Established for only a little over two years and still not profitable, Kling AI has already been tagged by capital with a valuation in the hundreds of billions of yuan.

As an AI video generation product under Kuaishou Technology (ASX: 01024), Kling AI was officially launched in June 2024 and has pursued commercialization through user subscriptions and enterprise API calls, with application scenarios covering advertising and marketing, film and short dramas, gaming, and other fields.

Recently, foreign media reported that Kling AI plans to initiate a Hong Kong listing process within the next 12 months and may file a listing application with the Hong Kong Stock Exchange as early as the beginning of 2027, aiming to raise at least US$1 billion. In response to the news about a Hong Kong listing, Kling AI told Phoenix WEEKLY Finance that "there is currently no relevant information."

In fact, even before the listing news emerged, Kling AI had already secured a high valuation. According to a financing announcement disclosed by Kuaishou in July this year, Kling AI planned to carry out an independent financing of up to about US$3 billion, with a pre-financing valuation of about US$15 billion, corresponding to a maximum post-investment valuation of US$18 billion, equivalent to more than 120 billion yuan.

What stands in contrast to the hundred-billion-yuan valuation is that Kling AI is still operating at a loss. Although its revenue in the first half of 2026 has already exceeded 1.5 billion yuan, continued R&D investment and computing power costs still make profitability a challenge.

Losses persist despite rapid revenue growth

In the global AI video generation sector, Kling AI has already entered the first tier. Zhang Yi, CEO of iiMedia Research, said in an interview with Phoenix WEEKLY Finance that Kling AI currently forms a duopoly competition pattern in China with ByteDance's Seedance and is also in the first tier of the global AI video generation industry, with certain advantages especially in professional content creation scenarios.

A research report released by AVIC Securities in July this year also pointed out that since its release in June 2024, Kling has iterated to the 3.0 series and was among the first globally to achieve native 4K direct output and precise motion control, with technical strength in the industry's first tier.

Continuous iteration of model capabilities has driven growth in Kling AI's commercialization revenue. Kuaishou's financial reports for 2025 and the first half of 2026 show that in 2025, Kling AI generated revenue of about 1.1 billion yuan. In the first quarter of 2026, its revenue exceeded 650 million yuan, up more than 300% year on year; second-quarter revenue further increased to more than 850 million yuan, up more than 200% year on year. In the first half of this year, Kling AI's revenue has already exceeded 1.5 billion yuan, surpassing its full-year level in 2025.

Behind the revenue growth, Kling AI's customer base is also expanding. As of June this year, its global user base had exceeded 100 million, with nearly 50,000 enterprise customers.

Rapid growth in revenue and users does not mean profitability has improved in tandem, and Kling AI has not yet shaken off losses. According to an announcement disclosed by Kuaishou in July this year, in 2024 and 2025, Kling AI's unaudited net losses were approximately 500 million yuan and 1.9 billion yuan, respectively. Chart: In July 2026, Kuaishou issued an announcement showing that in 2024 and 2025, Kling AI's unaudited net losses were approximately 500 million yuan and 1.9 billion yuan, respectively.

Entering 2026, losses may still be continuing. Some market sources said that for the full year of 2025 and the first quarter of 2026, Kling AI's cumulative losses exceeded 2 billion yuan. According to other media reports, Kling AI is expected to achieve single-quarter break-even only by the fourth quarter of 2028.

In reality, profitability difficulties are common across the AI video generation sector. There is an essential difference between AI video generation services and traditional SaaS. With traditional SaaS products, users' use does not bring significant new costs. Tokens sold by AI video companies such as Kling AI are prepaid credits; when users top up, ARR increases, but costs occur when users actually consume tokens and call models to generate videos.

"The problem is not only high computing power costs." Regarding the profitability challenge for AI video generation companies, Li Wangminghui, associate professor at the School of Electronic and Information Engineering and the National Maglev Transportation Engineering R&D Center of Tongji University, told Phoenix WEEKLY Finance that every call to a generative video service comes with actual inference costs. Videos generated by models may not meet requirements on the first try, and users may need to revise and regenerate repeatedly; these failures and reworks are actually costs.

Specifically, users usually need multiple generation and debugging attempts to obtain a video that meets requirements. The more tokens consumed, the higher the corresponding computing power expenditure. Even if a single inference can achieve positive gross margin, large fixed costs such as early-stage model training, hardware depreciation, and R&D expenses remain difficult to amortize. Combined with industry price competition and users' continuously rising demand for video quality, the benefits of technological cost reduction are easily consumed.

Zhang Yi believes that the benefits of falling computing power costs may be quickly consumed by industry price competition. What Kling AI needs more is to translate its technological leadership into pricing power and maintain premium capability through continuous technological iteration.

High valuation hides pressure to deliver performance

Without yet achieving profitability, Kling AI has already obtained a post-investment valuation of up to US$18 billion. How exactly this price should be measured has become a focus of market discussion.

At present, the primary market generally uses ARR multiples to value AI companies such as Kling. According to Kuaishou's disclosure, as of March this year, Kling AI's ARR was close to US$500 million. The US$18 billion valuation corresponds to about 36 times ARR.

According to Zhang Yi, valuation logic differs among different types of AI companies: foundational large-model companies are priced based on growth expectations from general-purpose platforms, AI office products are valued more like traditional SaaS, and Kling AI's valuation lies between traditional SaaS and AI platform premiums.

Is the US$18 billion valuation reasonable? "The core support for Kling AI's valuation mainly comes from rapid ARR growth, a relatively high proportion of overseas revenue, and participation by industrial capital such as Tencent, Alibaba, and Baidu," Zhang Yi said. In his view, the 36-times ARR valuation implies expectations that ARR will further grow to US$2 billion over the next two to three years and that gross margin performance will improve. Kling is still losing money, and failure to deliver on any link in the future could trigger a valuation re-rating.

"From the perspective of a technology researcher, I remain cautious about the valuation of such companies," Li Wangminghui said. The US$18 billion valuation not only reflects Kling AI's current commercialization progress, but also contains relatively high expectations for future market space and technological competitive position. She pointed out that Kling AI already has a certain scale of global users and continuously growing revenue, and combined with Kuaishou's accumulation in the video content field, these constitute support for its valuation. But current leadership in model capability does not mean it will still be leading three years from now. Generative AI technology spreads quickly, and new model architectures, training methods, and inference optimization could all change the industry's competitive landscape.

Intensifying competition also brings equity buyback constraints

In fact, Kling AI is no longer the only product in the AI video generation sector attracting attention. At present, products such as ByteDance's Seedance, Google's Veo, and Runway continue to iterate, and competition in the AI video generation industry has extended from model performance to price, cost, and commercial applications.

Regarding the future competitive landscape, Li Wangminghui believes that underlying video model research and development may concentrate in a small number of large enterprises with data, computing power, and research talent, but as technology spreads, open-source models advance, and inference costs decline, some standardized video generation capabilities at the application layer may gradually become homogenized, with more intense price competition.

This may mean that for Kling AI, model performance and user scale alone will make it difficult to maintain a competitive advantage over the long term. "Kling AI currently has a relatively large base of users and enterprise customers, but the number of users does not automatically equal user loyalty," Li Wangminghui said. What can truly form a long-term barrier is whether the model can deeply enter the actual production processes of industries such as film and advertising and solve issues such as shot consistency, brand standards, copyright, and content traceability.

Beyond technological competition, Kling AI also faces time constraints from the capital market. According to an announcement disclosed by Kuaishou in July this year, if Beijing Kling fails to complete an IPO before the agreed latest listing date, or before October 30, 2031, whichever is earlier, investors have the right to require Beijing Kling to repurchase all or part of their equity. The repurchase amount includes the original investment price plus a return calculated at a simple annual interest rate of 8%, minus dividends and distributions already received. Chart: In July 2026, Kuaishou issued an announcement showing that Kling AI's listing arrangements involve a repurchase clause dated October 30, 2031.

"There are still several years of buffer before 2031, and the repurchase clause is more like a bargaining chip in negotiations rather than imminent pressure. By comparison, changes in the listing window and market environment for Hong Kong AI companies may have a more direct impact on the pace of Kling AI's listing," Zhang Yi said.