---
title: "Tencent: A Rare Profit Reversal — How Long Will the AI 'Tuition' Last?"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/dolphin/post/45156515.md"
description: "The core biz. is roughly stable. However, escalating AI 'tuition'—ever-costlier investments—has pushed profits into reverse."
datetime: "2026-11-18T14:57:48.000Z"
locales:
  - [en](https://longbridge.com/en/dolphin/post/45156515.md)
  - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/45156515.md)
  - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/45156515.md)
author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)"
generator: "portal-rs"
---

# Tencent: A Rare Profit Reversal — How Long Will the AI 'Tuition' Last?

After the Hong Kong close on Nov 18 Beijing time, Tencent Holdings (0700.HK) released its Q3 2026 results. Overall, the quarter again put pressure on shareholders. In short, spending rose while earnings fell.

Revenue came in slightly below estimates, and profit posted its first YoY decline in four years. The core biz also missed, but that is not the main issue. The real drag is the ever-rising 'tuition' for AI.

So vs. the quarterly print, investors should focus more on the outlook for this 'tuition': under constraints on remote compute, how to treat prepayments, how compute will be deployed, and whether Capex targets will be revised. We suggest paying attention to the call.

In detail:

**1) Two glaring numbers: profit -6%, FCF down by RMB 70bn.**

Given this unprecedented AI investment cycle, a profit 'air pocket' at an asset-light Tencent was not a surprise. But most Tencent holders are fundamentals-focused value investors and insurers with low-to-mid risk appetite.

So no matter how the AI story is told, or how cloud leasing unit economics are framed, without showing up in the P&L, capital will not rush to buy in. Ex-AI products, Q3 OP was RMB 83.3bn, up 8% YoY; including net AI spend cut OP by RMB 15.2bn.

**2) Paying more, buying more — AI spend went even harder:** Tencent’s AI spend runs through two books: Capex in cash flow (incl. compute prepayments booked in operating cash flow), and Opex on the income statement. Including prepayments, Q3 Capex cash outlay hit RMB 151.1bn (mostly compute-related hardware).

Of that, RMB 65.1bn was paid and delivered in-quarter, which is generally the cross-company comparable Capex lens. But the market often uses the accrual basis of RMB 81.4bn for Tencent’s time-series comparison (RMB 16.3bn higher than RMB 65.1bn likely reflects amounts prepaid earlier).

Some leased compute and AI product development costs are expensed in R&D Opex. In Q3, non-comp components of R&D reached RMB 15.4bn, up almost RMB 5bn QoQ, roughly in line with Q2’s net increase of RMB 4.5bn. With budgets tight across the group, the incremental line is largely compute leasing and incremental depreciation.

**3) Core ops mixed: overseas games paused; cloud and ads held up**

Q3 revenue was RMB 207.3bn, up 7.5% YoY, missing consensus by RMB 3bn, mainly on Intl games and social network. Overseas games faced a tough comp from last year’s Supercell strength, plus ageing titles, a lack of fresh hits, and FX, driving a 9% decline and weighing on VAS. Domestic games, supported by evergreen titles like 'Delta' and 'Honor of Kings', were decent at +11%.

Social network revenue is tied to digital content subs and game channel rev-share in QQ/WeChat, and fell 2% YoY in Q3. This was likely dragged by Tencent Video and weaker game industry trends compressing channel rev-share. Deferred revenue fell RMB 4.3bn QoQ, and back-solvable VAS gross billings fell 3% YoY, pointing to softer paid content demand.

**Ads rose 18%. Despite a tougher 2H consumer backdrop, Tencent’s ecosystem proved resilient.** Beyond more time spent and higher load on Video Accounts, short-drama ads likely contributed as well (IAA typically ~70% of the market). Green-bubble short dramas scaled quickly after launching on WeChat mini programs; incentives added in Oct drew studios from red-bubble to green-bubble, and MAU has reached 150mn.

Cloud revenue is included in FinTech & Enterprise. On a standalone basis, growth accelerated to 35%, the high end of sell-side expectations and the only beat. WorkBuddy DAU exceeded 10mn, accelerating vs. 6.13mn in Sept.

4) **Profit pressure from rising IT and content costs**

Beyond R&D, costs were another source of variance on the expense side. Sales and G&A were controlled, with ratios stable QoQ. Surprisingly, ad GPM fell to 52%, down 500bps QoQ vs. a normal seasonal 100–200bps. Likely drivers include AI-related allocation, short-drama rev-share, and long-audio content royalties.

Game GPM continued to improve, as evergreen titles have low marginal spend and no IP rev-share, supporting higher margins. PC titles like 'Delta' also tend to carry higher margins.

FinTech & Enterprise GPM fell 200bps QoQ, likely due to mix shift with a higher share from lower-margin cloud. Payments also faced top-line pressure from a weak consumer environment.

**5) Buybacks turned frugal:** Q3 buybacks totaled HKD 7bn at an avg. price of HKD 443/share; YTD was HKD 30bn, and full-year likely below HKD 50bn. If compute spending cannot be deployed as planned, it remains to be seen whether shareholder returns will be stepped up.

**6) Detailed financials**

Tencent earnings call transcript: [‘Tencent (Trans): WorkBuddy scaling; Capex near peak’](https://longbridge.cn/zh-CN/dolphin/post/45151360)

## Dolphin Research View

Last quarter, we anticipated some 2H profit pressure, but the actual 6% decline was worse than expected. In this context, the market saw the company go the other way — spending more (more prepayments to lock compute) and buying back less (even as the stock weakened). For many long-only holders, even if they accept the AI embrace, this is a tough outcome.

**From the top line alone, we lean toward no major issues in Tencent’s core biz. Time-share erosion in the ecosystem can be patched and repaired by short dramas, which also add ad inventory.**

**The key question is how long the heavy AI spend lasts.** For this quarter’s RMB 151.0bn Capex cash outlay incl. prepayments and RMB 15.2bn net AI spend, investors likely have at least the following questions: (1) The company guided full-year net spend of RMB 36bn; by Q3 it had spent RMB 34.5bn. Will Q4 slow, or breach guidance and keep spending?

(2) WorkBuddy DAU rose 70% in two months, yet the lift to cloud revenue seems limited (from +20% to +35%, implying only ~RMB 1bn+ QoQ increase). Meanwhile, non-comp R&D rose by RMB 5bn QoQ for IT servers. Does this mean many WorkBuddy users are still on freebies, and the company is training behavior? If so, can management offer clearer goals and conversion targets?

(3) Management said FY26–27 would be peak investment years, but Monday’s new restrictions on remote compute add a critical question — how much of prepayments relate to restricted overseas GPUs, and how will compute deployment be planned post-effective date? The sharp rise in prepayments this quarter suggests some proactive front-loading — is that the case?

On valuation: pre-print, the stock was HKD 372.8, down 24% from the Aug peak. Mkt cap was ~HKD 3.4tn, implying ~11x this year’s Non-IFRS NP of RMB 260bn. Unless Tencent stumbles structurally with repeated strategic missteps, this is not expensive long term.

**Near-term positioning depends on clearer AI monetization pathways.** Hy4 is not domestic best-in-class yet, but it is table-ready. With Tencent’s ecosystem reach, user access is relatively easy; the key is a clear path and cadence to convert free users, helping investors gauge when the profit drag from AI spend bottoms.

**If management signals another year of aggressive spend and subsidies to nurture users, a further derating into sub-10x on sentiment is possible, given ongoing YoY profit pressure under heavy investment.**

China’s compute capacity should start ramping next year. With Iluvatar (Suíyuán) secured via Tencent channels and suppliers like Kunlun Chip scaling, the pressure from high-cost compute procurement may ease. We still need forward guidance on Capex allocation between onshore and offshore compute.

## Detailed analysis below:

## I. WeChat ecosystem traffic keeps expanding; AI spotlight shifts to WorkBuddy

Q3 WeChat/Weixin MAU reached 1.449bn, up 10mn QoQ; QQ MAU was 533mn, up 13mn. The social base remains solid.

The company noted Video Accounts’ total time spent rose over 20% YoY, with visibly higher engagement among younger users. This underpins strong ad growth.

Paid VAS subs were 253mn, down another 6mn QoQ, mainly from declines in Tencent Video membership, leading social network revenue to edge down to RMB 31.8bn. AI-wise, the spotlight was on WorkBuddy: by token consumption, it is already the most-used AI product domestically.

The consumer chatbot has not broken out, but Tencent has found an important position in productivity scenarios. On models, Hy4 (770bn total params; 49bn active params) launched in Nov, while Hy4 flash (224bn total params) targets lower inference costs.

WeChat’s 'Xiaowei' is now in beta with 100mn users, but that is still limited vs. the 1.4bn total — it remains in grayscale testing. A full rollout this year looks unlikely, with testing likely to expand instead.

WeChat is the largest entry point and Tencent’s trump card. When it truly opens up is the biggest swing factor — and the biggest cost variable. Given current C-end monetization, unless there is a B-end model to monetize downstream, treating it as a pure C-end entry could backfire.

## II. Games slowed: high comps + few new titles

Q3 total games revenue was RMB 66.5bn, up just under 5% YoY — the slowest since Q1 2024.

Domestic was RMB 47.6bn, +11% YoY, a notable slowdown vs. +17% last quarter. 'Delta Force' and 'Valorant: Radianite Ops' hit record DAU, and evergreen titles remained solid. Channel checks show 'Delta Force' still grew ~50% YoY, 'Honor of Kings' grew low single digits, and 'Peacekeeper Elite' slipped slightly.

**Incremental growth hinged almost entirely on 'Delta Force', which will soon face a tough second-year comp.** Intl was RMB 18.9bn, -9% YoY (-4% at constant FX). Last year had Supercell hits and new titles with one-off recognition, so the base was high — but the decline still exceeded expectations.

Deferred revenue at Q3-end was RMB 129.4bn, down over RMB 4bn QoQ, with YoY growth slowing from 11% to 6%, signaling softer spend across digital content, including games.

Back-solved VAS billings were ~RMB 94.1bn, -3% YoY. Ex-long-form video drag, game billings growth has slipped to low single digits. Q4 will see new titles, but none look like blockbusters. 'Honor of Kings: All-Chess' launched mid-Sept and may add some billings, but on a large base and with potential cannibalization of 'Honor of Kings' time, Q4 likely struggles to hold double-digit growth.

## III. Ads remain the growth pillar, but margins slipped

Q3 ad revenue was RMB 43.0bn, +19% YoY, driven by Video Accounts and green-bubble short-drama mini programs, while e-comm promo seasonality weighed on other ad slots.

By volume vs. price: WeChat MAU rose only 2% YoY, while ad revenue per user (the effective 'price') rose 16% YoY. Growth was almost all about 'price' — i.e., load and conversion.

Channel checks suggest Video Accounts ad revenue grew nearly 40% YoY in Q3, with current load under 6% and room to rise; Moments and mini programs grew 10%+. Thus, Video Accounts remained the main incremental driver, with green-bubble only launching late in Q3 and set to contribute more in Q4.

**In a weakening consumer environment, Tencent’s ad growth remains among the best in CN internet.** AI recommendations are lifting conversion, and Video Accounts load is rising from a low base, while short dramas add traffic and slots. But macro drag is mounting; do not assume high growth will persist into next year.

Q3 ad GPM was 52% vs. 57% in both the prior quarter and last year, with normal seasonality only 200bps — implying ~300bps from other factors. Management cited higher AI operating costs and content costs for short dramas (often 50:50 rev-share), and WeChat recently launched creator incentives.

**The flip side is that ad growth is getting 'expensive'.** In the past, opening load translated almost directly into profit; now you first spend on compute for recommendations and on content to build time spent. As the group’s 'buffer', ads can still flex revenue, but its ability to flex profit is weaker.

## IV. Tencent Cloud accelerated QoQ

Q3 FinTech & Enterprise revenue was RMB 63.7bn, with a slightly faster YoY growth vs. last quarter, mainly as Tencent Cloud accelerated from 20%+ to 35%.

Cloud was the bright spot. By our estimate, cloud is ~7–8% of total revenue; at +35% YoY, it added ~RMB 4bn of incremental revenue in a single quarter, ~70% of the segment’s YoY increase. By contrast, FinTech likely grew low single digits.

Management noted consumer lending and wealth mgmt grew, but commercial payments remain weighed by macro consumption. Near-term, we do not see much elasticity here.

Segment GPM was 50%, down 200bps QoQ. The reason is clear: GPU leasing and model inference carry lower margins than legacy cloud and payments. Faster cloud growth dilutes near-term margins, but the trajectory is healthy.

On monetization efficiency, management previously indicated paid-token GPM for WorkBuddy is around 60%, dragged by too many free users. Paid penetration among Tencent Cloud enterprise clients is still under 5%. The runway is long, but much of today’s usage is effectively 'given away'.

## V. Profit and cash flow fell, with heavier spend in COGS, Capex, and R&D

Q3 core OP (GP minus OpEx) was RMB 59.7bn, down 5% YoY, with a margin of 28.8%, nearly 400bps lower YoY. Compute investments were the largest headwind, including higher depreciation from elevated Capex and higher compute leasing booked in R&D.

Capex as a share of revenue rose from under 7% last year to 39%, which will depreciate over five years, dragging margins by 500–800bps per year. Bandwidth and server hosting fees have also accelerated for nearly a year, doubling in Q2 and Q3. Vs. 2023, compute spend is up by RMB 10–14bn per quarter, mostly in COGS for cloud clients’ inference, pressuring GPM.

Another item is depreciation/amortization from prior Capex, with compute-related depreciation typically booked in R&D for model training. Q3 R&D was RMB 32.4bn, +42% YoY; personnel costs were flat, while non-comp spend was RMB 15.4bn.

Pre-AI, this non-comp line was RMB 2–3bn per quarter. The incremental can be viewed as training compute — i.e., depreciation on GPUs and other chips procured via Capex, plus leased compute costs.

Sales expense was under 6% of revenue, up just 6% YoY, with promo and ad spend flat. Few new games this quarter reduced marketing needs, and WorkBuddy growth came more from free compute than ads.

Non-IFRS NP attributable to shareholders was RMB 66.5bn, down 6% YoY, broadly in line with Bloomberg consensus. -

Q3 recognized Capex was RMB 81.4bn, cash-paid-and-delivered Capex was RMB 65.1bn, plus ~RMB 86.0bn of compute prepayments. All-in, the cash outlay lens implies RMB 151.0bn spent this quarter to procure equipment — part delivered spot, part as prepayments.

Q3 operating cash flow was only RMB 2.6bn. Adding back RMB 86.0bn of compute prepayments, FCF was ~RMB 16.9bn positive. Since Tencent discloses Capex across three lenses per quarter, we elaborate below.

(1) Cash-paid and booked into PPE/CIP — i.e., paid this quarter and delivered. (2) Cash-paid but booked into non-current prepayments — i.e., deposits to lock compute hardware.

(3) Accrual-basis recognition — i.e., delivered equipment booked into PPE/CIP.

For cross-company comps, the cash-paid-and-delivered lens (Q3: RMB 65.1bn) is cleaner. But for Tencent’s own trend analysis, the accrual lens (Q3: RMB 81.4bn) is commonly used. PPE rose from RMB 149.9bn at year-start to RMB 242.5bn, while non-current prepayments surged from RMB 24.5bn to RMB 162.9bn.

**Different lenses aside, these will all turn into depreciation later, further weighing on margins.** From a cash lens, Q3’s RMB 151.0bn compute-related cash outlay front-loads pressure (requiring short-term funding), reflecting Tencent’s push to catch up in AI and secure the regulatory window; with this concentrated spend, cash flow should gradually improve.

\<End here>

Risk disclosure and statement: [Dolphin Research Disclaimer and General Disclosure](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

Dolphin Research on 'Tencent Holdings':

Aug 2026 earnings review: 'Tencent: All-out AI spend? One last step before a real comeback'

May 2026 earnings review: 'Tencent: No longer resting on laurels; AI is the only way'

Mar 2026 earnings review: ''Old' Tencent: time to shed old skin, AI heating up'

Nov 2025 earnings review: 'Tencent: No heartbeat games with Meta, just steady returns'

### Related Stocks

- [00700.HK](https://longbridge.com/en/quote/00700.HK.md)
- [80700.HK](https://longbridge.com/en/quote/80700.HK.md)
- [TCEHY.US](https://longbridge.com/en/quote/TCEHY.US.md)
- [TCTZF.US](https://longbridge.com/en/quote/TCTZF.US.md)
- [HTCD.SG](https://longbridge.com/en/quote/HTCD.SG.md)

---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**