Tencent Holdings Limited (700) Earnings Call Transcript & Summary

November 18, 2026

Hong Kong Stock Exchange HK Communication Services Interactive Media and Services earnings 81 min

Earnings Call Speaker Segments

Wendy Huang

executive
#1

Hello, and good evening to everyone. Thanks for being with us. On behalf of Tencent Holdings Limited, I welcome you to the webinar announcing our 2026 third quarter results. This is Wendy Huang speaking from the Tencent IR team. <<operator_script>> Please note that a recording is being made of today's webinar. Prior to beginning the presentation, we want to remind everyone that forward-looking statements are included in it; these statements are subject to a number of risks and uncertainties and might not come to pass in the future for various reasons. Information regarding general market conditions comes from a variety of sources outside of Tencent. Some unaudited non-IFRS financial measures are also contained in this presentation, and these should be considered alongside, rather than as a replacement for, measures of the group's financial performance prepared in accordance with IFRS. For a detailed discussion of risk factors and non-IFRS measures, please see our disclosure documents in the IR section of our website. Now let me introduce the management team joining us on the webinar tonight. Our Chairman and CEO, Pony Ma, will begin with a short overview. President, Martin Lau, will give a strategy review. Chief Strategy Officer, James Mitchell, will provide a business review; and Chief Financial Officer, John Lo, will finish with a financial discussion before we open the floor for questions. I'll now hand it over to Pony.

Huateng Ma

executive
#2

Thanks, Wendy. Good evening, and thank you all for being with us. The major progress we've made recently across models and applications has sped up Tencent's move into the new AI era. Hy4 has reached state-of-the-art AI model status in China, and Hy4 flash lowers certain inference costs to deliver cost performance for real-world use cases. By total daily large language model token consumption by users, we believe WorkBuddy has become the biggest AI application in China. Growth in our core services revenue picked up, driven by WorkBuddy, as we direct more compute toward Model-as-a-Service and GPU rental. Our core businesses remained resilient and robust, and a number of evergreen titles reached all-time peaks in DAU, while video accounts drove engagement up rapidly—particularly with younger users—by expanding its content offerings. Now, turning to our financial figures for the third quarter. Total revenue reached RMB 207 billion, an increase of 8% compared with a year earlier. Gross profit amounted to RMB 116 billion, up 7% year-on-year. If we set aside new AI products, gross profit would have been RMB 118 billion, reflecting an 8% rise year-on-year. Non-IFRS operating profit stood at RMB 68 billion, a decline of 6% versus the prior-year period. Setting aside new AI products, non-IFRS operating profit would have come in at RMB 83 billion, up 8% year-on-year. And non-IFRS net profit attributable to equity holders was RMB 67 billion, 6% lower than a year ago. Now let's look at our main communication and social network services. The combined monthly active users for Weixin and WeChat climbed to 1.4 billion, up both year-over-year and quarter-over-quarter. On the gaming side, Delta Force exceeded 58 million peak daily active users in October, which demonstrates how we can keep building a new game's popularity for years after it launches. Among productivity tools, WorkBuddy has emerged as the most used AI product in China when measured by large language model token usage. On the foundation model side, we launched the production version of Hunyuan 4 earlier this month. At the time of release, it placed first among Chinese models on the Agent Arena leaderboard, reaching China's state-of-the-art level. I'll now turn things over to Martin for the strategy review.

Chi Ping Lau

executive
#3

Thanks, Pony, and hello everyone, morning and evening wherever you are. Over the last 3 months, our AI-native businesses have made breakthroughs that truly transform them. Today I'd like to share some significant and exciting news about what we're doing in intelligence, applications and monetization. Let me begin with intelligence, meaning foundation models. We think the newest generation, Hunyuan Hy4, now stands at state of the art among China's foundation models, and it offers outstanding usefulness for real-world productivity. Among models of its size, Hy4 leads the category, and its abilities reach those of models that are even bigger. Take Hy4, for instance: on the Agent Arena leaderboard it ranked number one among Chinese models right out of the gate, and it scored highly across a wide range of benchmarks even though we did no optimization aimed at any specific one — which is pretty typical for plenty of other models. With Hy4, our team did a major scaling up of both pretraining and post-training, and we built in vision along with computer use abilities. As a result, compared with our previous models, Hy4 delivers big gains in following instructions, understanding long context, handling long horizon tasks, and deep research. And to bring inference costs down even more and drive broader uptake across additional real-world use cases, we released a tailored lightweight model called Hy4 flash. Hy4 flash is priced at one-twentieth of Hy4. Usage came in much higher than we had anticipated, and of all the models released this year, it drew the largest daily token consumption on Open Router. Hy4's capabilities please us, and so does the pace of its improvement, and we believe our intelligence journey remains at an early stage. Through putting Hy models into key products, WorkBuddy among them, we collect valuable user feedback that supports ongoing joint design of models and products and a positive iteration loop, especially for real-world scenarios; we have also lately consolidated multimodality development within Hunyuan; we hold the industry lead in 3D generation, and we are enhancing image generation while also building out video generation capabilities. Moving to applications: we have likewise seen breakout success in productivity AI with WorkBuddy. WorkBuddy has expanded at a tremendous pace, reaching more than 10 million daily active users and going through over 100 trillion tokens per day; in our view, it is now the top AI application in China when measured by large language model token consumption. WorkBuddy is also far and away the front-runner in the productivity AI space. Based on outside surveys, our internal figures show its advantage over competitors has widened in recent months. Although we had previously been careful about how fast the enterprise software market would grow in China, we now think that incorporating AI features will let spending on productivity software rise quickly and significantly, with WorkBuddy serving as both a major catalyst and a beneficiary. Here is the reasoning: First, AI services sidestep the issue of software piracy. Second, WorkBuddy serves many high-value enterprise scenarios—drafting documents, analyzing data, managing projects, doing research, and developing software—so the overall offering justifies its price. With monetization built on subscriptions plus token-based usage, buyers get into a paid experience through a modest flat entry cost and then increase spending as the value they gain grows. During the last seven days, WorkBuddy's annualized gross revenue went past USD 500 million—over twice what Tencent Meeting reached, a product launched seven years earlier. To some degree, what we're seeing mirrors the evolution of China's online gaming sector, where cloud connectivity similarly reduced piracy, and item-based monetization unlocked a much larger revenue opportunity once spending could grow in line with demand. We've already seen enterprises boost their outlays considerably when forward deployment engineers demonstrate how usage generates value. So we're working alongside partner companies to provide FDE support for WorkBuddy customers. Since the enterprise edition debuted in June, 60% of China's top 100 companies have become WorkBuddy customers. That strong initial traction is very encouraging, and we believe it's just the tip of the iceberg. At the moment we are putting money into growing how many people use WorkBuddy, and part of that involves giving free service for a while to particular groups of users, teachers among them. Even so, we have every confidence that as time passes we will keep making WorkBuddy's unit economics better. The ways we'll do that are lifting the share of users who pay, allowing each paying user to spend more on tokens, and pushing token costs lower via inference optimization, tailoring models, and efficiency gains in compute. We also think that connecting WorkBuddy with Tencent's ecosystem and with outside ecosystems can improve retention for WorkBuddy users, raise lifetime value, and strengthen the economics overall. WorkBuddy delivers depth when it comes to AI usage, while Xiaowei—the personal assistant built into WorkBuddy—will help AI reach widely and become something every Internet user in China can access. Testing for Xiaowei started in June, and we have kept expanding the pool of users trying it out, which has now hit 30 million DAU. We have introduced additional Xiaowei touch points across moments, official accounts and video accounts, which unlocks more usefulness in a broader set of scenarios and boosts interaction per DAU. At present we are bringing major mini programs and merchants on board and linking up with mini shops to make transactions easier. This works because Xiaowei is built on the purpose-built VLM model, tailored for cost efficiency and tuned for the Weixin ecosystem. Running Xiaowei doesn't cost much in terms of inference, and that figure is going to keep dropping. As a matter of fact, suppose Xiaowei were rolled out to all 1.4 billion Weixin users — even then, we'd anticipate Xiaowei's quarterly expense remaining lower than what Yuanbao was spending per quarter earlier this year. Each year, as compute efficiency multiplies, Xiaowei's capabilities will improve substantially. Eventually, certain inference workloads can run on users' own devices, bringing costs down even more. Because Weixin already has a network of merchants spanning payment and Mini Program services, Xiaowei can integrate powerfully with merchants and service providers at very little cost, producing synergies with them rather than competition and building win-win relationships. On the monetization side, we expect Xiaowei to serve as a multiplier that we apply to Weixin's already substantial revenue streams, compounding them further and faster. Advertising and transaction-based revenues present immediate opportunities for us, and over time we can introduce new revenue streams, including agent-to-agent facilitation services revenue from merchants and service providers. We are in the early stages of adding compute capacity at an accelerated pace, and we are making a portion of that capacity available to external paying customers through Tencent Cloud. Consequently, our cloud services business is now beginning to deliver financial benefits from our AI initiatives, initially from WorkBuddy, Model-as-a-Service and GPU rentals. During Q3, revenue from cloud services climbed 35% compared with a year earlier, an acceleration from 21% year-on-year in Q2 and the teens-level percentage year-on-year growth seen across 2025; WorkBuddy's third-quarter revenue came in at more than 5x what it posted in the prior quarter, fueled by a rising number of paying users and solid early uptake among enterprises. On Model-as-a-Service, revenue rose more than threefold quarter-on-quarter as we improved inference efficiency and tightened operational controls plus security guardrails for enterprises, giving them oversight of how their AI is deployed. Revenue from our GPU rental services increased by over 30% year-on-year — and quarter-on-quarter, thanks to enterprise demand for inference and better scheduling efficiency across our compute capacity. With that, I'll pass things to James for the business review.

James Mitchell

executive
#4

Thanks, Martin. Turning to the quarter, overall revenue was up 8% versus the prior-year period. Social networks contributed 15% growth, domestic games 23%, international games 9%, marketing services 21%, and FBS 31%. Gross profit increased 7% year over year, and within that, VAS gross profit was up 7%, marketing services 10%, and FBS 8%. Value-added services revenue came in at RMB 98 billion, up 3% year over year. Within that segment, social networks revenue declined 1% year over year to RMB 32 billion, driven by a drop in long-form video subscription revenue amid fewer top-tier releases, while app-based game item sales and audio subscriptions both grew. Domestic Games revenue rose 11% year over year, largely driven by Delta Force, VALORANT Mobile, and Honour of Kings. International games revenue fell 9% year over year, which comes to 4% on a constant-currency basis. That reflects a tough comparison base for some Supercell titles as well as the release of Dying Light: The Beast during last year's third quarter. Moving on to communications and social networks. Within Weixin, Green Bubble, our portal for mini drama series, posted fast user growth. It offers free-to-watch mini dramas that benefit from Weixin's social recommendations and from premium content, such as China literature novels turned into mini dramas. With WorkBuddy, developers can now automate the way they build and release mini programs, which brings down entry barriers and cuts development timelines. On video accounts, we boosted participation among younger users by collaborating more deeply on content with our in-house game music and TV services. We introduced an AI harness aimed at creators, which turns articles from their official accounts into videos and handles storyboarding, scripting, and media generation automatically. Year-over-year, total time spent on video accounts climbed over 20%, and we believe there is additional room to grow, since time spent per DAU is still below half of what industry leaders achieve. Turning to domestic games, Delta Force lifted its DAU by more than 50% year-on-year during the quarter, reaching a lifetime high; we enhanced the game's core extraction experience by adding a new map and introduced a new 5b5 tactical shooter mode. For VALORANT Mobile, both DAU and gross receipts reached lifetime highs, supported by popular blossom-themed and Chinese folklore-themed weapons. Honour of Kings also saw its average DAU climb compared with a year earlier, supported by a fresh aspect system that puts a new spin on classic heroes and by an arcade mode built for speed. Gross receipts for the title went up too, thanks to effective IP partnerships and dark fantasy costume designs. September marked the release of Honor of Kings: ACE, a hero card battler within the Honor of Kings universe, available on both mobile and PC. From its debut onward, the title has placed inside the industry's top 10 mobile games when measured by average DAU. Turning to our international lineup, VALORANT posted year-on-year growth in gross receipts during the third quarter, fueled by stronger player engagement and by the Champions 2026 collection. At the VALORANT Champions World Finals in Shanghai, we turned the event into a pop culture festival spanning games, fashion and outdoor experiences. Working alongside the creator of Steal a Brainrot, we brought that IP into PUBG Mobile's World of Wonder sandbox mode. The resulting map, Brainrot Royale, delivered the largest DAU increase the sandbox mode has recorded since its debut, and 30% of PUBG Mobile's DAU engaged with the sandbox mode. Monster Hunter Outlanders arrived on October 29 as a mobile action RPG, offering impressive monsters, the franchise's signature combat feel and a fresh adventure mode, and in week one it placed among the top 10 mobile games worldwide by downloads. Regarding Marketing Services, despite challenges across the industry, revenue increased 19% compared with the prior year, reaching RMB 43 billion, driven by gains in both impressions and pricing. Categories including consumer goods and local services contributed significantly to that expansion. We introduced AI-driven advertising offerings aimed at content publishers. For instance, we enabled mini drama providers to insert audio ads of ideal duration at ideal moments, which reduces interruption to viewers while enhancing monetization efficiency. Such AI solutions currently support about 80% of ad monetization for mini drama publishers. Ad impressions for Video Accounts increased by more than 50% compared with a year earlier, supported by video views and ad loads. We now deliver advertisements according to our content ranking system, which improves how relevant our ad recommendations are and boosts monetization. In addition, the Green Bubble Mini Drama gateway contributed significantly to marketing services revenue growth inside Mini Programs. Revenue for the FinTech and Business Services segment reached RMB 64 billion, an increase of 9% year-on-year. FinTech services revenue rose by a low single-digit percentage year-on-year, primarily because revenue from consumer loan and wealth management services went up. Commercial payments volume edged up year-on-year as transaction count increased, partly offset by a lower value per transaction amid softness in overall consumption. We introduced a payment app called TenPayGo aimed at foreign visitors to China. It brings together support for overseas cards and digital wallets along with transit ride codes, tax refunds, and AI-driven translation services. In Business Services, as Martin mentioned, revenue from our cloud services grew 35% compared with a year earlier, driven by stronger demand for AI-related offerings and for our cloud services generally. We have folded WorkBuddy into WeCom, which lets agents draw on authorized workplace context to finish tasks, share results across the organization, and convert organizational knowledge into capabilities that can be reused. I'll now hand things over to John.

Shek Hon Lo

executive
#5

Thanks, James. Turning to Q3 2026: our top line was RMB 207.3 billion, an increase of 8% versus a year ago. Gross profit came to RMB 115.9 billion, which is up 7% year-on-year. Operating profit was RMB 59.1 billion, down 7% year-on-year. We booked interest income of RMB 4 billion, 6% below last year's level, as the average interest yield came down. Finance costs were RMB 3.5 billion, compared with RMB 3.8 billion in the prior-year period; that reflects helpful ForEx movements, partially offset by higher interest expenses tied to increased debt. Our share of profit from associates and JVs was RMB 4.5 billion, versus RMB 7.8 billion of profit share in the same period last year. On a non-IFRS basis, the share of profit from associates and joint ventures came in at RMB 8.8 billion. That compares with RMB 10.3 billion in share of profits for the same period a year earlier. The main reason was lower profit from certain associates in the China Internet industry, after factoring in market consensus estimates and true-up adjustments. Income tax expense declined 7% year-on-year, to RMB 9.1 billion, primarily because of reduced tax expenses connected to certain overseas subsidiary. Without that impact, our income tax expense would have been higher year-on-year. Looking at the non-IFRS financial figures, operating profit stood at RMB 68.1 billion, a 6% decrease year-on-year. When new AI products are excluded, operating profit was RMB 83.3 billion, which represents an 8% increase year-on-year. Profit attributable to equity holders came in at RMB 66.5 billion, a 6% decline year-on-year. Diluted EPS stood at RMB 7.27, 5% lower. Now on to third-quarter gross margin. The overall figure was 56%, roughly flat compared with a year earlier. By segment, VAS gross margin climbed 3 percentage points year-on-year to 64%, helped by a favorable mix shift toward high-margin games developed in-house. Marketing Services gross margin reached 52%, a drop of 4 percentage points year-on-year, reflecting AI-related operating costs and content costs tied to building out our mini drama service. For FinTech and Business Services, gross margin came in at 50%, essentially flat compared with a year earlier. Turning to Q3 operating expenses: selling and marketing costs were up 6% year-on-year, reaching RMB 12.1 billion, driven largely by promotions aimed at boosting uptake of our AI native offerings, WorkBuddy among them. Research and development spending rose 42% year-on-year to RMB 32.4 billion, mainly due to increased R&D investment in enhancing our AI model family, pushing forward Weixin AI initiatives, and building out AI capabilities across our products and services. G&A, excluding R&D, increased 2% year-on-year to RMB 11.7 billion, reflecting our discipline in managing non-AI-related expenses. Our non-IFRS operating margin for the first quarter was 33%, a decline of 5 percentage points year-on-year. Excluding new AI products, the non-IFRS operating margin stood at 40%, roughly stable year-on-year. As I wrap up, I want to run through a few important cash flow and balance sheet figures. Capital expenditure statements tied to operations reached RMB 63.6 billion, a 257% increase compared with a year earlier, reflecting our stepped-up spending on AI infrastructure — funding improvements to the Hunyuan model family, supporting WorkBuddy inference requirements, building out AI capabilities throughout our products and services, and satisfying outside demand for our cloud offerings. Payments for nonoperating CapEx totaled RMB 1.5 billion. Free cash flow landed at negative RMB 69.1 billion, a figure driven mostly by substantial compute purchases. Excluding the prepayments associated with compute procurement, free cash flow would in fact have been positive, at RMB 16.9 billion. Regarding net debt, our position was RMB 13.7 billion, compared with net cash of RMB 58.2 billion on 30th of June 2026 — the change largely comes from the prepayments I referred to earlier, partially offset by RMB 8.4 billion of net cash inflows from net divestment activity within our investment portfolio. As of the end of September 2026, the fair value of our stakes in listed entities, excluding subsidiaries, stood at RMB 505.4 billion, and the carrying book value of our holdings in unlisted investees, excluding subsidiaries, came to RMB 397.2 billion.

Wendy Huang

executive
#6

<<operator_script>> The first question comes from the Robin Zhu from Bernstein.

Robin Zhu

analyst
#7

I suppose I have two questions. First, we've had 2 quarters now with very heavy CapEx and prepayment spending. I'm curious where Tencent stands in this spending push to acquire compute. How far along would you say we are toward reaching a good compute level, by whatever measure you use? And then how do the tighter U.S. restrictions affect both current and future compute deals? Does that also affect the acceleration we saw in cloud and other parts of the business going forward? Second, WorkBuddy has performed very well. I'd like to know whether management could size the opportunity ahead in terms of user traction, whether it's mainly power users who make up the addressable market, and the revenue and profit potential. Could the semis restrictions put this at risk?

James Mitchell

executive
#8

Robin, let me handle the first group of questions about prepayment CapEx and, I think, the effect of advanced semiconductor restrictions on cloud revenue growth. On prepayments, the second and third quarters were unusually high, and we've now entered the long-term contracts we intended to enter. So our prepayment outflows will come down; in fact, some quarters may see prepayments reverse because of contract cancellations—contracts we prepaid earlier but are now exiting following recent changes. As for CapEx apart from prepayments, the present run rate—we view it on a cash CapEx basis, which I believe was RMB 65 billion rather than the RMB 81 billion accrual CapEx, since if you combine prepayments and CapEx, you should pair prepayments with cash CapEx. Otherwise you'd be double counting by using prepayments and accrual CapEx. So in any case, at the cash CapEx level, we believe that's roughly at peak levels, and cash CapEx may fluctuate around this peak through the remainder of 2026 and 2027 on a quarterly basis. After that, we expect it to come down. Regarding how the restrictions affect our cloud business, we actually don't see any impact, because our cloud business today mainly runs inference, particularly inference for highly efficient models like Hy4 flash and other flash models. We're now at a point where that inference can run very effectively on a surprisingly small number of domestic chips. So I think over the last 12 months there have been some transformations in inference productivity. As a result, we can drive a much higher volume of tokens on domestic chips specifically than we would have imagined a year ago. And it's inference that drives cloud revenue. So in fact, we believe our cloud revenue will not be substantially hurt by the changes.

Chi Ping Lau

executive
#9

Okay, I think there are a few points to add to what James covered. First, on CapEx: the capital we're spending at present is really aimed at infrastructure outlays so that our AI native business can get off the ground. It's somewhat similar to constructing a factory so you're able to run training and produce tokens. The way it works is that once you've got that in place, you can go on using the same facility to train a new generation of models and also keep generating tokens. That's the reason that after we've made that outlay in '26 and '27, as we move into '28 the trend should improve given this characteristic, since you're able to keep using the infrastructure we've paid to build during 2026 and '27. Second, the pricing we're actually seeing in '26 and '27 is elevated. As we get into '28, even if you need to purchase the same item, you ought to expect the price to fall fairly substantially. Third, and I think this matters most, now that we've worked on training and inference for our models, we're seeing compute efficiency multiply rather rapidly year-over-year. That means we'll be able to achieve better training efficiency and also a far greater number of tokens as the years pass with the same facility. So given these factors, we actually view '28 as much better, and beyond that too. Now, regarding the effect on our AI business from the newest tighter restrictions, I'd call it unfortunate, but there are a few points. First, it does hold everyone back. It holds everyone back in China, so everyone has less compute, less advanced compute, to work with, regardless of when you began leasing or building capacity outside of the China. Second, where training is concerned, it will be affected. Yet our training efficiency with Hy4 is in fact very high, and for the next version, it would probably take us a longer time to deliver that next version in the short run. But over time, the domestic supply chain should come on board. On inference, as James said, the models we've built are in fact very efficient. As a result, it shouldn't be a problem for the domestic supply chain to cover it, and regarding WorkBuddy as a product, WorkBuddy runs on very efficient models to begin with, but it also runs with a whole range of different models, and it receives a very diversified supply of tokens, some of which we provide ourselves. A portion of those are in fact supplied by the model companies — the different model companies that WorkBuddy hosts. And in fact, WorkBuddy has turned into the largest customer for nearly every open weight model in China. Over the longer term, we think all Chinese players would really need to depend on domestic supply chains along with domestic chips, and we do see the industry moving forward on that very quickly. With the most recent restriction, we expect that to accelerate things even more, since all development effort would really need to be concentrated on accelerating the domestic supply chain. Now when it comes to WorkBuddy, we're extremely, extremely excited about the growth we've seen lately. And if we look at the AI business, the way AI business differs from the Internet age is that in the Internet era, if you double your business — and that was actually regarded as a really excellent outcome — in the AI era, you typically see a very successful product or model grow 10x; and I think WorkBuddy genuinely belongs in that category. So we're very pleased to see that kind of growth traction. And WorkBuddy is at scale in terms of both DAU and DAU, and DAU keeps rising very quickly. It's already the biggest AI application in China measured by tokens delivered. If you view the AI business on an intelligence delivered basis — tokens serve as a decent proxy for that — it is already by far the biggest AI application in China. And token growth is also rising and speeding up. We really see it as a great model, a new model in the enterprise market. We laid out that there's no piracy. You really can't have a small team of people building a mixed shift WorkBuddy inside the organization. And WorkBuddy essentially delivers a whole package of different functionalities. So it isn't just one piece of software; it's almost like an entire package of software functionalities handed to you. And I think probably most importantly, it's really a great business model as we begin working through monetization, in that the free tier genuinely lets you acquire users very effectively. Then the paid tier really lets you get some people to pay and gives them a fairly substantial amount of token. And then for the truly heavy users, they keep adding spending on tokens, and you've seen users spend as much as RMB 40,000 per month buying tokens. So when you look at that, it's really quite different from SaaS businesses in China in the past, where you could only charge a flat fee through subscriptions. So we really feel good and excited about WorkBuddy. And as time goes on, I think — first of all, it holds a fantastic position in the market, and when people think about productivity and AI, they think about WorkBuddy. As it gathers more and more data, that genuinely improves our model as well as the models that work alongside us. At the same time, it lets users build up more and more skills and keep more and more of their work products in WorkBuddy's memory. And as it becomes part of how businesses run their workflows, we believe it will be very sticky; so we're really enjoying the phenomenal growth of WorkBuddy.

Wendy Huang

executive
#10

We will take the next question from Alex Liu from Bank of America.

Alex Liu

analyst
#11

I've got two questions. First off, at the recent Tencent 28th anniversary, I believe management discussed organizational change. Could you share a bit more about the strategic thinking behind that, and how broad the changes should be across the different business units and in resource allocations? And where do you expect the earliest financial impact to appear?

Chi Ping Lau

executive
#12

Well, we're at a very early stage of this, right? But the insight we've seen is that, number one, AI really can be applied across various businesses to boost the efficiency of how those businesses are run. But to be able to do that, you actually sort of need a pretty significant reorg of the organization, and you also need to reengineer a lot of the work processes. So if we can actually pull that off, right, then existing businesses can actually run with fewer people, right? But at the same time, what we're seeing is that AI actually lets small teams leverage AI to explore new ideas, build new businesses and build new growth drivers, right? So if we can actually run an existing business with fewer people and actually sort of redeploy those people to build new growth drivers for the existing business as well as to build new businesses, right, then I think that would be the ideal, right? And that's the whole idea behind this. And as we build our AI native business, we get more and more excited about this opportunity, because we see it in our model team. We see it in our WorkBuddy team. We also see that when WorkBuddy is applied to our customers, it actually helps improve efficiency in a great way, right? So that's why we basically say this is a time for us to make bigger changes to the organization; and if you ask when — what are we seeing already, right? I think the fact that our model team was actually completely rebuilt from scratch, and within a year they were able to push our model to a top-tier position in China — that's one example. I think the WorkBuddy team was really sort of nobody, right, 6, 7 months ago, and they built such an exciting business. I think that's another example. And we hope to see more and more of that over time.

Wendy Huang

executive
#13

We will take the next question from Alex Yao from JPMorgan.

Alex Yao

analyst
#14

I have two questions. First, regarding the global AI footprint: as you bring Hunyuan and WorkBuddy to users beyond China and expand the cloud business internationally, how do you decide where AI workloads are best hosted, whether on domestic infrastructure or in international regions? And how much flexibility do you have to shift that mix when conditions in different markets change? My second question concerns the economics of WorkBuddy. On one hand, we're pleased to see user numbers and token consumption growing very quickly; on the other hand, many highly efficient yet very cheap flash models have been introduced into WorkBuddy, which arguably will lower the overall blended pricing for token consumption there. So could you help us understand the economics between volume growth and pricing decline? And do you still see plenty of opportunity to ramp up monetization even as the per-workload cost or price falls?

Chi Ping Lau

executive
#15

On where hosting should happen right now, I think that's a very complicated question, right, because there are restrictions on GPU usage. But I would look at it this way: if market opportunities exist, we'll be able to figure out a way to solve it, right? So right now we see a fantastic growth opportunity for Hunyuan 4 and WorkBuddy in China, but we also see a pretty good chunk of Hy4 usage actually coming from outside China, right? To serve that, we could handle a portion of it from inferencing capacity in China. Over time, we actually sort of want Hunyuan 4 promoted around the world, and we may find partners to do that, right? And there are inferencing partners around the world who can actually do that, so I think that's one aspect. As for WorkBuddy, over time I think we see great opportunities for growing that outside China as well. And it's software. It can connect to different models, and as a result, right, it doesn't necessarily require inferencing infrastructure to support it, because it's very diversified in terms of token supply. So I think that's that. And regarding cloud, right, if it's non-GPU related, right, we can actually build capacities ourselves outside China. I think there are still a lot of such opportunities, because our cloud service is actually very cost competitive. When it comes to GPU, right, then unfortunately we can't actually offer high-end GPU anymore, right? That said, we might be able to collaborate with certain partners in various locations, so I believe that is — should we identify business opportunities, and we do believe that for both Hunyuan 4 and the Hunyuan family of models as well as WorkBuddy, right, and cloud, we could uncover plenty of growth avenues beyond China given the efficiency, given the cost competitiveness, given the quality, and then we would be able to secure the infrastructure needed to back it up. Regarding how flash models get used inside WorkBuddy, in my view flash models are excellent customer acquisition models, right, since they let us provide very cost competitive tokens to free users, right? And then as time passes, when free users genuinely want additional capability, right, there's a path for them to move into subscription. And then as time passes, when people kind of need to handle increasingly complicated work, right, they will graduate to more capable models, right, and then they will spend more on tokens, right? I think the flash model is genuinely a great method for acquiring customers, while the more capable models are great methods for monetizing further by delivering greater value to users. This in fact kind of aligns perfectly with what we stated about WorkBuddy, namely that it actually monetizes at every one of the different points along the demand curve. So it is only helpful in that respect.

Wendy Huang

executive
#16

We will take the next question from Colin Liu from BNP Paribas.

Colin Liu

analyst
#17

There are a couple of things I'd like to ask. First, on international games: at what point do we expect revenue from international games to return to positive growth, given that GTA is absorbing so much attention and players' mind share? Do you believe that will in fact push back when the inflection point might arrive? My second question ties to one already raised. You mentioned the reorganization will take time, yet AI investment is currently hitting OpEx and dragging on overall profitability, so is there any nearer-term cost-saving effort or planning you could share with us right now?

James Mitchell

executive
#18

Colin, on your first question about international games, I'd say in fairness that our slowdown came before the build-up to GTA VI and stemmed from several internal and external factors that we are working to address; going forward, we think we hold a solid number of evergreen games, several still early in their growth curve. We have plans for new games that we hope and believe will turn evergreen. And after a stretch this year with none of those premium games, next year we will also go back to a more normal cadence of more upfront sales-driven games. So that covers international games. More broadly, the global game environment is one where innovation has slowed to some degree. And with AI now, there are opportunities to speed innovation back up. At present that centers more on producing in-game content and on preproduction for new games rather than full production and post production for new games. But overall, we have led when it comes to bringing innovation to game technology, to game business models, to full game play, both in China and more recently internationally. That is what we will aim to keep doing; and if we succeed at it, then we believe our international game revenue growth will take care of itself.

Chi Ping Lau

executive
#19

On the efficiency question, again I want to restate how we view our current operations. We split things into two buckets. There's the existing core business, and there's the AI native business. With an AI native business, it's only natural to pass through an investment phase before revenue starts coming in and then profit follows. We've witnessed that pattern many times at Tencent. Still, given how large the investment is this time, people feel somewhat anxious about it. That said, we're genuinely excited by what we've observed and lived through over the last 6 months. Our models and our products have made a genuine breakthrough, and we can now say this business is on track to become very valuable to us over time. Turning to the existing businesses, we believe that with AI arriving and what AI can do to enable things, plus what we've learned from building the AI native businesses, there's a great deal of efficiency to be gained inside the core businesses. It will come both from extracting greater efficiency and from letting teams create more revenue and growth opportunities. That's another initiative we'll be driving forward. We also see strong opportunities here in the period ahead. That's our view on your reorganization question.

Wendy Huang

executive
#20

We will take the next question from Kenneth Fong from UBS.

Kenneth Fong

analyst
#21

Two things I'd like to ask. The first concerns Xiaowei. Since we've started pilot testing and are going deeper on integration with mini programs, could you share any initial feedback from users? And what milestone must we hit before we roll it out more widely? Over the longer term, how should we view the monetization timeline for Xiaowei, the TAM, and the route to monetization through advertising or perhaps subscription, especially given early signals from consumer agents such as Meta Muse? My second question is about the advertising gross margin, which narrowed significantly this quarter because of AI investment as well as a larger mini drama mix. As we think about the ad margin trajectory going forward, what's the right way to model it, and in particular, how soon could inference efficiency improvements stemming from the AI investment be enough to counteract an AI cost drag of this kind? And regarding the mini drama content costs, what portion of that impact would represent a structural drag on margin in the near term?

Chi Ping Lau

executive
#22

Regarding Xiaowei, we have been making it available to an increasing number of users in stages. We have also been adding more touch points inside Weixin, which now number more than 30, such as Moments, Official Accounts, and Video Accounts. What we have seen is that as these touch points are added, an increasing number of people begin to use them. Interactions per DAU keep climbing too. Xiaowei now sees 30 million DAUs, and the count of interactions for each of those DAUs has been going up as well; we sensed that users broadly find it quite handy. Utility has improved across social and content scenarios, and there's plenty of frictionless use at convenient touch points; even so, many people — if you set it beside meals — would say it isn't as strong as Muse, which is intentional, correct? Since our actual aim is for Xiaowei to launch first not as fully capable agents offered to a limited group of people prepared to pay for it, right? Our goal is for Xiaowei to reach every Weixin user, and as it links up with the various touch points and with the ecosystem — mini programs included — it will keep getting more capable. That said, we need time for more mini programs to support CLI interaction through our Xiaowei integration, so many of these capabilities will be built up gradually. There's also the cost side to consider: if we're going to offer this to all Weixin users, cost becomes a real question. That's the reason we rely on a compact model, and we're confident that even if we roll it out to all users, right, it will not be such a big cost, right? Actually, the quarterly cost, as we mentioned, will be lower than the ramp-up cost at the start of the year, even if we roll it out to every user; then the question becomes what direction do we take from here? The answer is that we see a clear tailwind, since compute efficiency is rising quite dramatically. In fact, it multiplies by several times annually. So with each upcoming Xiaowei version released, you can expect capability to leap several times over. In a year or a year and a half, Xiaowei's abilities are going to be much stronger than what we see now. Eventually, it'll be just as capable as news. Yet we're able to reach every user within Weixin. That means we can provide this service to everybody at a price we're able to handle. Once we get to that point, we think it would serve as a huge multiplier for the value of the Weixin ecosystem, since users will be engaging with their agents all the time, and it will be woven into their everyday routines tightly and automatically, and they'll be able to carry out transactions, find content, manage schedules in some way, and get many things done within Weixin just by interacting with Xiaowei. When we consider Weixin’s existing revenue sources, taken together they total several hundred billion RMB annually, and a boost of ten or twenty percent on top of that would represent a substantial sum, which would make Xiaowei extremely profitable in our view. That, I’d say, is the big picture. And we expect to have a clear view of it in the coming six to twelve months.

James Mitchell

executive
#23

On your second question, Kenneth, about the advertising gross margin: for many quarters now, we've been steadily improving our ad tech stack. Sometimes an upgrade's cost lands at the beginning of a quarter, and the benefit doesn't show up until later in that same quarter or the next one. A bit of that occurred this quarter, but it isn't new information. What was genuinely new this quarter was the success — really the breakout success — of mini dramas in the Green Bubble gateway; and on the positive side, that contributed substantially to our advertising revenue. The downside is that we're still in the very early days of getting our mini drama ecosystem truly off the ground, and of persuading a number of independent mini drama studios — studios whose content can't be found on our main competitors' apps — to bring their content into Green Bubble. As a result, some costs are front-loaded as we work to fuel that mini drama traffic growth and revenue growth. Mini drama also holds appeal over the longer term, since alongside mini games it demonstrates how we can kind of activate these new closed-loop activities inside Weixin, and for quite a while now games have been both a major driver of in-app purchases and a major source of WeChat advertising revenue for us. Advertising contributions from mini dramas are now keeping pace as well. The largest opportunity—still something we're working toward—sits with mini shops and shifting toward capturing additional e-commerce through a closed-loop approach. Conversely, absent the unexpected boost from mini dramas, our ad revenue growth would have been slower; that, I believe, simply reflects a broadly challenging macroeconomic climate, and in particular a tough backdrop for consumption and advertising. Even so, we've kept growing throughout that climate. We've kept outpacing the market. Thanks to AI, closed-loop initiatives like mini dramas, and increased traffic for video accounts and similar channels, our goal is to continue expanding faster than the broader advertising market.

Wendy Huang

executive
#24

We will take the next question from Alicia Yap from Citi.

Alicis a Yap

analyst
#25

I have two questions. The first one: we've seen recent progress from certain frontier models — OpenAI, Gemini, Meta, and the Chinese labs as well — which seems to indicate that advanced capabilities in interactive storytelling, in word generation, and in AI-driven characters for gaming environments are emerging. So does Tencent see these developments as a competitive danger to traditional gaming engagement, or as a possible catalyst for a new generation of AI native games? Where does management believe the biggest opportunity or risk lies for Tencent Games over the coming 3 to 5 years? And a follow-up on the advertising point you just made. James, I believe you spoke about softening macro consumption and about the offsetting factor, which includes mini dramas advertising. How should we think about this — given that we're also starting from a higher base? Should we expect greater — I mean, normalization pressure in 2027? Or do we believe AI-driven monetization and additional market share gains will keep offsetting — will keep providing some of those offsetting factors?

James Mitchell

executive
#26

On the second one, I believe I've already answered that we're optimistic about our capacity to keep outgrowing the market; and I don't sense that we face any particular base pressure. My view is simply that the advertising market has already been weak for several months in 2026. So I wouldn't describe 2026 as a spectacularly good year for advertising; and ultimately these things are cyclical. At some point, consumer confidence and consumer spending will recover. And once that happens, advertising spending will rebound; so if anything, I probably see more upside than downside risk for 2027, but time will tell. As for your first question, AI can represent either a threat or an opportunity for the game business, for the game industry globally. We've observed that in certain categories of games, such as hyper-casual, single-player mobile games, there have been waves of AI-created copycat games appearing very quickly as a result of people running decompilers on existing games. Lately, the big model laboratories have taken action to stop that decompilation or piracy process, which appears to have curbed the AI copycatting, but we're very early in the evolution. I would say the work involved, the expertise needed not only to create but also to operate, sustain, and grow over time a multiplayer evergreen game that is multi-platform and multimodal is orders of magnitude greater than the work needed to create a single player hyper-casual mobile game; and obviously, those multiplayer gameplay-based, multi-platform, multimodal evergreen games are where we make most of our game revenue. So AI can speed up parts of that work, as I said when answering a question earlier, like the preproduction drafting of what a game could be, in effect. But as things stand today, AI cannot take over the main production process or the main game operation process.

Chi Ping Lau

executive
#27

I also don't believe the game industry lacks titles, correct? So being able to offer an ever-growing number of games doesn't truly shift the dynamics of gaming — what matters is how you distribute and draw in a large user base consistently, right? And so I think it will actually bring about a new generation of AI-native game experiences, and those experiences can stand alone, but they can equally be folded into evergreen game platforms, right? And I think we've seen many instances where, when an innovation appears, evergreen game platforms are able to incorporate those game experiences into the platforms and in a sense gain from that strong idea. And at the same time, AI game generation tools can genuinely help push game platforms to evolve toward more and more user-generated content, right? It becomes in a sense both a first-party and a third-party game platform, right? You'd have professionally generated content from the game team. But at the same time, you in a sense have a space where users can create content of their own, right? And I think we've already seen that working very well for Peacekeeper Elite, right? And we expect that as more and more of these tools arrive, right, it would genuinely help speed up the evolution of many of our evergreen games into evergreen game platforms. And that would be enormously positive for us.

Wendy Huang

executive
#28

We will take the next question from John Choi from Daiwa.

Hyungwook Choi

analyst
#29

I have a couple of questions. First, regarding returns on AI investment. With these new AI products launching, I believe we've seen operating profit drop fairly noticeably this quarter. So given that these new AI products are already putting a meaningful drag on operating profit, what milestones should we be watching for to gauge the returns on these investments? And when do you expect those earnings drags to begin narrowing? That's my first question. My second question concerns your cloud revenue. I think we've observed a meaningful acceleration to 35% this quarter. And I think management shared more detail on the model-as-a-service GPU rental. Given all the compute constraints we've seen both domestically and globally, how should we think about the quality and sustainability of this AI-related cloud demand? And what revenue and margin opportunity should we expect going forward?

Chi Ping Lau

executive
#30

On AI investment, right, I think first of all we sort of have to make the investment, right? And then the benchmark is really whether we're creating valuable businesses and products from that investment, right? Before looking at the financials, let's look at whether we're creating value from that investment. And before our Hy3 and Hy4 launched, right, it was uncertain whether our training investments were generating valuable models. But now with Hy4 and Hy4 flash coming out and we see the explosive endorsement from users by burning through a lot of tokens, right? I think we can put a check on it, right, and say the investment in training is now generating very valuable products. And as we look to the future, future generations of models will get more and more exciting. And then when we look at the products, right, first of all we actually sort of invested in Yuanbao and you could say Yuanbao wasn't as successful as we expected. Now Yuanbao is actually back to not just not focused on growing user base, but focused on growing its capability, and it's actually sort of growing its capability very nicely. And recently, it launched the AI search function, which I think is also an addition to the overall product offering, but we actually cut spending on Yuanbao quite significantly. But then the next product we actually launched was like WorkBuddy, and WorkBuddy has become a phenomenal success, right? So now we can actually put a check and say all the investment we put around AI products is generating positive value. And we're in a very good position to keep expanding on these valuable businesses and look for new opportunities, too, right? And over time, we know these products will generate, first of all, revenue and then — well, first of all, usage and then revenue and then over time, profit. The exact timing of that is unknown to anyone. We’re truly in the middle of a generational technology revolution, correct? And I think it’s very hard to create a business the way you construct a building and then seek rental return from it; but when we consider the investment they’ve made and we’re producing very valuable results, right, then we feel confident that over time, it will bring in revenue and it will eventually produce profit.

James Mitchell

executive
#31

Regarding cloud revenue growth, I believe what you were asking is whether restrictions on importing advanced semiconductors might interfere with that growth. From our perspective, that won’t happen for us, since the things propelling faster cloud revenue growth include WorkBuddy and Model-as-a-Service. For WorkBuddy and the models we deploy inside Model-as-a-Service, these are very efficient models being deployed, so they can be run on domestic chips, which the new limitations do not cover. Also, Hunyuan 4 flash costing 95% less than Hunyuan 4 illustrates this point. In recent months there have been transformative gains in how many tokens fixed compute or GPU capacity can produce; that lets us obtain considerably more mileage from the compute already in place, while simultaneously we are adding more compute via the domestic supply chain; together those two forces ought to support our cloud revenue growth ahead.

Wendy Huang

executive
#32

We will take the last question from Ron Keung from Goldman Sachs.

Ronald Keung

analyst
#33

My first question concerns profit growth in our core business. When I look at operating profit and exclude the new AI products, growth has decelerated. So what I'd like to hear is which cost items are swinging the most for our core business? Is it the mini dramas, or is it attributed depreciation? And how should we think about that underlying high-quality profit growth that used to define the core business? What's the right way to think about growth — the profit growth outlook over the coming quarters? Second, perhaps you could discuss Yuanbao Search, since we're on the topic of agentic capabilities and all of that, and I also notice we're pursuing some initiatives around Yuanbao Search. So what's the reasoning behind returning to the search vertical? And in what ways are we innovating within that vertical?

James Mitchell

executive
#34

Let me try to take the first question, about operating profit growth in the core business, or the non-new-AI-products business. There are two drivers to consider. One driver is the soft consumption environment. That directly shows up as slower growth in our financial services, fintech services business. It would also have shown up as slower growth in our marketing services business. However, as I noted, the mini drama series acted as a catalyst that let us partially counterbalance that, even though getting that ecosystem off the ground came with substantial expenses of its own. So we confronted that macroeconomic headwind, and it hurt our margins. The second element is that, as you likely know, a lot of component pricing spiked sharply toward the end of last year — DRAM, for instance, was roughly 400% higher year-on-year by early this year. During this quarter, you observed our established offerings—not only the fresh AI lineup but also the traditional ones—taking on a portion of the cost pass-through stemming from this year’s earlier surge in component input prices, memory in particular. At this point, we along with, I believe, numerous other firms have acted to monitor and control our use of those parts, especially memory, far more carefully. And I suspect that partly because of this, memory pricing has lately been leveling off. We anticipate that for the coming several quarters, the core business will contend with those two cost pressures I referenced, after which it should work past them—first as we absorb this stretch of unusually elevated component costs, and later as the macro environment eventually settles down. Those two factors, then, explain why core operating profit growth is slower.

Chi Ping Lau

executive
#35

When it comes to Yuanbao Search, this really reflects an acknowledgment that search is genuinely a very important capability in the AI era. But it has also been completely transformed by AI agentic capability. So as we continue enhancing Yuanbao's capabilities, the search function inside Yuanbao grows stronger and stronger, to the point where we truly saw it as an atomic capability that can actually be offered to other Internet and technology companies in order to generate revenue. That is why we have essentially opened it up and begun offering it as a service and providing it to other companies. So this reflects, first, that AI has been transforming search capability; second, that Yuanbao, in the course of delivering its service, has built up a very significant capability in that area; and third, that there is actually external demand. So we have launched it and over time would see some revenue from it.

Wendy Huang

executive
#36

Thanks, Martin and James. Appreciate everyone being here for our results webinar today. That brings the call to a close. If you'd like to review our press release along with other financial information, head to the IR section on our company site at www.tencent.com. A replay of the webinar will be up shortly as well. Thanks again, and see you next quarter.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Tencent Holdings Limited transcript — plus 132,000+ transcripts from 7,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.

Get the API View API docs →

This call discussed

For developers and AI pipelines

Programmatic access to Tencent Holdings Limited earnings transcripts and 132,000+ others is available through the EarningsCalls.dev REST API. Plans from $39.99/month — full transcripts, speaker segments, full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.

More tools for stock research

OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in Tencent Holdings Limited or any ticker on your watchlist.

StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.