the valuation section.

One structural point that matters for reading everything below. Tencent’s income statement places its share of profits from associates and joint ventures below the operating-profit line — confirmed in the FY2012, FY2013 and FY2025 income statements. Operating profit therefore contains no earnings at all from the large minority-stake portfolio described further down, and the two can be looked at side by side without counting anything twice. The company makes the point itself: excluding the share of associates’ profits from both periods, its second-quarter 2025 non-IFRS attributable profit would have risen 20% rather than the reported 10% (second-quarter 2025 results, 13 August 2025).

Reads on the thesis: roughly half the company — marketing services plus fintech and business services, 49.8% of FY2025 revenue — is a geared play on Chinese nominal economic activity, in management’s own words (second-quarter 2022 results: “we generate approximately half of our revenues from FinTech and Business Services as well as Online Advertising that directly contribute to, and benefit from, overall economic activity”). That is not a moat question; it is a market exposure the destination has to carry.

The full history

Twenty-five years of revenue, every year on the record

The point of running the whole arc rather than the customary five or ten years is that a short window crops out exactly the events that tell you what a company is: the two state-imposed ruptures, the single revenue decline, and a restatement made in the first year an accounting standard applied. Revenue grew from RMB49.1 million in 2001 to RMB751,766 million in 2025 — a 15,318-fold increase, roughly 49% a year compounded over twenty-four years. Measured from the first post-listing year instead, FY2004 to FY2025, revenue rose 657 times (36.2% a year) and profit attributable to shareholders rose 510 times, from RMB441.1 million to RMB224,842 million (34.6% a year).

Where each figure comes from. 2001–2005 from the five-year summary in the FY2005 annual report; 2006–2010 from the FY2010 annual report; 2011–2015 from the FY2015 annual report; 2016–2020 from the FY2020 annual report; 2021–2025 from the FY2025 annual report. Tencent was private until June 2004, so 2001–2003 exist only as retrospective comparatives inside a later filing, never as a contemporaneous annual report. Overlapping years were reconciled across filings and agree exactly — for example the FY2005 and FY2008 annual reports both give 2004 revenue as RMB1,143,533 thousand, and the FY2010 and FY2013 annual reports agree on 2009.

The standard ten-year measures, FY2016 to FY2025

The five headline measures on one page, each taken from a five-year summary printed by the company itself, so the ten years are covered by two filings that overlap and agree. The trend lines further down this page are drawn straight from this table and from nothing else.

Sourcing, cell by cell. Revenue, gross profit and profit attributable for 2016–2020 come from the five-year summary in the FY2020 annual report and for 2021–2025 from the five-year summary in the FY2025 annual report. Operating profit for 2019–2023 comes from the five-year summary in the FY2023 annual report and for 2021–2025 from the FY2025 annual report; the three overlapping years (2021, 2022, 2023) agree exactly across the two filings. Dividend per share is the final dividend recommended in each year’s own annual report, each one independently confirmed as the prior-year comparative in the following year’s report. Why operating profit is “n/d” for 2016–2018: from FY2023 Tencent reclassified interest income and net gains and losses from investments to below the operating-profit line and restated the comparatives only as far back as 2019 — the cause is printed in Note 2.2 of the FY2023 annual report, and the restatement of FY2022 alone was RMB124,879 million. The operating profits published contemporaneously for 2016, 2017 and 2018 (RMB56,117m, RMB90,302m and RMB97,648m) are therefore on a wider definition; splicing them onto the restated years would manufacture a break in the series, so they are left as an explicit gap. The dividend column is in Hong Kong dollars because that is the currency the company declares in; the rest of the table is renminbi.

Before → after: a restatement the company made against itself

In its FY2004 annual report Tencent reported 2004 net profit of RMB446,702 thousand and gross profit of RMB727,317 thousand. In the FY2005 report those same 2004 figures appear lower — RMB441,119 thousand and RMB725,408 thousand — with the cause printed on the page: “The adoption of IFRS 2 requires retrospective application to all share options granted to employees after 7 November 2002 and not vested as at 1 January 2005.” A downward restatement, disclosed with its reason, in the first year the standard bit. It is a small number and a large signal about disclosure behaviour.

Where the revenue sits

Segment mix, growth and the margin question the company does not answer

Every figure in this section is from the Tencent 2025 Annual Report and the FY2025 annual results announcement of 18 March 2026.

The four lines add exactly to RMB751,766 million. The margin column is left as an explicit “not disclosed” on every row rather than filled with an estimate: no segment operating profit appears anywhere in the twenty-two annual and twenty-two interim reports held. That absence is the single largest measurement gap on this company, because it is what would settle whether the payments-and-cloud segment is diluting a very high-margin games business or whether the games business is simply not as profitable as assumed.

Two precision points, because the loose version of each is wrong. First, marketing services is the fastest-growing segment at +19%, but it is not the only line growing faster than the group: value-added services grew 16%, also ahead of the group’s 13.9%. Second, FinTech and business services at +8% is the slowest of the three main segments — the small “others” line grew more slowly still, at 4%. Both corrections come from the table above.

Before → after: the margin, and what it is doing

The significance of the margin series is that it is not a growth series. Read the four years together and the profit growth of the last three years turns out to be mostly a recovery, not an expansion.

From FY2022 to FY2025 revenue rose 35.6% while operating profit rose 117.9% and the margin widened by 1,214 basis points. Before: a reader taking FY2025’s headline profit growth at face value would extrapolate mid-teens profit growth. After: around two-thirds of the three-year profit growth is attributable to the margin returning from a crackdown-and-pandemic trough, which is a one-time repair. The margin cannot travel another twelve points from 32.1%, so the honest forward arithmetic is roughly revenue growth plus a fraction of a point of margin drift, not the recent profit growth rate repeated.

Reads on the thesis: this is the arithmetic that forces the destination work below. If the margin cannot repeat its repair, the only remaining sources of value are the growth of each vector and what artificial-intelligence spending does to the margin from here.

The moat, read on output

Wide, and measurably no longer widening

The three advantages below are each tested in three steps: name the mechanism, measure it, then check whether rivals are closing the gap. Sources are the Tencent 2025 Annual Report, the NetEase Annual Report 2025 and the Kuaishou Annual Report 2025.

Three competitive advantages are usually claimed for this company: the messaging network, the games franchise, and a payments-into-cloud flywheel. Each has to be named, measured, and then checked against what a rival’s filing shows. The results are uneven.

The network — width proven, widening disproven. The mechanism is the strongest kind there is: a two-way messaging graph under real identities, bound to a payment instrument, dense with mutual connections and embedded in mini-programs, official accounts and payment rails. The number is now on the record and it cuts both ways: 1,418 million combined monthly users is an enormous, verified moat — but it grew 2% in the year and 0.3% in the quarter, the twelfth consecutive year of deceleration.

Each row’s prior-year comparative, where the filing prints one, chains exactly to the previous year’s reported level across independent annual reports — a self-reconciling series. Where the filing states a rounded percentage and the arithmetic gives a slightly different one, both are shown rather than one being silently preferred.

Why the level matters. 1,418 million against a Chinese population of roughly 1.41 billion means the network has essentially run out of new people in its home market. Everything additional must now come from monetising each person harder rather than adding people — which is exactly what advertising revenue growing 19% on a flat user base looks like. That is a genuine form of deepening; it is not the same thing as the moat getting wider.

The games franchise — durable, not demonstrably widening. The real switching cost in a live-service game is the friend group inside it, not the game. But value-added services bundles domestic games, international games, music, video and social subscriptions into one line with no games split and no games margin, so the franchise cannot be measured directly. What can be measured is the rival: NetEase’s games and related value-added services were RMB92.1 billion of its RMB112.6 billion of revenue, its games segment grew 13.4% in the second quarter of 2025 (RMB22.8bn against RMB20.1bn), and it does so at a 64.3% gross margin against Tencent’s 56.2%.

The payments-and-cloud flywheel — not visible in any disclosed number. The claim is that ubiquitous payments generate merchant relationships and transaction data, which pull through cloud and business services. The segment that would show it grew 8%, the slowest of the three main segments, and Tencent discloses no payment volume, no cloud revenue and no segment margin. A flywheel that is the slowest-turning part of the machine is not evidenced as a flywheel. This is stated as an absence of proof, which is what it is.

The rival nobody can measure. ByteDance, which owns Douyin, is private, files nothing, and appears in no exchange archive. It is the main competitor for Chinese attention and advertising money, and no filing anywhere on earth allows a check on whether the gap is opening or closing. Kuaishou, the listed short-video company, reported 410 million daily active users on a one-way following graph — a different metric on a structurally different kind of network, and deliberately not placed in the same column as a two-way monthly figure.

The scaled-innovation engine — input undisclosed, output read instead. The framework also asks whether scale funds a widening research-and-development lead over rivals — the scaled-innovation treadmill. Tencent does not disclose research and development as a separate line; it is folded into general and administrative expenses, and has been for twenty-two years, and it is the only member of its peer set (NetEase, Kuaishou and Alibaba all disclose theirs) that does not. Under the amended #25 this does not make the innovation moat “unmeasurable” and does not void the moat verdict: the engine is read on its outputs — Weixin was built internally rather than bought, its iteration cadence and delight are what produced a 1,418m graph, and the ad-monetisation output is compounding at 16.7% a year. The undisclosed input is re-classed as a transparency demerit under #18 (compromise C18), and it matters more each year AI spending rises. It is set out in full at the one metric that matters.

Reads on the thesis: the moat is a network moat evidenced by its outputs — the graph, the engagement and the rising advertising ARPU. The technology input is undisclosed, but that is a transparency flag on the page, not an absence of moat.

Future 1 of 5

Games — a 47% share of a market that has stopped adding players

What it is worth today. About RMB242 billion, 32% of group revenue: domestic games of RMB164.2 billion, up 18% (fact, FY2025 annual results announcement) plus international games of RMB77.4 billion, up 33% (32% on a constant-currency basis; fact, FY2025 annual results announcement, business review). The full-year figure is disclosed in the annual report’s business review, beside domestic games of RMB164.2 billion and social networks of RMB127.7 billion — the three summing to the RMB369.3 billion value-added-services line.

The size of the market it plays in. China’s domestic games market sold RMB350.79 billion in 2025, up 7.68%, to a player base of 683 million, up 1.35% — both described as record highs by the industry association that compiles them. That is an industry anchor, not a filing, and it is not linked because no primary URL can be stamped on it. The single most important thing in those two numbers is that they disagree. A market growing 7.68% on a player base growing 1.35% is a market growing almost entirely by charging its existing players more. That is the same signature as Tencent’s own user saturation, one layer up, and it has a mathematically shorter runway than growth that comes from adding players.

Why Tencent wins in it — argued from its own moats, not asserted. First, distribution: Weixin and QQ are the discovery and social layer for Chinese mobile gaming, which is the network moat pointed at games and the one advantage rivals cannot copy. Second, evergreen economics: Honour of Kings and Peacekeeper Elite are ten-year franchises still named as growth drivers in the first, second and third quarters of 2025. The switching cost in a live-service game is the friend group and the account history inside it — a cost of disentangling, not of rebuilding, which is the durable kind. Third, owned overseas studios — Supercell, Riot and the PUBG MOBILE and Brawl Stars portfolio — give an international leg that does not depend on the China regulator.

The counter-case, written with the same care. Tencent’s recognised domestic games revenue of RMB164.2 billion is 46.8% of that RMB350.79 billion market. (The two are not strictly like for like — recognised revenue against market sales — and true share on gross receipts is higher, not lower.) A share near a half is not a share-gain story; it is a market-growth story. Tencent cannot meaningfully take more of China; it can only grow with a market compounding at 7.7% and decelerating. Meanwhile NetEase — the credible number two, one-seventh the size — grew games and related services to RMB92.1 billion and earns a higher return on equity, 22.0% against 21.1%.

The single deciding uncertainty in this vector: can Tencent create a new evergreen franchise, or is it harvesting two ten-year-old ones? Delta Force, named as a driver in the second quarter of 2025, is the live test case. If the answer is “harvesting”, domestic games is a slow-decay annuity and the base case above is generous.

What to watch, on a one-to-three-year view. (1) Domestic games gross receipts growth rather than recognised revenue — the company’s own releases repeatedly flag the gap. (2) Whether Delta Force is still named as a growth driver in the FY2027 releases, or has faded. (3) Whether Honour of Kings and Peacekeeper Elite are still described as growth drivers rather than stable contributors. (4) International games growth, stripped of currency movement, staying at or above 10% for four consecutive quarters. Reads on the thesis: this vector supports the destination but cannot drive it — there is no share left to take at home.

Future 2 of 5 — the highest-value vector, and it gets the most depth

The WeChat advertising engine — taking share, at 28% of Meta’s rate

Where it stands. Marketing services revenue was RMB144,973 million in FY2025, up 19%, against a China digital advertising pool this analysis estimates at roughly RMB1.0–1.1 trillion growing at 8–10%. That is a share of about 13.8%. The verified fact is the +19% growth itself; the share-gain reading is analyst inference, and it is the single most bullish item in the analysis if it holds. It holds only on this analysis’s own estimate of the pool growing 8–10% a year: both third-party vendors cited elsewhere on this page put the China digital-advertising pool growing at 17–18%, and at that rate +19% is roughly in line with the market rather than share-taking. The claim is therefore flagged, not asserted as fact.

The market-size numbers are weak and are handled as such. One vendor puts China digital advertising at about USD 145.4 billion by 2030; another puts it at USD 141.0 billion in 2025 rising to USD 266.6 billion by 2029. The two disagree by roughly 80% on the same horizon. This analysis therefore uses the lower anchor and grows the pool at 4.5–6% nominal — deliberately below both vendors, because an advertising market cannot durably outgrow the economy it advertises into. Neither vendor is linked: neither is a primary source.

Why Tencent wins here, from its own moats. First, advertising load is a dial Tencent controls and has deliberately kept low. The clean way to see the headroom is revenue per user — the measure the industry writes as ARPU, average revenue per user, and which is used throughout this page in its expanded form. Marketing services revenue divided by combined Weixin and WeChat monthly users is RMB102.2 per user per year in FY2025, against roughly RMB360 for Meta worldwide. Tencent monetises its network at about 28% of Meta’s rate — the only place in this business where a tripling is arithmetically available without acquiring a single new customer. Second, the transaction closes inside the app: Mini Programs mean the click, the checkout and the payment happen on Tencent rail, an attribution advantage ByteDance structurally cannot match for the same merchant. Third, artificial intelligence as a targeting multiplier — the company subtitled its second-quarter 2025 results “Utilising AI in Games and Marketing Services”. That is a management claim; the +19% outcome is consistent with it, though it does not prove causation.

The counter-case, at equal weight, and it is a governance argument rather than a market one. Advertising load has a hard ceiling, and it is not Meta’s. WeChat’s franchise rests on being the utility Chinese people must use — messaging, identity, payment, civic services. Filling it with Meta-density advertising is precisely the harvesting behaviour the framework’s second gate prohibits, and management has said so in effect for a decade by keeping load low. So the Meta comparison sets an upper bound Tencent may be structurally unable — and unwilling — to reach. The honest reading is that RMB360 is a ceiling that will never be touched, and even the bull case should assume something like RMB230–270 by 2035.

The single deciding uncertainty: whether advertising revenue per user can keep compounding at low-to-mid teens on a user base that is not growing, without the advertising load itself becoming the harvest. The whole destination hangs on that question, which is why the one metric that matters is drawn from this vector and from nowhere else.

What to watch. (1) Advertising revenue per user, year on year — the primary metric, with its break trigger set out below. (2) Whether Tencent ever begins disclosing Video Accounts advertising revenue or daily users — the disclosure itself would be a bullish signal; continued silence is an opacity flag. (3) Marketing services growth against the estimated pool for four consecutive quarters. (4) The gross margin of the marketing services segment, which is not disclosed today and is a named gap. Reads on the thesis: this is the vector that decides whether the destination is the base case or the bear.

Future 3 of 5

Payments, cloud and business services — the slowest-growing third of the company

The number that matters more than any market-size estimate: this segment grew 8% in FY2025 while the group grew 13.9% and advertising grew 19% (2025 Annual Report). FinTech and business services is 30.5% of Tencent’s revenue and it is growing at 58% of the group’s rate. Eight percent is not a cloud-boom growth rate. It is a regulated-utility growth rate, and it has been telling us what this segment is for several years.

Why the right-to-win is weakest here. In payments, Weixin Pay is genuinely a duopoly rail alongside Alipay — but the take-rate is administratively influenced rather than competitively earned, and the state is an explicit co-claimant. A duopoly whose price is set by the regulator is a utility, not a moat that widens. In cloud, Tencent is not the China leader, and the only hard product evidence in the filings held is thin and stale: the second quarter of 2022 reported TDSQL database revenue up 30% and above 5% of cloud revenue, with a third-party study calling it the leading distributed database in China. That claim has been re-checked and fails: the study is four years old and undated. It is retired from use rather than repeated. Tencent discloses no cloud revenue figure at all — a named gap.

Market-size anchors for this vector are worse than for advertising and are recorded rather than relied on: one forecaster puts China’s hybrid cloud market at RMB337.0 billion by 2030 at an 18.4% compound rate; another puts total China cloud computing at about USD 140.5 billion by 2030 but China public cloud at only about USD 30.3 billion — a 4.6-fold inconsistency inside one vendor’s own product set. Nothing on this page is built on any of them, and none is linked because none is a primary source.

The single deciding uncertainty: whether demand for artificial-intelligence inference converts Tencent Cloud from an 8%-growth also-ran into a 20%-growth franchise — or whether it merely gives Tencent a reason to buy graphics processors for a business that will not earn their cost of capital. That is also the hinge of the next section.

What to watch. (1) Segment growth breaking above 12% for two consecutive quarters. (2) Any disclosure of cloud revenue as a standalone line. (3) Any change to the regulated payment take-rate. (4) Any segment gross-margin commentary. Reads on the thesis: this vector is where the capital-cycle argument below bites hardest — it is the commodity third of the company.

Future 4 of 5 — the vector the previous valuation did not charge for

Artificial intelligence: an explicit cost, and a possible new leg

Artificial intelligence is modelled here in both directions, because the evidence says it acts in both at once and the two effects sit in different places in the profit-and-loss account. Treating it only as a cost is the bear’s error; treating it only as a new leg is the bull’s.

The cost, sized — and this is where the ratio matters more than the amount

FY2025 capital expenditure was RMB79,198 million. On its own that is a record and tells a reader almost nothing. As a ratio it is 10.53% of revenue, down from 11.6% the year before — and the sustainability read is the point: a company spending a tenth of its revenue on capacity while its operating margin rises is not on a treadmill. What makes it a live question is that management has spoken of capital expenditure in the “low teens” as a percentage of revenue, and Tencent came in below its own guidance.

Why it under-spent, and why that is not reassuring. The reason given by the company’s chief strategy officer is chip supply, not restraint: he pledged “a substantial increase” in 2026 capital spending, “especially in the second half, as more China-designed AI chips become available”. Management has separately guided artificial-intelligence product investment above RMB36 billion in 2026 — more than double the RMB18 billion spent in 2025 on the HunYuan model and the Yuanbao assistant. As ratios of FY2025 revenue those are 4.8% and 2.4% respectively. One third party models group capital expenditure reaching RMB165 billion by 2027. All three are management claims or third-party interpretation, reported in the press rather than in a filing; none is linked, because no primary document can be stamped on them.

The possible new leg — and the honest verdict on it

Tencent monetises artificial intelligence in three observable places, and none of them is a separate revenue line: advertising targeting (the +19% above), the cost of producing game content, and cloud inference. Yuanbao, the consumer assistant, has no disclosed revenue, no disclosed users and no disclosed engagement anywhere in the filings, despite RMB18 billion of 2025 spending pointed at it. That is a named gap and an opacity flag.

The judgement made here, stated plainly: artificial intelligence is not modelled as a fourth revenue leg, because there is no disclosed number to build one on and the framework forbids filling a gap with a vendor estimate. It is modelled as (a) an explicit capital-expenditure line in every scenario, flowing through depreciation to the margin, and (b) a growth multiplier applied inside the advertising and cloud vectors. The three scenarios differ chiefly in whether (b) pays for (a): in the bear it does not, in the base it roughly does, in the bull it more than does. Stated that way, artificial intelligence is the swing variable of the entire destination — which is why the bear weight is set as high as it is.

What the disclosed capital-spending numbers say — in both directions

The reassuring reading. Intensity fell from 11.6% of revenue in FY2024 to 10.53% in FY2025 while the operating margin rose from 31.5% to 32.1%. Fourth-quarter 2025 capital expenditure of RMB19,632 million was 46% below the RMB36,578 million of the fourth quarter of 2024. On the most recent quarter, spending is falling year on year, not exploding.

The cautionary reading, weighted equally. Absolute capital expenditure still rose 3.2%, from RMB76,760 million to RMB79,198 million. The quarterly path is lumpy — RMB12,983 million in the third quarter against RMB19,632 million in the fourth — so the exit run-rate is above the full-year average. Management describes the fourth-quarter payments as “primarily to support our AI efforts” and the 2025 interim report describes capital expenditure as “primarily to support our AI-related initiatives”. Both are management’s characterisation, not an audited measure.

A definitional wrinkle recorded rather than smoothed. The same filing reports fourth-quarter 2025 capital expenditure two ways: RMB19,632 million in the operating-metrics table and RMB22.4 billion of “payments for capital expenditures” in the cash-flow narrative. These are an additions basis and a cash-payments basis respectively — a definitional difference, not an error. The table series is used throughout this page and the difference is named here.

The sensitivity of the margin, measured on disclosed figures rather than modelled

Hold FY2025 revenue of RMB751,766 million constant and vary only the margin. The point is the size of the dial.

Arithmetic only: revenue held at the FY2025 figure and multiplied by each historically observed margin. A reversion merely to the FY2023 margin — a level this company actually printed three years ago — would remove about RMB44 billion of operating profit, roughly eighteen percent. A full return to the FY2022 trough would remove RMB91 billion. And the company discloses no sensitivity for any of it: no margin sensitivity, no discount-rate or terminal-growth disclosure, no goodwill-impairment sensitivity, no quantified artificial-intelligence spending plan, no forward guidance of any kind. The absence was searched for and is genuine. It is sharper because a direct rival discloses more: Alibaba’s FY2026 annual report states on its face that its impairment testing rests on “the expected growth in revenues and operating margin, timing of future capital expenditures, an estimate of weighted average cost of capital and terminal growth rate… The periods of the financial forecasts generally range from three to five years.”

What to watch. (1) Capital expenditure as a share of revenue against the “low teens” guidance. (2) Whether depreciation growth outruns marketing-services growth in any two consecutive quarters — the direct signal that the cost is beating the benefit. (3) Any first disclosure of Yuanbao users or artificial-intelligence revenue. (4) China’s policy on artificial-intelligence chip supply. Reads on the thesis: this vector is the swing variable, and the reason the value below is a range rather than a number.

Future 5 of 5 — a value block, not a demand vector

The minority-stake portfolio, and how to count it exactly once

Alongside the operating business Tencent holds minority stakes in a large number of other companies, some listed and some not. Because these are equity-accounted associates rather than subsidiaries, accounting rules do not require the company to publish what they are worth. It publishes the figure anyway in the annual report, every reporting period — disclosure that runs against management’s own near-term interest, since it invites the market to mark the stakes down when they fall.

The two columns are on different accounting definitions and must not be added as though they were the same thing: the listed column is a market fair value, the unlisted column is a historic carrying value. That is why they are shown as two columns and never as one total.

Counting it exactly once. Tencent’s income statement puts its share of profits from associates and joint ventures below the operating-profit line. Operating profit therefore contains none of the portfolio’s earnings, and the portfolio can be looked at as a separate asset without being counted twice. Interest earned on the treasury book sits below the operating line too, so the same is true of the net cash figure. One residual impurity is disclosed rather than glossed: a line reading “other gains/(losses), net −3,177” does sit inside operating profit and may contain investment items — it is a loss, so if it distorts anything it makes the operating base look slightly worse, not better.

What the trend shows, and the confound that stops it being a conclusion. The listed portfolio was marked at RMB726,244 million in mid-2020 and RMB672,700 million at the end of 2025 — down about 7% over five and a half years. Over the same span the company’s own non-IFRS profit attributable to shareholders rose 112%, from RMB122,742 million to RMB259,626 million. That looks like a large pool of capital that has not compounded. It cannot be stated as such, for one specific reason: in December 2021 Tencent distributed its JD.com stake to its own shareholders in specie, and in November 2022 it did the same with Meituan. Those distributions handed value directly to shareholders and left the portfolio — and both announcements have now been ingested: JD.com — 457,326,671 Class A shares, one for every twenty-one Tencent shares, December 2021 and Meituan — about 958,121,562 Class B shares, one for every ten Tencent shares, November 2022. Neither announcement puts a value on the distribution, so the amount handed out still cannot be netted off the portfolio’s change in mark. The correct description is therefore unverified, not destroyed. But the direction is now established from primary documents: this portfolio has been distributed, not compounded — which is why the destination work below treats it as a static store of value rather than a reinvestment engine. One nuance, so the associated income is not mis-scored as low-quality: the FY2025 share of associates’ profits (about RMB33.5bn, FACT, FY2025 AR) and the listed book’s fair-value gains are the yield on a deliberately-built, well-chosen portfolio with a proven realisation mechanism — the in-specie distributions above turn the mark into shareholder value at scale. It is volatile and partly unrealised, so it stays a named nuance rather than a clean pass; but it is strategy-driven, not noise.

One figure explicitly not used. A carrying value of RMB323,116 million for listed equity investments circulated in earlier work on this company. It could not be reconciled to any line in the filings and nothing on this page is built on it. The verified disclosure — the RMB672.7 billion fair value — is used instead. The implication of the gap between fair value and carrying value is itself the interesting part: a large amount of value sits off both the income statement and book equity precisely because associates are equity-accounted rather than marked.

The volatility is the point

Disclosed listed fair value, quarter by quarter, from the company’s own results announcements (RMB bn): 3Q21 1,196.6 → 4Q21 982.8 → 1Q22 606.0 → 4Q22 585.1 → 2Q23 435.4 → 4Q24 569.8 → 3Q25 800.8 → 4Q25 672.7 → 1Q26 547.1. That is a RMB254 billion decline in the two quarters to the first quarter of 2026, with no disposal-versus-mark bridge disclosed, and a RMB377 billion single-quarter drop in the first quarter of 2022 only partly explained by the JD.com distribution. The absence of that bridge is a named gap.

How this block is valued in the numbers below, and the haircuts are stated rather than buried. The listed fair value of RMB672.7 billion is taken at 0.60 / 0.70 / 0.85 in the bear, base and bull cases — for disposal tax, no intent to liquidate, and a holding-company discount. The unlisted carrying value of RMB363.1 billion is taken at 0.30 / 0.50 / 0.65 — for illiquidity, stale marks and the risk of write-downs in Chinese private technology. Not zero, because refusing to value real optionality is inaccurate rather than conservative.

Why the listed block earns the upper end of its haircut range. The listed book appreciated strongly through FY2025 — from RMB569.8bn at end-2024 up to RMB800.8bn at the third quarter before settling at RMB672.7bn at year-end, +18% on the year (tier-1, Tencent results announcements). Two features make that mark more trustworthy than a typical holding-company book: it is already carried at fair value (a live market mark, not a cost-based carrying value that could mask a stale gain), and it has a demonstrated monetisation path — the JD.com and Meituan in-specie distributions realised the mark at or near its stated value and handed it straight to shareholders. Together those support taking the listed block toward the upper end of the 0.60–0.85 haircut range. This does not change the SOTP inputs already used above (still 0.60 / 0.70 / 0.85, RMB672.7bn), and the unlisted RMB363.1bn stays conservatively haircut — it is a note on the quality of the mark, not a re-rating of the number.

This is the single largest non-operating swing factor in the valuation. Give the portfolio no credit at all and the achievable return falls to about 11.56%; take both blocks at 100% with no discount of any kind and it rises to about 16.18% — against the 14.23% base. So only full, un-haircut portfolio credit lifts the return over the 15% bar, and every case stays a hold. It is named as the principal valuation compromise (C7). Note also that the only way this dial alone clears 15% is to take both the listed fair value and the unlisted carrying value in full — no disposal tax, no holding-company discount, no illiquidity discount — while the same disclosure shows the listed block fell RMB254 billion in two quarters.

Reads on the thesis: this block is additive to the operating value and only to it — the investees’ revenue is not in Tencent’s revenue line, and their profits sit below the operating-profit line, which is exactly why the valuation below is built on operating profit and operating cash flow rather than on net profit.

The centrepiece

Industry structure and the capital cycle — where the returns actually go

Vector-by-vector work tells you how big a business gets. It does not tell you what the returns look like when it gets there. That is an industry-structure question, and on this company it is the biggest single driver of value. It is answered here in four steps: where capital is flooding in; who survives; whether a consolidated industry’s returns normalise above the cost of capital; and the policy overlay.

Step one — where capital is flooding in, measured on a rival’s own audited filing

Fact, from the Alibaba FY2026 annual report, year to 31 March 2026: income from operations fell from RMB140,905 million to RMB50,150 million — a 64% collapse — with the margin going from 14.1% to 4.9%, on revenue that grew 2.7%. The interim is starker still: income from operations of RMB40,353 million against RMB71,235 million, margin 8% against 15%, adjusted operating profit down 44% and non-GAAP net income down 43%.

The counter-fact, weighted equally. Tencent’s operating margin went up in the same window — 31.5% to 32.1%, and its non-IFRS operating margin from 36% to 37%. Tencent chose not to fight the instant-commerce war at Alibaba’s and Meituan’s intensity. That is measurable capital discipline and genuine peer-relative franchise strength, and it is the strongest single argument for the base case over the bear.

But note what the discipline has not yet been tested against. Tencent declined the instant-commerce war. It has not declined the artificial-intelligence capital-expenditure war — it under-spent its own guidance in 2025 because of chip supply, and has guided to more than doubling artificial-intelligence product spending in 2026. The brake so far has been availability, not choice. When the brake releases, the discipline gets its real test. That is the bear case in one sentence.

Step two — the shakeout, argued from hard tests rather than adjectives

Sources: Tencent 2025 Annual Report and first-quarter 2026 results; Alibaba FY2026 Annual Report; NetEase Annual Report 2025; Kuaishou Annual Report 2025. Every ByteDance cell is an explicit “not disclosed” with the same reason, because the company is private — it is not omitted quietly.

The uncomfortable peer fact, restated because it is the strongest bear input after Alibaba’s margin. On return on equity for the latest reported financial year, Tencent is third of four — 21.1% against Kuaishou’s 26.3% and NetEase’s 22.0% — ahead only of the company that just destroyed RMB90 billion of operating profit. The framework’s ultimate moat test is superior peer-relative return on capital. Tencent does not clear it. The moat is wide; on this measure it is not the widest. Survival is not the question — Tencent survives every scenario on this page on a net-cash balance sheet. The question is what returns look like on the other side.

Step three — does a consolidated industry’s returns normalise? Answered against named historical analogues, not assertion

Three consolidations that did normalise — returns settled above the cost of capital. United States airlines, 2005–2019: four carriers reached about 80% of the domestic market and, after forty years of value destruction, earned returns above their cost of capital from 2014. What decided it: bankruptcy physically removed capacity and the fleets did not come back; buyers were atomised individual travellers; loyalty programmes and airport slots gave real differentiation; and no state actor funded new entrants. United States wireless, 2019–2020: three national carriers, revenue per user stopped falling. What decided it: spectrum is a genuinely scarce, state-auctioned input, capital intensity deterred entry, and the regulator permitted the consolidation. Chinese beer, after 2016: the survivors shifted from a volume land-grab to premium pricing and margins expanded, because the state tolerated consolidation, brands supported real differentiation, and capacity closures were allowed to happen.

Three that did not. Chinese solar, 2011–2013 and again 2023–2026: consolidated in name twice, normalised neither time, because exit was blocked — local governments kept refinancing capacity to protect employment and tax revenue — the product was fully commoditised, and the state was a participant with non-profit objectives rather than a referee. The 2024–2026 “anti-involution” campaign is the state’s own admission that the market mechanism failed to clear overcapacity it had itself financed. Chinese display panels, 2010–2020: state-financed capacity, permanent overcapacity, returns below the cost of capital for a decade despite the industry becoming a Chinese near-monopoly globally. Chinese local services and instant commerce, 2015–2026 — the closest analogue of all: Didi “won” ride-hailing in 2016 and immediately faced aggregators; Meituan “won” food delivery and is now in a subsidy war with Alibaba and JD. The mechanism is the key: the entrants were never capital-constrained start-ups. They were adjacent platforms with hundred-billion-renminbi balance sheets treating the war as strategically necessary rather than economically justified. A shakeout requires the losers to run out of money. When the challenger is Alibaba, it does not run out of money — it runs its margin down from 14.1% to 4.9% and keeps going.

Reads on the thesis: every scenario below is built on this split verdict, and the structural finding it produces — that the danger is defensive spending in a competed industry rather than the loss of the core — is the finding stated at the top of this page.

The destination, sized

Three futures to 2035 and 2045 — each a business story, none a percentage move around the price

Each scenario below is defined by business outcomes — users, share, unit economics, structure — never by a percentage above or below today’s quote. Each is a coherent story across all five vectors plus a state of the industry, and each carries observable one-to-three-year signposts, so a reader can tell within a year or two which one is happening.

BASE — “Rich, saturated, disciplined” · 55%

Weixin’s user base never grows again: it is 1,418 million and it stays there. Tencent monetises the same people harder — advertising revenue per user compounds from RMB102 toward about RMB240 by 2035, with advertising load rising but staying visibly below Meta density. Games grows with a China market decelerating from 7.7% toward 4%, Tencent holding about 47% share and adding an international leg at 10% then 6%. Payments, cloud and business services stay what the 8% already says they are: a regulated, competed, low-margin utility growing at 8% then 6%. Artificial-intelligence capital spending rises to a sustained 11% of revenue and the advertising-efficiency gain roughly pays for it — the margin drifts from 32.1% to 33.0% on mix, with the extra depreciation netted off. State of the industry: the split verdict above holds.

Signposts: advertising revenue per user growing 12–18% a year · capital expenditure 10–12% of revenue · payments-and-cloud growth 6–9% · domestic games gross receipts growing mid-single-digit · the net share count still falling about 2% a year.

BEAR — “The compute tax” · 25%

The chip curbs relax and the brake comes off. Tencent joins the capital-expenditure war it has so far been unable rather than unwilling to fight; spending settles at 12% of revenue on a much larger installed base, and depreciation compounds faster than the revenue it enables. Artificial-intelligence assistants begin disintermediating the advertising unit — the user asks an assistant instead of browsing a feed — so advertising revenue per user decelerates to 8% then 4% while advertising load rises, which is the harvesting signature. The instant-commerce and assistant wars reach WeChat’s door and Tencent finally has to respond. The state takes more of the payment and cloud profit pool. Games share erodes from about 47% toward 32% as NetEase, miHoYo and artificial-intelligence-native studios take player time. The operating margin compresses from 32.1% to 25.0% by 2035 — still a good business, but operating profit is flat in nominal terms for a decade. State of the industry: the solar column wins the argument, and the differentiated core turns out to have been protected by an absence of attackers rather than by a genuine barrier.

Signposts: capital expenditure above 13% of revenue · depreciation growth exceeding marketing-services growth for two consecutive quarters · advertising revenue per user growing below 10% for two consecutive years · payments-and-cloud growth below 5% · combined users turning negative · the buyback shrinking the count by less than about 1% a year.

BULL — “The attention rail monetises” · 20%

WeChat becomes the transaction and assistant layer for Chinese commerce, not merely its messaging layer. Mini Programs plus Video Accounts plus Weixin Pay close the loop — discovery, transaction and payment on one rail, with attribution ByteDance cannot match — and advertising revenue per user compounds at 17% then 11% toward about RMB270 by 2035, still a quarter below Meta’s current rate. Marketing services reach about 26% of a China digital advertising pool Tencent is already taking share of. Inference demand converts Tencent Cloud from an 8%-grower into a 14%-then-10% grower, and the artificial-intelligence capital spending earns its cost of capital through the advertising engine rather than through renting out compute. International games scale into a genuine second franchise. The operating margin expands to 36.0% by 2035 on mix, because the incremental revenue is the highest-margin revenue Tencent has. State of the industry: the wireless column — and the state, through the anti-involution campaign, actually helps by suppressing the subsidy wars.

Signposts: advertising revenue per user growing above 18% a year, sustained · Tencent first discloses Video Accounts or Yuanbao metrics · payments-and-cloud growth above 12% for two consecutive quarters · marketing services above 25% of group revenue by 2030 · the operating margin above 34%.

The destination, sized in the three lines that matter

FY2025 figures are facts from the 2025 Annual Report. Every 2035 and 2045 cell is analyst inference, built upward from the five vector sections above. Three consistency checks were run and are stated rather than assumed: the vector rows sum to the revenue totals; the base 2035 revenue of RMB1,429.9 billion sits below the previous valuation’s year-ten revenue of RMB1,518.5 billion, so the rebuilt destination is more conservative than the one it replaces despite being built from the bottom up; and the ratio of operating cash flow to operating profit is ramped from FY2025’s actual 1.2546 to 1.30–1.40, rising in the bear case because depreciation is a non-cash charge — which deliberately makes the bear harder, not easier, to argue.

The probability assigned to each future, and the reasoning for each

These are the analytical engine’s stated call, published with the reasoning attached precisely so that any one of them can be overwritten without rebuilding the model. They were set from the vector and structure work above, before the discounting was run, and they are not calibrated to produce any particular answer.

Reads on the thesis: the probabilities are now the assumption carrying the most value in this analysis — which is why they are published with their reasoning rather than merely with their result.

The value

Value the whole business as a fair value, then deduce the return the price actually delivers

The bottom line. If you buy Tencent at today’s price of HKD 471.80, you would earn about 14% a year — just below the 15% we require. It falls a little short on our best guess and further short on a few reasonable changes. So it is a hold, not a buy: a great business, but the price gives you no cushion — and it has risen about 6.5% since our last look, moving further from a buy. (The price is struck to the previous trading-day close, HKD 471.80 as at 1 August 2026, so it doesn’t wobble with intraday moves — this is a ten-year analysis, not a day-trade.) Here is how we get there, in plain terms.

How we value it. We treat Tencent as one whole business worth three things added together: (1) the spare cash the operating business should throw off over the next ten years — the cash left after running it and paying for its capital investment, including the full AI build-out; (2) what its pile of investment stakes is worth; and (3) its net cash. Because cash in the future is worth less than cash today, we shrink those future amounts back to today’s money using a safe government-bond yardstick. That gives what the whole business is worth, which we set beside what the market charges for the whole business.

But “what it’s worth” is not the decision — the return you’d earn is. The deciding question is simple: if you pay today’s price, what yearly return do you actually earn as that cash arrives? We want at least 15% a year. Everything below builds to that one number.

The safe-bond yardstick. To bring future cash back to today’s money we use 5.2% — the yield on long-dated US government bonds, which fits Tencent because its shares trade in Hong Kong dollars and the HK dollar is pegged to the US dollar. (A Chinese government-bond rate would be lower and would make the business look worth even more, so 5.2% is the more cautious choice for the “what it’s worth” figure.) The cash-flow forecasts and the durability assumption are our estimates; the 2025 balance-sheet figures are facts from the accounts; the HKD 471.80 price is market data.¹

The one assumption that decides it: how durable is the cash machine? Most of the value is the cash Tencent keeps producing after 2035. We can’t forecast every future year, so we roll all of it into one number — the business in 2035 is worth 14 times that year’s spare cash, a “good and durable, but not growing forever” assumption you set, not one dialled to hit the price. It matters enormously: turn that dial and the yearly return moves about six points, and on our central setting it already sits just below the 15% line — only the most optimistic durability setting clears it, and every other reasonable choice is lower still. See the box.

Fair value, whole-company — sum of the parts, by scenario

Net cash of RMB107,145 million at 31 December 2025 is a fact from the 2025 Annual Report, as are the portfolio’s RMB672.7 billion listed fair value and RMB363.1 billion unlisted carrying value; the haircuts (weighting to RMB648bn) are stated in the portfolio section. The operating value is the sized destination discounted at 5.2% over a 10-year explicit horizon to 2035, with a 14× terminal on the 2035 free cash flow. Per-share figures use 9,250 million shares — a derived number, not a disclosed one — and RMB 1 = HKD 1.087; both are a known limitation: four different share counts circulate, 8.90bn to 9.52bn, and a 4% count error is a 4% error in every per-share figure here.

The achievable return — the decision

From the same probability-weighted forecast, solve for the annual return that equates the destination to the price:

On the central case a buyer at HKD 471.80 earns an implied 14.23% a year — about eight-tenths of a point below the 15% bar. And every one of the model’s own reasonable variations is lower still. Change one convention at a time:

The central case and every reasonable variation except the most optimistic terminal sit below 15%. The honest read is “fairly-valued-to-slightly-rich against the required return,” a HOLD / WATCH, not a buy. The 1.75× fair-value gap is a discount-rate artefact and buys no safety.

What the pass hinges on — the terminal multiple, the headline sensitivity

This is the largest single swing factor, so it leads the sensitivity work. Hold the same v29 forecast fixed and move only the terminal free-cash-flow multiple:

These are same-forecast v29 numbers. The often-quoted 13.83% is a different thing — the prior method’s figure at its own g-derived ~9× terminal — and is used on this page only as the prior-method comparison, never as a v29 sensitivity point.

The whole call lives in this table. The business, the forecast and the probabilities are identical across the three rows; only the terminal-durability convention moves, and it moves the return ~6 points. At 9× the name fails outright; at the house 14× it fails by ~0.8 of a point; only at 20× does it clear. That the case needs the most optimistic durability dial just to reach the bar is why the verdict is a HOLD / WATCH, not a buy.

Margin of safety, and the buy level

At HKD 471.80 there is no margin of safety — and the price now sits above break-even. The central case is already below the bar (14.23%), and at ~HKD 450 the return would merely break even at 15%, so today’s HKD 471.80 is above that break-even line. A genuine margin of safety requires the achievable return comfortably above 15%, which means a materially lower entry (the return rises as the price falls):

Even that ladder is the central-case forecast; on the 20-year horizon or the top-of-range share count the whole ladder shifts down. At the same-forecast 9× terminal the name does not clear 15% at any price near today’s. This is a below-the-bar, assumption-dependent read, not a fat pitch — a HOLD / WATCH at HKD 471.80.

The other dials, and why none of them decides

The terminal multiple is the largest swing, and every other dial points the same way here. The horizon (20-year → 13.40%) and the share count (9,520m → 13.69%) sit below the bar too, and the stricter own-FCF definition pushes lower still — which is why the central 14.23% needs the most optimistic terminal just to reach 15%. The operating-margin dial, by contrast, is muted: the bear scenario already carries a 25.0% 2035 margin at 25% probability, so margin risk is priced in the weighting rather than needing to be re-priced in the base.

And Tencent’s own disclosed sensitivity to any of this: none found. No margin sensitivity, no cost-of-capital or terminal disclosure, no impairment sensitivity table, no quantified capital-expenditure plan, no forward guidance of any kind (equipment useful life is disclosed — 2 to 10 years — but not split out for AI-specific assets). Where no sensitivity is disclosed, the opacity is itself the flag — and it is sharper because Alibaba discloses more.

The north star

The one metric that matters — WeChat advertising revenue per user

A note on the name. This measure is what the advertising industry calls ARPU — average revenue per user. It is written out in full throughout this page, and the version used here is specific: marketing services revenue divided by combined Weixin and WeChat monthly active users, which is an advertising ARPU rather than a total-revenue-per-user figure.

If you could see only one operating number on this company, it should be this one. The framework tests a candidate metric against four questions, each a pass or a fail rather than a feeling: does it sit where the moat converts into owner cash (a driver, not a result)? Does it turn before the reported financials? Can management flatter it? Would a smart competitor fear it?

First, the honest part: the previous falsifiable test does not survive those four questions

The 26 July verdict used Tencent’s FY2026 operating margin as its falsifiable test. Run against the four questions it fails all four.

So the FY2026 operating-margin test is demoted. It is not discarded — it is retained below as what it actually is, a thesis-breaker gauge: the number that tells you the bear case has arrived, after it has arrived. It is no longer the falsifiable test.

The metric selected, and the series

The ratio is an inference computed from two separately disclosed facts, both in the 2025 and 2021 reporting. The +16.7% headline is the ratio of the two disclosed ARPU levels (102.2 ÷ 87.6). An alternative “+17.1%” comes from deflating the filing’s +19.4% marketing-services growth by its rounded “+2%” user growth; this page uses the cell-consistent +16.7% so the headline reconciles to its own table. The two missing years are a real gap and are shown as such rather than filled. Backfilling the full series is a named open remediation item for the next re-run.

Why it passes all four tests. One: it sits exactly where the moat converts into owner cash. The moat is a 1,418-million-node real-identity network, and there is precisely one place that network turns into cash at a high incremental margin. This metric measures it, and it is upstream of revenue, upstream of margin, upstream of everything. Two: it leads. Revenue per user turns quarters before the operating margin does, because the margin only moves once the mix shift has already happened — and it is the metric that discriminates between the scenarios: the bear is “8% then 4%”, the bull is “17% then 11%”. Three: it is hard to flatter. Both inputs are separately disclosed headline figures reported for other reasons; neither can be moved by an accounting choice, a one-off gain or a reclassification below the operating line. Four: a smart competitor would fear it. ByteDance’s entire commercial position in China rests on owning attention Tencent has historically declined to monetise hard, so a rising number here is the direct quarterly measure of that position eroding.

The one named limitation, and its direction stated correctly. Marketing services (the numerator) includes non-WeChat inventory — QQ, media, Tencent News — while the denominator is WeChat-only monthly users. So on the level, the pure-WeChat ARPU is lower than RMB102.2, because the numerator carries revenue earned off users the denominator does not count — which means the headroom against Meta’s ~RMB360 is actually larger, not smaller. Separately, on the growth rate, the slower-growing non-WeChat mix understates the pure-WeChat ARPU’s growth, so the true WeChat compounding is faster than the blended +16.7%. The clean series — Video Accounts and Mini Program advertising per WeChat user — is not disclosed and is recorded as a gap.

And this metric resolves the framework’s hardest question about this name. The second gate asks whether growth comes from extracting price from trapped customers. Tencent’s user base has stopped growing, so every renminbi of future growth in this vector comes from monetising the same people harder — which is either skilful monetisation of a genuinely under-monetised asset or harvesting. Read against advertising load and user time, this is the only number that tells the two apart. That is why it is the one metric.

The runner-up, and why it is not the primary

Games gross receipts growth, domestic and international, stripped of currency movement. It covers the other profit engine and it is explicitly leading on Tencent’s own testimony: the first quarter of 2026 noted revenue growth lagging gross receipts on deferred-recognition timing; the third quarter of 2024 described international games revenue growth “substantially lagging gross receipts”; and the first quarter of 2024 printed international games revenue up 3% while gross receipts were up 34%. A thirty-one-point gap in a single quarter is the definition of a leading indicator; deferred revenue mechanically converts one into the other over the following two to four quarters. Its limitation, named: Tencent gives gross receipts as commentary rather than always as a number, so the third test is only partly satisfied. Hence runner-up.

The watch-triggers — explicit, numeric, and thesis-breaking

The one input Tencent does not disclose — read on output, flagged for transparency

Tencent does not disclose research and development expense as a separate line item. It is folded into general and administrative expenses. That is established by an exhaustive sweep of the complete ingested corpus — twenty-two annual reports, twenty-two interim reports, ninety results announcements from 2004 to 2026 and sixty earnings-call transcripts — in which exactly one incidental research-and-development figure appears anywhere.

Tencent is the only member of its own peer set that does not separately disclose what it spends on technology. The framework’s principle applies by name: where no figure is disclosed, the opacity is itself the flag. This is not an accusation of bad faith — folding research and development into general and administrative expenses is permissible and Tencent has done it consistently for twenty-two years, which is the opposite of an opportunistic change. But consistency is not transparency, and this framework grades what can be verified.

The only container disclosed is general and administrative expenses, of which research and development is an unquantified subset. The quarterly figures retrievable from the results announcements (RMB m) run 17,189 in the third quarter of 2020, 18,967 in the first quarter of 2021, 23,862 in the third quarter of 2021, 26,669 in the first quarter of 2022, 25,419 in the second quarter of 2023 and 26,289 in the third quarter of 2023: roughly doubling to early 2022 and then flattening. That is an upper bound on research and development and nothing more. It cannot distinguish the two readings the test exists to distinguish — a widening moat, where scale funds a programme sub-scale rivals cannot match, from a treadmill, where spending rises as a share of revenue merely to hold position.

The stress test

Forces of change — every load-bearing claim dated and re-validated

A competitive claim is a snapshot with a shelf-life, not a permanent truth. Every claim this analysis leans on is therefore dated, given its evidence tier, and re-checked against today — 31 July 2026 — across five standard forces: technology, demand shift, geopolitics, regulation and the capital cycle. The individual “Dated 29 July 2026” stamps below mark when each claim was last stress-tested in the forensic run and are carried forward and re-validated on this 31 July page; none has changed status. Claims that fail re-validation are retired, not repeated. One has already failed: the 2022 third-party claim that Tencent’s TDSQL was the leading distributed database in China rests on an undated four-year-old study and has been withdrawn from use.

Technology

The disruption that would actually break the thesis: artificial-intelligence assistants disintermediating the advertising feed. If a user asks an assistant to buy something instead of browsing a feed, the advertising impression — the unit Tencent sells — stops existing. Dated status: not yet observable. No metric in the filings has moved. The signpost is advertising revenue per user decelerating while advertising load rises. This is the primary bear mechanism and it currently has zero confirming evidence, which is why the bear carries 25% rather than more.

The counter-disruption, and it favours Tencent. Whoever owns the endpoint captures the value of artificial intelligence, not whoever owns the compute. Tencent owns 1,418 million endpoints and monetises them at RMB 102 a year against Meta’s roughly RMB 360. Still true today, and it is the single largest quantified upside in the destination.

Demand and substitution

The substitution question that matters is not “where do users go” but “where does user time go”. Tencent competes with ByteDance and Kuaishou for the same finite attention. Kuaishou’s FY2025 average daily users were 410 million, above 400 million for four consecutive quarters, and it grew revenue 12.5% at a 14.5% operating margin while earning the highest return on equity in the peer set at 26.3%. ByteDance is not measurable at all: no primary filings exist anywhere. That is the largest structural gap in this competitive analysis, and it is not fixable. Dated 29 July 2026: the substitution risk is real, live, and only partially observable.

Two facts anchor the demand side and both still hold. Combined users of 1,418 million at 31 December 2025, up 2% on the year and 0.3% on the quarter, on a twelve-year path from +41% to +2% — the most important single fact in the destination. And China’s games market growing 7.68% in 2025 on a player base growing 1.35% — the market itself is now monetisation-led rather than user-led, one layer above Tencent, with the same signature.

Geopolitics, macroeconomics and commodities

The macroeconomic exposure, in management’s own words (17 August 2022): “we generate approximately half of our revenues from FinTech and Business Services as well as Online Advertising that directly contribute to, and benefit from, overall economic activity.” Still true today, and the proportion has risen — those two segments were 49.8% of FY2025 revenue. Roughly half of Tencent is a geared play on Chinese nominal economic activity. That is not a moat question; it is an exposure the destination has to carry, and the base case’s 4.5–6% advertising-pool growth embeds it. On balance-sheet endurance: net cash of RMB 107,145 million at 31 December 2025 and RMB 146.9 billion at 31 March 2026 — Tencent survives every scenario on this page. Endurance is not the question.

Regulation and policy

The two-sided reading of anti-involution, which most commentary gets one-sided. The bullish leg: it is the state trying to stop exactly the subsidy war that took Alibaba’s operating margin from 14.1% to 4.9%. If it works, Chinese platform returns normalise upward and the bull structure state becomes more likely. The bearish leg: it is simultaneously a demonstration that the state sets industry economics by decree, and an industry whose returns are administratively determined cannot be underwritten on competitive analysis alone. Dated 29 July 2026. Net verdict: mildly positive for industry returns, clearly negative for predictability — and it is carried in the bull structure state only.

The capital cycle — and the gauge that tells you the bear has arrived

The full study is the centrepiece section above. The dated version: capital is flooding into Chinese artificial-intelligence compute, and returns on that compute will be poor — every large platform is being told by its board and its government to build the same thing at the same time, with no differentiation in the output and no exit mechanism for the losers. The consequence for Tencent is precise and it is not the obvious one: the risk is not that the core gets competed away; it is that Tencent is forced to spend defensively in an industry whose returns on that spend get competed to zero, to protect a core that was not actually under threat. The margin, not the revenue, is the battleground. The gauge is below.

What would break the structure thesis, by force

Honest summary of the stress test. Four of the five forces currently show no confirming evidence of damage, and the fifth — the capital cycle — shows damage at a rival, not at Tencent. That asymmetry is precisely why the base case carries 55% and the bear 25% rather than the reverse. The bear case is a well-evidenced mechanism with no confirming observation yet, and that is exactly the kind of risk that deserves a real probability rather than a full one.

Is Tencent the discarded quality? Every investment frenzy finances overcapacity in one place and creates a pool of unfairly discarded quality nearby. Dated verdict, 1 August 2026: partially, and less so than it was. The shares are at HKD 471.80 against a fifty-two-week range of HKD 411.00 to 683.00 — still nearer the low than the high. On the v29 method the achievable return at HKD 471.80 sits just below the 15% bar — a central-case 14.23% that only the most optimistic terminal lifts over the line — so the discount is thin, not deep: falsely-condemned quality bought for real margin of safety needs a lower entry, not HKD 471.80.

Re-run trigger: all of the above is re-validated at the FY2026 interim results. Any dated claim older than four quarters is re-checked before it is relied on again.

The uncontrollable part

Regulatory history, measured in the company’s own numbers

The single hardest number in this section: in the first quarter of 2022 Tencent’s own results announcement attributed a 1% decline in domestic games revenue, to RMB33.0 billion, to “the direct and indirect effects of the minor protection measures”. Dated re-check, 29 July 2026: decayed as a live driver — domestic games grew 18% in FY2025. But the precedent stands: the state can remove a segment’s growth by decree, and did.

Chinese consumer internet has a feature no operator can manage away: the state is a co-claimant on the profit pool, through games licensing, minors’ protection rules, financial-technology rules and data rules. This is not an abstraction on this company — it has happened twice in seven years and both times it is visible in the filings.

The significance is not that another intervention is imminent — nothing in the evidence points to a live one, and FY2025 is a record year on every line. It is that a business can be legislated out of part of its profit pool without any competitor winning that profit, and no competitive advantage defends against that. Two occurrences in seven years is a base rate that can be described but not forecast: neither trigger was visible in any filing beforehand.

Two structural risks the company itself put in writing

The ownership structure. Foreign shareholders do not own the mainland operating entities; they own contractual rights over them, an arrangement usually called a variable-interest entity. Tencent disclosed it plainly in its very first post-listing annual report, in 2004: the registered shareholders “are required under their contractual arrangements with the Group to transfer these interests in Tencent Computer to the Group or the Group’s designee upon the Group’s request at a pre-agreed nominal consideration.” The single largest structural feature of the equity, disclosed in year one in plain language. Its enforceability has never been tested against a hostile Chinese state.

Currency, in management’s own words. From the 2014 interim report: “there are limited cost-effective hedges against the fluctuation of RMB and no effective manner to convert a significant amount of non-RMB currencies into RMB, which is not a freely exchangeable currency.” A Hong Kong dollar-denominated claim on renminbi earnings therefore carries a conversion and repatriation exposure the company states it cannot hedge away.

Named gaps in this section’s evidence. Four of nine standard forensic checks were not run in this review — receivables against revenue, full cash conversion, auditor rotation and finance-director turnover, and related-party transactions. Separately, no primary data on insider disposals or share pledges is held. The two largest holdings are now on the primary record (FY2025 annual report): Prosus/Naspers 22.80% via MIH Internet Holdings B.V., and founder Ma Huateng 8.82% — see the People section.

Competitive context

Peer benchmark — and why these three rivals and no others

Why these three rivals, and no others

The selection rule, stated rather than assumed. A peer set is not “the top three” by size or by reputation. It is chosen so that each name maps to a distinct part of the subject’s moat, and so that every name files primary accounts a reader can check. On that rule the set is:

• NetEase (9999.HK) — the games moat. The credible number two in Chinese games and about 82% one product line: games and related value-added services were RMB92.1 billion of its RMB112.6 billion of revenue. It is the cleanest available read on whether Tencent’s games franchise earns a premium. It does not: NetEase earns a higher gross margin and a higher return on equity at one-seventh the size.
• Kuaishou (1024.HK) — the listed proxy for the short-video threat. The competitor that actually matters for Chinese attention and advertising money is ByteDance, and ByteDance cannot be measured. Kuaishou is the only listed short-video company filing primary accounts in the same market, so it stands in as the measurable proxy for that threat — explicitly as a proxy, never as a substitute.
• Alibaba (9988.HK) — the advertising, cloud and payments moat, all three at once. The only rival that competes with Tencent simultaneously in digital advertising, in cloud and in payments, and the only one large enough for its capital decisions to set the industry’s returns. Its FY2026 accounts are also the hardest live evidence on this page for the capital-cycle argument.

Two names are deliberately absent, and both absences are stated rather than hidden. ByteDance is private, files nothing on any exchange, and appears in no archive. On any reading it is the most important competitor for Chinese attention and advertising money, and no filing anywhere on earth allows a check on whether the gap is opening or closing. It is recorded as the largest structural gap in this analysis, and it is not fixable by any effort available here. miHoYo, the other significant games challenger, is private for the same reason.

Two structural handling rules follow, and they are handled structurally rather than by footnote. First, NetEase files no standalone Hong Kong interim report: its quarterly results release serves that purpose under Listing Rule 13.48(1), so the absence is a permitted filing practice and not a disclosure failing — it is why one NetEase cell below is an explicit “not disclosed”. Second, Alibaba’s financial year ends on 31 March, not 31 December, so it gets its own table with its own labelled columns and is never placed in a shared row with the December-year companies.

A single company’s numbers mean nothing until the same numbers are read off its rivals’ own filings. Every figure below is taken from that company’s own annual report — no screening service, no third-party database, no recollection. The tables have been rebuilt to one standard: one measure, over one stated period, on one accounting definition per column; every cell either a real figure or an explicit “not disclosed” carrying its reason; and any residual non-comparability written down rather than quietly averaged away. Rows are ordered by the measure, not by who the subject of the dossier is.

The structural problem, handled structurally. Alibaba’s financial year ends on 31 March, not 31 December. Its “FY2026” runs to 31 March 2026 and overlaps Tencent’s calendar FY2025 by only nine months — and the difference is not cosmetic here, because Alibaba’s margin collapse is concentrated in the April–September 2025 half (income from operations RMB40,353m against RMB71,235m a year earlier; margin 8% against 15%). Alibaba therefore gets its own table, labelled with its own year-end, and is never placed in a shared row with the December-year companies.

Table 1 — the December-year set: scale and margins, year to 31 December 2025

Sources, cell by cell: Tencent 2025 Annual Report · Kuaishou Annual Report 2025 · NetEase Annual Report 2025.

Residual non-comparability, disclosed not normalised. Tencent and Kuaishou report under International Financial Reporting Standards; NetEase reports on a United States accounting lineage with reconciliation notes, so the operating-profit definitions differ slightly at the edges. NetEase reports in thousands of renminbi and has been converted to millions here. All three report in renminbi, so no currency conversion is involved. ByteDance, which owns Douyin and is on any reading the most important competitor for Chinese attention and advertising money, is absent because it is private and files nothing anywhere — recorded as a gap, not silently omitted.

Table 2 — return on equity, ordered by the measure, both bases shown

Return on equity is profit expressed as a percentage of the money shareholders have in the business. It is the cleanest single comparison of quality, and it is also the easiest to get subtly wrong, because there are two defensible denominators.

How each cell is built, so the definitions are visible. Numerator in every case is profit attributable to shareholders. Denominator in the like-for-like column is the average of opening and closing total equity. Kuaishou: 18,617 ÷ ((62,024 + 79,584) ÷ 2) = 26.3%. NetEase: 33,760 ÷ ((142,410 + 165,013) ÷ 2) = 22.0%. Tencent: 224,842 ÷ ((1,053,896 + 1,241,065) ÷ 2) = 19.6%. Tencent’s 21.1% uses average attributable equity — 224,842 ÷ ((973,548 + 1,154,152) ÷ 2). The rivals’ attributable figures ARE retrievable and are now shown: Kuaishou 26.30% and NetEase 22.6% (an earlier “not disclosed” on these cells was a search failure, not an issuer gap, and is corrected). The ordering is the same under either basis — total-equity 26.3 / 22.0 / 19.6, attributable 26.30 / 22.6 / 21.1 — so Tencent is third of the three December-year names on return on capital whichever denominator is used.

Table 3 — Alibaba, shown separately for its 31 March year-end

Source: Alibaba Group Fiscal Year 2026 Annual Report and Fiscal Year 2026 Interim Report.

The two columns overlap by nine months and are not the same period; that is the whole reason for the separate table. Alibaba’s label for its operating line is “income from operations”, a slightly different construction from the IFRS operating profit used by Tencent, and is not renamed here.

Reading the peer tables

Table 4, and the three claims the cells will and will not support

Table 4 — operating margin, prior year against latest year

Direction matters as much as level, so the margin is shown as a before-and-after with both periods named in the cell rather than as an arrow.

NetEase’s prior-year cell is now filled — FY2024 operating profit RMB29,584m gives a 28.1% margin, a +3.7-point improvement — correcting an earlier “not disclosed” that was a search failure, not an issuer gap. Stated plainly and against the thesis: of the three December-year names, Tencent shows the SMALLEST operating-margin improvement — +0.6 points, behind NetEase’s +3.7 and Kuaishou’s +2.5. Tencent’s margin is the highest in level, but it is improving the least. Alibaba’s two periods are March years and are labelled as such inside the cells; its change is a within-company comparison rather than a cross-company one.

What the cells support — and three loose claims corrected against them

Claim 1 — growth. Among the three December-year companies, Tencent grew fastest: +13.9% against Kuaishou’s +12.5% and NetEase’s +7.0%. The honest description is fastest of the three, by 1.4 percentage points over Kuaishou, at more than five times Kuaishou’s revenue — growing quickest while being much the largest of that set is the harder version of the achievement, and it is real. What the cells do not support is “the largest in the peer group”: Alibaba is bigger, at RMB1,023,670 million against Tencent’s RMB751,766 million, although over a year ending three months later.

Claim 2 — margins. Tencent’s operating margin of 32.1% is the highest of the three December-year companies, but NetEase is 30 basis points behind at 31.8% and the difference is noise. On gross margin Tencent is behind: NetEase earns 64.3% against Tencent’s 56.2%, the highest of the three for which the figure is disclosed (Alibaba’s is not disclosed comparably). NetEase is roughly one-seventh Tencent’s size and about 82% one product line — games and related value-added services were RMB92.1 billion of its RMB112.6 billion. So the far larger, far more diversified business earns no margin premium over the small focused one. There are two readings, and both are demerits rather than proof of an advantage: either the wider ecosystem adds no margin, or a very profitable games business is being diluted by a lower-margin payments-and-cloud business. Which one it is cannot be settled, because segment margins are not disclosed.

Claim 3 — return on capital. Ordered by the like-for-like (average total equity) measure: Kuaishou 26.3%, NetEase 22.0%, Tencent 19.6%, Alibaba 9.6% over its March year. On the attributable-equity basis — now retrievable for all three December-year names — the ordering is unchanged: Kuaishou 26.30%, NetEase 22.6%, Tencent 21.1%. Tencent is third of four either way. It is not the return-on-capital leader of its own peer set.

Before → after: what the symmetric-precision audit changed in these tables

Third-tier material

Social and video sources — leads, never facts

The items below are third-party interpretation, not evidence of record. A single post or video is a lead: it raises a question to be checked against the primary filings. Only an aggregate pattern — complaint-count trends, resolution rates, review-score trajectories, regulator tallies, weighted by credibility — can corroborate anything, and even then it is checked against filings first. This material is listed for transparency about what was reviewed, and because being explicit about what was rejected is part of the record.

This is not an aggregate. It is four voices, one of them doubled, and one of them the company itself. It does not meet the standard for corroborating anything, and no statement on this page rests on it. Every hard claim was nonetheless tested against the primary filings, because a lead is worth checking even when the source is worth nothing.

Provenance for all five items: captured from visible transcript panels in a browser; transcript origin auto-generated or machine-generated captions; each stored file self-labelled as a derived lead and not a filing.

The ledger

Every named compromise, C1 to C19

The framework’s rule is that every compromise is named in the verdict: an unnamed weakness is a blind spot, a named one is a position. Each entry below says what the compromise is, what pays for it, and what would make it intolerable. This v27 run re-classes C18 (from “moat unmeasurable” to a transparency demerit under #18) and partly remediates C11 (the Prosus and founder holdings are now disclosed figures, not recollection); C19 (uncharged capital expenditure in the prior model) stands as fixed.

What is not on this page, and it is the only thing withheld: the author’s own holding in the name. The verdict, the opportunity type, the scorecard, the gate outcomes, the intrinsic-value target, the buy level, the compromise ledger and the sized destination are all published above.