Meitu (Trans): Monetization Pivot to AI Compute Credits

Dolphin Research's Trans of Meitu FY26 interim earnings call

I. Key takeaways

1. Shareholder returns: interim DPS up ~50%; buyback ~75% executed

a. The BOD approved an interim DPS of HK$0.067 vs. HK$0.045 in 1H25. Management reiterated its practice of paying a fixed ratio of Adj. NP.

b. Of the HK$300 mn buyback announced in Mar, ~HK$230 mn has been repurchased to date. The remainder will be deployed in light of market conditions and biz. development.

2. On not issuing a 'positive profit alert': no direct link to performance

a. Under the Securities and Futures Ordinance and Listing Rules, a profit warning/alert is triggered by a material deviation vs. market consensus that constitutes inside information, not by good or bad results per se.

b. With added quarterly operating disclosures this year, management sees a lower likelihood of material deviations. As a result, pre-announcements will naturally decrease and should not be read as a results signal.

3. Three YoY base changes affecting reads

a. The cosmetics supply-chain service in Beauty Solutions is classified as discontinued in the 2025 AR. YoY comparisons here are on a continuing-ops basis.

b. Productivity apps growth of 40.1% excludes the stock-photo licensing biz. exited in 2025, which contributed ~RMB27 mn revenue in 1H25.

c. Adj. NP rose 39.5% YoY including discontinued ops for historical comparability. On another basis, it was ~RMB620 mn, up 40% YoY.

4. Two drivers of GPM compression and the cost-structure hedge

a. GPM was 71.5% vs. 75.3% a year ago. The decline reflects faster growth in AI-native productivity apps within Imaging & Design, where API costs are a higher share of revenue, and a lower mix of higher-margin ads. Mature apps such as Meitu Xiuxiu and BeautyCam saw stable margins.

b. Ads revenue was RMB410 mn, down 4.4% YoY, due to declines in brand/traditional formats while programmatic continued to grow. Management remains cautiously optimistic on ads.

c. Third-party API costs remained in the mid-single-digit percentage of Imaging & Design revenue, with over 96% of generation in 1H from in-house models.

d. Channel fees of RMB380 mn were the largest cost item, with a stable ratio to Imaging & Design revenue. Compute and cloud-related costs were RMB130 mn, over half tied to inference compute directly linked to user consumption.

e. R&D expense was RMB470 mn, up 4.8% YoY. Excluding the high base from last year's general foundation model training, growth would be ~11%.

II. Detailed call notes

2.1 Management remarks

1. Content strategy: extend tools into a 'create–distribute–accumulate' flywheel

a. On creation, Meitu Xiuxiu and BeautyCam anchor mass entry and cash flow. Wink, Kaipai, MVLAND, and RoboNeo extend creation into personal avatars, virtual avatars, MVs, voice-over, and AI short dramas.

b. On distribution, Kaipai extends from creation into promotion. Meitu Xiuxiu boosts exposure via Recipes, stickers, filters, and brand/IP co-creation, while ZCOOL enables discovery and transactions within the designer community.

c. On accumulation, personal/virtual avatars, music rights, virtual singers, and brands/products built in Meitu Design Studio are stored as reusable assets, feeding the next round of creation and distribution.

2. Intl market selection: focus on creator mix, not population size

a. Internally, creator density is split into 'casual' and 'professional' metrics. Casual creators post daily content without monetization; professionals monetize content.

b. Findings in Brazil ran counter to intuition: there are many content posters, but few who can monetize as pros. The gap between a large casual base and low pro density is the product opportunity.

c. Kaipai's growth in China follows the same logic: users in health, insurance, and real estate understand their biz., but not topics, scripts, or video production.

3. Meitu Design Studio: shifting from shelf e-com assets to social content commerce

a. The consumer decision path has flipped. It used to be seeding on social, then deciding on detail pages on Amazon and other shelf platforms; now the loop closes directly within social livestreams and KOL content feeds.

b. Cross-border sellers have inverted their resource allocation. Previously ~80% on shelf e-com assets and 20% on social; now 80%–90% on social assets and 20% on shelf.

c. Two-step product path: Step 1 (past two years) built capabilities for e-com platform assets and accumulated product, brand, and AI model assets. Step 2 extends into social content commerce, focusing on product and service content.

4. Monetization mix: subscription tiers cap spend; AI compute points take over

a. ARPPU for productivity apps is ~150% of that in life-use scenarios, and management expects the gap to widen. The current gap reflects subscription tier caps—users willing to pay more have no avenue to do so.

b. As Agent capabilities improved in 1H across products, users handled more complex tasks at higher frequency, driving a sharp rise in AI compute-point consumption with actual workload.

c. High-value users have emerged. One product has several thousand premium-service users with annualized spend ~10x overall ARPPU, with user count up 6x vs. Dec 2025. Another product, within months of launch, has a monthly ARPPU of ~RMB220, comparable to pro-creation app pricing.

5. Resource allocation for new products: small teams to test, scale on signals

a. Given the rapid change in AI, validation must be faster. Early opportunities are pushed to market by small teams with limited initial spend to validate demand and business models.

b. Only upon clear user growth and monetization signals does the company add R&D and marketing resources, dynamically adjusting investment to maintain iteration speed while controlling early exploration costs.

2.2 Q&A

Q: How were needs for new products like Kaipai, Picchi, and MVLAND identified, and why did ARPPU ramp so fast?

A: All needs were mined from user segmentation in existing products, and high ARPPU comes from choosing inherently high-ticket verticals. Demand for MVLAND was validated 12 years ago by Meipai—launched in May 2014, DAU hit 5–6 mn within two months, based on MV effects from user-shot videos.

MVLAND is the AI-native version, initially targeting pro musicians to meet social distribution needs for songs, then moving top-down to the mass market once pro aesthetics are met.

Picchi addresses new portrait-processing needs from younger users and KOLs in Meitu Xiuxiu and other apps—remembering retouching preferences, batch processing, and stronger 'internet vibe'. Kaipai originated from the teleprompter feature in BeautyCam, revealing demand for talking-head videos.

ARPPU is largely vertical-driven: pre-AI, making one MV could cost RMB50k to low hundreds of thousands. Products in such verticals naturally command far higher ARPPU than life-use apps.

Q: What KPIs determine whether new products are kept or terminated, and where are the future product boundaries?

A: No quantified kill-threshold disclosed; products are reviewed via user and ARR data for integration or adjustment. The aim is to focus limited resources on validated products.

Boundaries are not limitless. AI content-creation tools naturally extend into content distribution, which then accumulates into IP and rights assets, forming a content flywheel. The company will validate only around this flywheel, not unrelated tracks.

Q: Does moving from creation to distribution to accumulation mean shifting from asset-light tools to heavy ops? What are the business model and financial implications?

A: Distribution is an extension of existing post-creation needs, not a new biz.; the team scales from the existing distribution team without building anew. Distribution is a common capability across gaming, film/TV, and music, and has always been core. The company is extending from distributing its own products to distributing creator content.

Previously, users of Kaipai or MVLAND had to open accounts themselves or use third-party tools for traffic buying, product selection, commerce, and livestreaming. Now, by filling this service gap, distribution and asset accumulation feed back into creation, forming a new loop and boosting stickiness.

Q: What are the company's strengths and gaps in content distribution and asset accumulation?

A: Strength: proprietary channel matrix. Gap: a mature external-channel distribution service platform is still under construction. The matrix of Meitu Xiuxiu, BeautyCam, Meitu Design Studio, and ZCOOL directly supports distribution. Many IPs have co-launched with Meitu products with strong data, making Meitu Xiuxiu a validation channel for IP content.

The gap lies in broad external channels across social, online video, music, and e-com, and how to empower users via traffic buying. Management notes this is not a pure disadvantage, but an area where a mature distribution service platform has yet to be built and is a key goal.

Q: Overseas MAU growth slowed. What changes has management observed in overseas markets?

A: No direct explanation for the MAU slowdown; instead, management shared a 'creator distribution' mechanism from a U.S. marketing-video product. This product is akin to a marketing-video Midjourney. Studios and creative agencies use its templates for social marketing videos, with ARR up meaningfully in two months.

The mechanism is instructive. Traditionally, budgets are given to agencies or MCNs to find hundreds of KOLs, with little transparency. The new approach lets existing creators, who would post anyway, tag the product source, submit links if content goes viral, and receive incentives based on likes, shares, and comments. This effectively spends the entire budget on already-validated viral content, bypassing middlemen and leveraging long-tail creators.

The prerequisite is that the product's output is inherently social-ready. Management believes MVLAND, Kaipai, and RoboNeo fit this.

Q: Several fast-growing products are video-focused. Will the company prioritize video productivity apps?

A: Yes. Likely 80%+ of products will use video as the primary format going forward. Fast growers like Wink and Kaipai are video-centric, aligning with the surge in creator output and content consumption, and with Meitu's 280 mn MAU base of video creators and consumers.

All incubating products are also video-first. Video is far more demanding than images in data training and CV algorithms. The company positions itself as productivity tools centered on social creators.

Q: Two products in design and video completed brand refreshes this year. What is the logic and how are metrics trending?

A: Kaipai's overseas version pivoted to marketing video with ARR over US$1 mn; another product remains in pre-PMF validation. The latter's ARR trails MVLAND and Picchi, with internal org changes pending, and targets e-com sellers' social commerce needs, mainly marketing videos.

The pivot for Kaipai overseas reflects the U.S. market's high density of professional creators. Trained on pro video tools, they view Kaipai for the masses as 'de-featured' and have low download intent. Hence the shift from live talking-heads to keyword or asset-driven generation of marketing videos.

The company believes many social commerce categories still require real on-camera presence to build trust, making live shooting hard to replace. This direction will see more investment in LatAm and other regions with many casual creators and low professionalization.

Q: How do MVLAND's user profiles differ at home and abroad, and which features drive ARR growth?

A: Domestic vs. overseas is ~2:8, with paid users mainly established studios and long-tail indie musicians. China is ~20%, with the rest overseas.

Studios used to produce ~500 songs a month, first pairing simple lyric videos for YouTube/Spotify, and only making MVs for the ~10 out of 100 that took off. Now it is reversed—spending a few hundred dollars to quickly make an MV per song, greatly expanding the validation pool.

Indie musicians first create a 30s chorus demo to launch a challenge on short-video platforms, then complete the full track within two to three days if it pops. MVLAND helps render the chorus into artistic visuals to improve validation efficiency.

The next growth cohort is brands and merchants using marketing MVs for new launches or promotions.

Q: With overall MAU stabilizing, what is the playbook to grow MAU and better monetize the mature user base?

A: Plan for steady, modest MAU growth, with focus on overseas MAU and especially global paid users. Monetizing the mature base returns to fundamentals—identifying user needs and building solid features, and users will pay.

There will also be more promotion of paid features, vs. the prior emphasis on free-user acquisition.

Q: How should the current valuation be viewed, and what anchors should the market use?

A: There is no perfect comp; start from first principles rather than forcing a comp. Management suggests assessing the future shape of the track, the company's profit and cash flow growth, then seeking comps accordingly, which is more reasonable than forcing a comparison first.

Q: Globally, who are the most competitive peers?

A: Mainly startups. Imaging apps have evolved from 1.0 PC to 2.0 mobile, to 3.0 (e.g., Wink, Kaipai), and now 4.0 AI-native apps, where most content can be generated without live shooting.

In 4.0, the most competitive players are largely startups without legacy product or go-to-market constraints, redefining products and marketing with newer approaches. The company is reorganizing internally to match 4.0.

Q: AI compute-point consumption more than doubled in 1H. Is this a new biz. model, and any 2H guidance?

A: Confirmed as a different logic from subscriptions and ads, but no separate guidance. Subscription tiers have natural caps; users cannot pay more even if they want to. AI compute points rise as long as products keep creating value in production and users keep using them.

Management expects compute points to be embedded in more new products. Based on current revenue, it remains on a very rapid growth path, with no major change in sight near term.

Q: Market feedback suggests the company is cautious in external user acquisition for new products, risking a missed lead in the sprint phase. Will 2H strategy change?

A: Yes. Shift from broad free-user acquisition to concentrating spend on paid features and entitlements. Most marketing will remain KOL-driven on social, but the focus moves from mass reach to paid-user growth.

The reason: ongoing AI compute consumption keeps costs rising, and AI apps are becoming consumer products. The prior logic of acquiring free users first, then upselling subscriptions on that base, will change to some extent.

Separately, the company is building a social content marketing mechanism akin to performance ads—encouraging more creators to post across major platforms and rewarding based on saves, comments, and shares. While less 'hard' than ad buying, it aligns better with creator-led dissemination and should be more precise in theory.

Q: Wink ranked high in multiple countries in 1H. What is the 2H product strategy?

A: Acknowledge insufficient localization; continue to focus on quality restoration and shift to region-based, granular R&D. In China, Wink grew on video beautification and quality restoration, but video beautification is led by Chinese aesthetics, extending to CN/JP/KR and Asia with regional and cultural limits.

Overseas growth is mainly driven by quality restoration, a global common need. Multiple rounds of benchmarking suggest the company is best-in-class here, and growth came with minimal localization, clearly outperforming video beautification.

Deeper work revealed wide regional differences in desired texture, tone, and grading after restoration, as well as facial skin rendering in video. Hence the world is segmented into major regions, with tailored R&D to drive globalization.

Q: GPM fell while GP rose. How to view margin trajectory?

A: No quantified range; management aims to keep margins relatively stable. This pattern is common in the industry, and one should not infer a trend from a single period.

Three levers: (1) shift focus from MAU/free-user stacking to high-ARPPU users to lift high-margin mix; (2) third-party API pricing should fall as competition intensifies; (3) as in-house vertical models reach sufficient quality, they can replace pricier third-party APIs.

<End of text>

Risk disclosure and disclaimer:Dolphin Research Disclaimer and General Disclosure