Tencent's HK$4 Trillion Market Cap Tug-of-War: Gaming Revenue Sparks Debate, AI Office Agent Unexpectedly Tops Charts

After a single-day plunge of 7.05% on July 22, Tencent Holdings (00700.HK) saw its market capitalization once again approach the psychological threshold of HK$4 trillion. During intraday trading on July 23, Tencent's share price turned positive, with its market cap recovering to HK$4.08 trillion, stabilizing for the moment.

In fact, Tencent's market cap has been in a state of violent oscillation throughout 2026: it breached HK$5 trillion in March, fell below HK$4 trillion in June, and only reclaimed the HK$4 trillion mark on July 2. The curve itself represents the market's repeated repricing of the question: "When will AI investment pay off?"

Behind this tug-of-war lies intense market disagreement over the outlook for Tencent's gaming revenue and the expected returns on its artificial intelligence investments. Just as market anxiety was spreading, a third-party report revealed that WorkBuddy, Tencent's AI-native office agent, quietly topped China's desktop rankings in June, adding a new variable to the debate over "falling behind" versus "returns."

Gaming Data Decline Sparks Misreading, Citigroup Rushes to Call a "Buying Opportunity"

The immediate trigger for the July 22 sell-off was market concern over declining revenue from Tencent's core cash-flow business: gaming.

According to data from mobile app analytics firm Sensor Tower, global player spending on mobile games across the App Store and Google Play reached $6.1 billion in June, down 7% month-over-month. For the entire second quarter, global mobile game spending totaled $19.1 billion, a decrease of approximately $900 million compared to the same period last year. As Tencent is one of the world's largest game publishers, the weakness in the global market was quickly interpreted by the market as a negative signal for Tencent's gaming business.

However, Citigroup explicitly stated in subsequent analysis that this panic-driven sell-off represented a misreading of tracking data. Citigroup attributed the July 22 decline to a confluence of three factors: a market rotation of capital into AI hardware, concerns over a decline in Tencent's Q2 gaming revenue, and pessimistic expectations that increased AI investment by Tencent would drag on profits. Multiple factors combined to amplify negative market sentiment.

Citigroup further explained that, considering the deferred revenue recognition mechanics of the gaming industry and the strong performance of Tencent's PC games, the market's overreaction to quarterly data was irrational. The bank forecasts that Tencent's China domestic gaming revenue will still grow 8% year-over-year in Q2, and while a quarter-over-quarter decline of 3.9% is possible, seasonal fluctuations do not affect the fundamental strength of the company's gaming business. On this basis, Citigroup explicitly stated that the July 22 share price decline created a buying opportunity for investors.

Investment bank Bernstein also recently spoke out, rebutting some of the AI cost narratives circulating in the market. The firm bluntly stated that claims about platforms needing to indefinitely subsidize inference costs are "grossly inaccurate and excessively pessimistic."

From "Leaky Boat" to "Standing Firm," AI Monetization Path Still Being Explored

The core question hanging over Tencent has always been when AI investment will yield sufficient returns.

At the May annual general meeting, Tencent Chairman and CEO Pony Ma offered a candid response to prior skepticism using a "sailing" metaphor. He admitted: "A year ago we thought we had boarded the ship, but later discovered the ship was leaking. Now we feel we've found our footing, but we still can't sit down comfortably — we hope the ship can move a bit faster." These remarks acknowledged the detours Tencent has taken and the trial-and-error costs incurred in AI exploration.

Regarding how AI investment will translate into revenue, Tencent executives had already set the tone in earlier communications: monetization remains at a relatively early stage, and leading companies have yet to introduce very mature monetization models. This requires a long-term perspective, looking at what supplements might exist beyond the subscription model.

At the June Tencent Cloud AI Industry Application Conference, Tencent Senior Executive Vice President Dowson Tong directly addressed the assessment that "Tencent's AI is slow," acknowledging there were indeed areas for improvement, but emphasizing that AI is a marathon, where the competition is not about starting speed but long-term endurance in scenario deployment. Shunyu Yao, Tencent's Chief AI Scientist who joined from OpenAI late last year and proposed the ReAct agent methodology, pinpointed the core of the strategic pivot: the focus of AI's second half is shifting from pursuing model parameters to the collaborative design of models, products, and scenarios.

From a technology deployment perspective, Tencent's pace is accelerating. Bernstein noted in its report that Tencent's Hunyuan Hy3 large model has achieved significant improvements compared to the April preview version. The WeChat AI assistant "XiaoWei" began a gradual rollout to some users in June, with early feedback being positive. Officially launched on July 6, Hunyuan Hy3's smaller-sized models are claimed by the company to match the performance of flagship models with 2 to 5 times the parameter scale, and it was open-sourced under the commercially friendly Apache 2.0 license. API pricing is set at 1 yuan per million input tokens and 4 yuan per million output tokens, with cached input as low as 0.25 yuan. This pricing structure forms the cost foundation for a "low-barrier experience" on the product side. According to Tencent's disclosures, after WorkBuddy integrated Hy3, task success rates improved from 72% to 90%, and average processing time was reduced by 34%.

Data released by market research firm Analysys shows that in the second quarter of this year, visits to China's desktop AI-native office agent platforms increased significantly, with Tencent's WorkBuddy surpassing 20 million visits in June, ranking first in the category. Recently, WorkBuddy also launched a standalone app.

However, some market analysts have soberly pointed out that given WeChat's extremely large user base, the future AI transformation of WeChat will still face significant computational cost challenges.

The "Laggard's" Counterattack: WorkBuddy Tops the Office AI Track

While outsiders were still comparing which traditional office software — DingTalk, WeCom, or Feishu — had more comprehensive AI features, the core of competition had already shifted from "feature stacking" to "agent deployment."

According to a report by Galaxy Business Observer, Analysys' "China Office Agent Platform Market Insight Report for Q2 2026," released on July 20, showed Tencent's WorkBuddy topping the domestic desktop AI-native office agent platform rankings with 20.97 million visits in June. Its visit volume even exceeded the combined total of second-place ByteDance's TRAE IDE and third-place Alibaba's QoderWork. The ranking includes 17 mainstream domestic desktop products, with total visits in June reaching approximately 60.62 million. By comparison, WorkBuddy's monthly visits were just over 8 million when it launched in March.

Camp	Listed Products	Combined June Visits
Tencent-affiliated	WorkBuddy, CodeBuddy, QClaw, Marvis	32.62 million
ByteDance-affiliated	TRAE IDE (China version), Kouzi	~14.41 million
Alibaba-affiliated	QoderWork, Wukong	~9.19 million
Other products	Combined	~4.40 million

Data source: Analysys "China Office Agent Platform Market Insight Report for Q2 2026." At the individual product level, WorkBuddy ranked first with 20.97 million visits, while Alibaba's QoderWork ranked third with 7.88 million visits. Note: the ranking's statistical scope covers PC webpage visits — neither daily active users nor paying enterprise accounts — reflecting desktop entry-point mindshare rather than revenue.

In the more mature traditional collaborative office software track, QuestMobile data shows that DingTalk, WeCom, and Feishu together command a combined market coverage of 92%, effectively dividing up China's domestic market base. The three players' strategic divergences have widened further in this round of upgrades: DingTalk pursues an industry-specific Agent route, opening up agent capabilities for enterprises to connect with internal business systems, focusing on a tightly controlled "AI command system" and rapidly scaling through its base in traditional industries like manufacturing and government services. Feishu deepens its strengths in knowledge collaboration, using AI to connect documents, projects, and organizational knowledge bases, while also entering the developer space with TRAE IDE, with steadily rising penetration in internet and creative industries.

Tencent has chosen a differentiated dual-track approach: on the consumer side, WorkBuddy focuses on out-of-the-box task execution capabilities, capturing the desktop office entry point. On the enterprise side, WeCom anchors its core moat in private domain connectivity, embedding AI directly into customer engagement business processes — helping sales staff automatically organize customer needs and identify deal-closing bottlenecks, creating differentiation from competitors focused primarily on internal office efficiency.

Notably, WorkBuddy's lead does not rely solely on Tencent's self-developed models. The product itself is a model aggregation platform, simultaneously integrating multiple large models including Hunyuan, DeepSeek, Zhipu GLM, and Kimi. Users access these via a credit-based top-up system — regardless of whose model runs underneath, Tencent collects the fees. After Hy3 was first integrated into WorkBuddy on July 6, according to the Analysys report, over 60% of users who actively select a model chose Hy3 — aggregating external models while strengthening the self-developed foundation constitutes Tencent's two-pronged approach.

Just a year ago, Tencent was still mired in "laggard" controversy in this track. DingTalk and Feishu had long since deployed AI meeting summaries, intelligent approvals, and other features. Feishu's revenue surpassed 3 billion yuan (approximately $443.9 million) in 2025, with year-over-year growth exceeding 100% in Q2 2026, demonstrating strong momentum. Meanwhile, WeCom's AI feature rollout consistently lagged half a beat. Internally, Tencent also reflected on holding a strong hand — large models, cloud computing, and a social ecosystem — yet failing to produce a mass-market star product. Multiple teams including QClaw, Marvis, and WorkBuddy were racing internally, with resources dispersed and fragmented market presence.

The shift in thinking was followed by rapid execution. Tencent first completed internal resource consolidation, transferring relevant business and some teams from the QClaw product center to Cloud Product Division Six, merging them into the same system as WorkBuddy (the QClaw product itself continues to operate independently), clenching dispersed resources into a fist. After the Hy3 large model launch on July 6, model inference costs dropped significantly, directly paving the way for a low-barrier product experience. Tencent's Q1 2026 financial report showed that WorkBuddy, still in its early stages, already had active user retention exceeding 60% and paying user retention exceeding 80%. On the enterprise side, WeCom began internal testing of its intelligent assistant "Dayuan" in June, grafting AI capabilities — together with WorkBuddy Enterprise Edition and the Agent Suite — into the connection scenarios of 14 million enterprise organizations and over 750 million WeChat users.

Tencent is not alone in shifting from "internal competition" to "integration." According to Caijing magazine, Alibaba will integrate three Agent products — QoderWork, Wukong, and MuleRun — into "Qianwen Office," to be helmed by new DingTalk CEO Chen Yusen. Internally, ByteDance is advancing deep integration between Doubao and Feishu. This means the June rankings represent only the "first half," with significant potential for ranking changes in the next quarter.

Financial Fundamentals and Areas to Watch

Returning to the performance level, in the first quarter of this year, Tencent's total revenue reached 196.5 billion yuan (approximately $29.1 billion), up 9% year-over-year; gross profit was 111.3 billion yuan (approximately $16.5 billion), up 11%. Among this, China domestic market gaming revenue was 45.4 billion yuan (approximately $6.7 billion), up 6% year-over-year — as the 2026 Chinese New Year holiday fell later than in 2025, revenue growth lagged behind domestic gaming gross receipts growth. International market gaming revenue performed more strongly, growing 13% year-over-year to 18.8 billion yuan (approximately $2.8 billion). Marketing services revenue, supported by AI-driven advertising recommendation model upgrades, reached 38.2 billion yuan (approximately $5.7 billion), up 20% year-over-year, with the growth rate further accelerating from 17% in Q4 2025.

On the most contentious market debate — "AI investment dragging on profits" — Tencent's Q1 report already provided a scale reference.

Metric (Q1 2026)	Amount / Ratio	YoY
Total Revenue	196.5 billion yuan	+9%
Gross Profit	111.3 billion yuan	+11%
Operating Profit	75.6 billion yuan	+9%
Operating Profit (excl. new AI products)	84.4 billion yuan	+17%
Operating Profit Margin	38.5%	roughly flat
Operating Profit Margin (excl. new AI products)	43.0%	39.9% in prior year
Capital Expenditure	31.9 billion yuan	+16%

Data source: Tencent Q1 2026 Results Announcement (May 13, 2026). The difference in operating profit before and after excluding new AI products is approximately 8.8 billion yuan, corresponding to roughly 4.5 percentage points of margin — this is the direct basis on which the market prices "AI investment." Notably, the operating profit margin excluding AI actually rose rather than fell, increasing from 39.9% last year to 43.0%, indicating that the profitability of the core business has not weakened.

Tencent will report its second-quarter earnings on August 12. At that time, market doubts about gaming revenue fluctuations, the specific impact of AI investment on profit margins, and the commercialization progress of emerging businesses like WorkBuddy will all face a concentrated data-driven reality check.