In October 2008, Wu Xinhong launched a 'foolproof PS' tool online, which surpassed one million users within two months. This software, later known as Meitu XiuXiu, capitalized on the golden age of mobile internet—an era dominated by Tencent and Meituan, where 'user scale was everything.' On its IPO day in December 2016, Cai Wensheng declared: 'Everyone regrets missing Tencent. This story may well repeat with Meitu.'

Yet Meitu failed to deliver on market expectations. Its smartphone venture, positioned against Xiaomi, incurred a RMB 500 million loss in 2018 and shut down the following year. Its e-commerce platform, Meitu Beauty, ceased operations after just a year. Its social media app, Meipai, clashed with Douyin and Kuaishou, with monthly active users (MAUs) plummeting from 110 million to 5.62 million. Even its foray into cryptocurrency—investing over USD 100 million in Bitcoin and Ethereum in 2021—became one of its few operational highlights.

From 2017 to 2021, Meitu accumulated nearly RMB 2 billion in losses, with its market cap shrinking from nearly RMB 100 billion to HKD 4 billion by July 2022, a 95% decline over four years. A company that chased every trend ended up battered by each one.

The turning point came in 2023. Wu Xinhong replaced Cai Wensheng as chairman, betting everything on imaging—now prefixed with 'AI.' The management's tone shifted: no more talk of becoming 'the next Tencent,' but instead embracing a 'small and beautiful' approach.

Focusing on the interim report, in the first half of 2026, Meitu’s total revenue reached RMB 2.213 billion, up 22.1% year-on-year, slightly exceeding market expectations. Gross profit was RMB 1.581 billion, up 15.9% year-on-year, with a gross margin of 71.5%, down 2.2 percentage points year-on-year. Net profit attributable to the parent was RMB 619 million, up 40.0% year-on-year. Adjusted net profit attributable to the parent was RMB 652 million, up 39.5% year-on-year. Profitability exceeded market expectations. UBS estimated a +33% year-on-year increase, while the company’s positive profit guidance indicated a range of 36%–40%, with the actual +39.5% falling at the upper end of the range.

It seems Meitu, after chasing trends for 17 years, has finally found its moment.

(The financial report summary is below, in RMB 100 millions. All data sourced from Wind and corporate reports, compiled by Financial Records.)

On paper, Meitu is a steadily profitable software company. Narratively, it aims to squeeze into (fit into) the valuation framework of AI application companies. In H1 2026, Meitu disclosed its AI productivity application ARR of approximately RMB 620 million, up 47.8% year-on-year—fast growth, but still modest in scale. The market, however, remains unimpressed, with Meitu’s market cap sliding over 40%.

After reviewing the financials, we believe that while Meitu attributes its performance growth to AI advancements, it still relies on rigid growth in MAUs and paying users. In terms of models and business models, Meitu has yet to offer a compelling narrative distinct from its competitors.

Thus, even as Meitu shifts its pricing model from traditional SaaS to token-based billing, its essence remains that of an internet company.

01

Growth Engine Shifting, Customer Base Remains Solid

During the reporting period, Meitu’s total revenue in H1 2026 reached RMB 2.213 billion, up 22.1% year-on-year. The previous year’s growth was just 12.3%, marking a nearly 10-percentage-point acceleration in one year.

Breaking down the revenue structure, the old engine has stalled. Advertising revenue fell 4.4% year-on-year to RMB 415 million. All growth came from imaging and design products: RMB 1.769 billion, up 30.9% year-on-year, accounting for nearly 80% of total revenue.

The real shift occurred within imaging and design products. Productivity application revenue surged 40.1% year-on-year, now accounting for over 18% of the total.

In H1 2026, Meitu disclosed its AI productivity application ARR of approximately RMB 620 million, up 47.8% year-on-year. Productivity application MAUs reached 33 million, an all-time high. The business model has evolved beyond traditional SaaS. Beyond subscriptions, it introduced 'AI computing power points' billed by token. User spending on computing power points rose over 46% quarter-on-quarter in Q1 and Q2, totaling approximately 113% for the half-year.

Profitability begins with gross margin. H1 gross margin was 71.4%, down 2.2 percentage points year-on-year. The culprit is clear: AI inference costs are token-based, and the share of high-compute productivity applications is rising.

Net profit attributable to the parent was RMB 619 million, up 40.0% year-on-year (on a continuing operations basis), nearly double the revenue growth rate.

Leverage came from across-the-board cost control. Total operating expenses in H1 were RMB 1.024 billion, up just 6.3% year-on-year—less than one-third of revenue growth. The combined expense ratio fell from 52.9% to 46.3%. With a MAU base of 282 million generating organic growth, customer acquisition and fulfillment no longer rely on heavy spending.

Today, AI is transforming not just corporate workflows but also the vast ecosystem of individual creators (OPCs) and small-to-medium businesses (SMBs). E-commerce sellers need batch-generated, cross-platform visual assets. Podcasters, educators, insurers, and real estate agents demand low-cost marketing videos. Cross-border merchants seek localized design capabilities. In these scenarios, content directly drives revenue.

Willingness to pay for production tools far exceeds that for entertainment—the soil for Meitu’s current growth.

Kaipai’s MAUs, paying subscribers, and ARR all surged over 100% year-on-year. Vmake’s overseas MAUs grew over 54% year-on-year, with ARR of approximately USD 5 million. MVLAND and Picchi both exceeded USD 500,000 in ARR within months of launch. Meitu’s design studios house sellers’ asset libraries, brand templates, and collaboration workflows—switching tools would cost far more than switching models. The moat lies in demand-side user habits and scenario lock-in.

Another axis is globalization. By 2025, Meitu’s overseas MAUs exceeded 100 million, outpacing mainland growth for the first time. In H1 2026, overseas MAUs for productivity applications grew over 80% year-on-year, with paying users overseas growing roughly twice as fast as those in the mainland. With domestic traffic saturation an industry consensus, Meitu’s global expansion offers attractive capital returns.

03

No Model Investment, Valuation Capped—But No Mistake in Choice

Yet the brighter the interim results, the starker the capital market’s indifference. In May 2025, Alibaba subscribed to RMB 250 million in Meitu convertible bonds, driving the stock from below HKD 3 to HKD 12 as the market priced it as an 'AI application leader.' A year later, the stock fell back below HKD 4, with a market cap of approximately HKD 19.6 billion—a HKD 35 billion evaporate (evaporation).

The market’s current AI pricing logic is blunt: premium for compute narratives.

Consider the numbers. ByteDance’s AI capex plan for 2026 has been raised to over RMB 200 billion. Alibaba’s three-year, RMB 380 billion investment plan had executed approximately RMB 190 billion by end-June 2026. Tencent’s Q1 2026 capex alone hit RMB 31.94 billion.

How much did Meitu spend in H1? Approximately RMB 30 million. ByteDance’s daily compute purchases exceed Meitu’s entire H1 capex. Meitu’s HKD 19.6 billion market cap cannot buy even a quarter’s worth of GPUs from ByteDance.

Big Tech players doubling down on AI seek 'big outcomes.' Meitu lacks the scale to compete.

More nuanced is Meitu’s own choice. H1 2026 R&D spending was RMB 473 million, up just 4.8% year-on-year, allocated to vertical models and applications. The 48.9% increase in interim dividends confirms its cash cow positioning.

Facing the industry reality that 'large models devour software,' management abandoned in-house foundation model development to narrow its strategic focus. Our assessment: strategically sober, but valuation-capped.

For application companies like Meitu, the valuation discount from abandoning models is inevitable. Peer experience shows that joining the model arms race is suicidal—Meitu’s 2024 damage control was correct.

Among peers: B612 Kakuji’s parent, Snow, remained unprofitable in 2024, monetizing mostly via ads. Wondershare bet on in-house AI large models (Tianmu), seeing H1 revenue fall 7.09% year-on-year and net profit plunge RMB 68.98 million—a 30% widening loss.

Three companies in the same track ( track : track/field), two diverging paths: in-house models failed to serve as a moat; victory still hinges on actual product capabilities.

04

Commercially, Meitu Still Resembles an Internet Company

If Meitu insists on positioning itself as an 'AI application company,' it is somewhat wishful.

Superficially, 15% productivity revenue cannot justify AI company valuation multiples. The choice not to develop large models further alienates capital markets.

Internally, UBS, referencing Anthropic’s prospectus, offered a simpler judgment on AI-era business models: constrained by compute, AI currently leans toward extracting more revenue from fewer users. If income still scales with user base expansion, the model remains grid-based—an internet business model.

For Meitu, revenue per paying user (revenue ÷ monthly paying users) has trended downward over the past three years, aligning more with an internet model of declining marginal costs and user acquisition-driven growth.

Thus, AI for Meitu is not an identity but a production asset. New monetization—compute points billed by token—adds a revenue stream beyond subscriptions. New customer acquisition narratives—the 'AI application' label helped secure USD 250 million in Alibaba convertible bonds.

Yet QuestMobile data shows Doubao’s MAUs hit 382 million in June 2026, dominating China’s AI app market. ByteDance’s Jimeng AI reached 13.525 million MAUs in Q1 2026. Features like 'remove passersby' or 'AI image expansion' are free on Doubao. Meitu’s paid feature list is being commoditized by Big Tech.

If free models dismantle paywalls, paying users should churn. To be sure, Meitu is not without advantages. Technology can democratize, but aesthetics cannot. Aesthetic sensibilities are embedded in templates, filter parameters, and interaction defaults—as well as in e-commerce sellers’ asset libraries, brand templates, and collaboration workflows. Model leaderboards cannot measure this, nor can open-source communities replicate it. RoboNeo’s 20-person team surpassed 1 million MAUs in its first month by leveraging such productization capabilities.