From Dependency to Independence: Valuation Soars to Tens of Billions

Written by / Meng Huiyuan

Edited by / Li Jinlin

Layout by / Annalee

In June 2026, a financing announcement sent shockwaves through the venture capital community: Kuaishou's AI video generation subsidiary, Kling AI, officially launched its first funding round post-spin-off, with a pre-money valuation of $18 billion (approximately RMB 122 billion).

How extraordinary is this figure? As of early June, Kuaishou's parent company held a market capitalization of approximately $27 billion. Kling's valuation as a subsidiary already equates to roughly 67% of the entire Kuaishou empire.

A business established just two years ago, now valued at nearly 70% of its parent company's market cap, is nothing short of a bombshell in the industry and beyond.

Crucially, Kling is not an isolated case.

ByteDance's Doubao AI is quietly laying the groundwork through an "Independent Equity Incentive Plan," while Alibaba has merged its Tongyi Large Model Division and Future Life Lab to form the Token Foundry Division. Baidu's Kunlunxin has submitted an IPO application to the Hong Kong Stock Exchange... Clearly, a wave of AI asset spin-offs is surging from the "tech giant greenhouses" toward "independent capitalization."

But why are tech giants rushing to spin off their AI businesses? Does the video generation sector offer room for independent listings to thrive?

Costs, Valuations, and Strategy: The Three Critical Calculations Behind Tech Giants' AI Spin-offs

Undoubtedly, the rate at which AI businesses burn through cash is challenging traditional internet companies' financial statements.

Take Kuaishou as an example. Its Q1 2026 financial report revealed total revenue of RMB 33.7 billion, up 3.4% year-on-year; adjusted net profit reached RMB 3.4 billion, a sharp 26.3% decline. Meanwhile, Kuaishou's full-year 2026 capital expenditures are projected to hit RMB 26 billion, up approximately RMB 11 billion from 2025, with nearly all increases allocated to AI computing power—Kling being the top priority. Q1 R&D spending reached RMB 3.6 billion, up 9.8% year-on-year, with the financial report explicitly attributing this to increased AI investments.

The problem lies in Kling's commercial returns lagging far behind cost growth.

In Q1 2026, Kling AI generated over RMB 650 million in revenue, with an ARR (Annualized Revenue Run Rate) approaching $500 million (approximately RMB 3.6 billion). However, its costs are on a ten-billion-yuan scale.

Huang Lichong, President of Universal Capital International, stated in media interviews that spinning off Kling essentially transforms the AI business from a "cost center sustained by the parent company" into an "independent, market-driven AI tech firm with self-sustaining financing," using external capital to shoulder the hefty investments in video large models.

ByteDance faces similar pressure from soaring computing bills.

Its Doubao large model processes over 120 trillion tokens daily, with B-side market share for Volcano Engine's MaaS reaching 49.5%. Yet Doubao only launched its first tiered subscription plans (RMB 68/month, RMB 200/month, RMB 500/month) in May 2026, marking the nascent stages of commercialization.

Source: IDC's "Latest Analysis Report on China's Enterprise-Level MaaS Market Landscape"

Industry insiders describe this predicament: "The hotter Doubao gets, the more it's invoked, the greater the computing consumption, and the deeper the losses. Free access is a traffic strategy, but free is also killing cash flow."

Beyond cost pressures, the allure of valuation unlocking looms larger.

Morgan Stanley had previously valued Kling's segment at just $6 billion. However, as the spin-off plan advances and market valuation logic for pure AI firms evolves, the same assets now command a pre-IPO valuation of $18 billion—a threefold difference.

The core reason such AI businesses command high valuations post-independence lies in their treatment within consolidated financial statements as profit-eating cost items. Once independent, they're revalued based on sector scarcity, revenue growth, and market imagination, with price-to-sales ratios of several dozen times becoming commonplace.

Of course, deeper strategic considerations underpin these tech giants' AI asset spin-offs.

One aspect is opening capital exit pathways. ByteDance, valued at approximately $550 billion overall, faces prolonged obstacles to a comprehensive listing, constraining exit options for shareholders outside public markets. Against this backdrop, ByteDance launched the "Doubao Independent Equity Incentive Plan," granting stock options tied to Doubao's performance to Seed Division employees, establishing valuation benchmarks and organizational foundations. While ByteDance currently states it "won't consider spinning off" AI4S businesses, the independent equity plan's implementation lays the groundwork for Doubao's future independent capitalization.

The other aspect is achieving independent commercial model breakthroughs. On June 8, Alibaba established the Token Foundry Division, led personally by CEO Wu Yongming, integrating star products like Happy Horse and Happy Oyster with the Tongyi Large Model to accelerate AI commercialization. Separating AI businesses from the parent company helps escape internal resource allocation and pricing constraints, enabling rapid growth under independent commercial models.

Can Computing Cost Growth Be Outpaced by Technological Iteration?

If spin-offs represent rational choices after tech giants crunch the numbers, then for AI products like Kling, the prerequisite for securing independent financing and even an IPO hinges on one question: Can their unit economic model prove viable?

From an absolute growth perspective, Kling delivers impressive results. Over two years since launch, its quarterly revenue soared from RMB 150 million in Q1 2025 to RMB 650 million in Q1 2026, a year-on-year surge exceeding 300%. ARR climbed from $100 million in March last year to nearly $500 million this March, a fourfold increase in one year. The company raised its 2026 full-year ARR guidance from $300 million to $500 million.

Source: Kuaishou's official Weibo account

In terms of revenue structure, roughly 70% of Kling's revenue comes from B-side APIs and enterprise services, with overseas markets contributing 70%. By Q1's end, Kling had amassed over 60 million global users, serving over 30,000 enterprise clients and developers. From an application perspective, using AI to produce micro-short dramas can slash costs to below one-third of traditional levels and shorten production cycles by over 60%. In film and television production, the TV series "Peaceful Years" compressed a two-month production task to under two weeks after adopting Kling.

However, compared to global AI video competitors, Kling's revenue scale remains in its infancy.

Overseas video generation leader Runway is valued at approximately $5.3 billion, with an actual mid-2026 ARR around $150 million and a full-year target of $265 million. Kling's lofty valuation implies The premium offered by the capital market is Runway Several times over , reflecting expectations of scarcity as the "world's first AI video stock," but also demanding stricter validation of its unit economic model.

Because alongside revenue growth comes rigidly climbing computing costs. Video generation inference demands roughly 2,000 times the computing power of traditional text large models, with physical costs ranging between $0.5 to $1 for generating a 1080P video.

In contrast, the lesson from Sora's shutdown is stark: Sora's cumulative total revenue from in-app purchases reached just approximately $2.1 million (approximately RMB 14.51 million), yet daily computing costs to maintain operations hit $15 million—meaning its cumulative earnings couldn't cover a single day's computing expenses at peak demand.

Kuaishou's situation, while better than Sora's, still faces structural cost pressures.

Of the projected RMB 26 billion in 2026 capital expenditures, the vast majority flows to Kling's computing power procurement and model iteration. Although Kling has achieved positive marginal profits at the inference level, with ARR quadrupling in a year, the parent company's capital expenditures have simultaneously doubled.

Fitch Ratings' Asia-Pacific Corporate Rating Director, He Jingxing, remains cautious: "Fitch currently does not anticipate Kling will make a highly significant contribution to Kuaishou's profitability or cash flow in the coming years."

A deeper issue is that Kling fundamentally reversed its revenue structure in Q1 2026, with B-side APIs surging to approximately 60% and P-side professional user subscriptions dropping to around 40%. The core risk points in its unit economic model have shifted accordingly, from "declining C-side payment willingness" to "price competition pressure in B-side API invocations" and "scaled computing cost control capabilities."

The true meaning of a "commercialization inflection point" lies not in how fast ARR grows, but whether the unit economic model holds—whether revenue from a single video generation can cover computing costs and achieve scaled profitability. The question is: Can computing cost growth, which rises with usage, be outpaced by unit cost reductions from technological iteration?

This presents a dilemma.

If video generation tools price far above user psychological expectations (e.g., users expect a single generation cost below $0.1, while actual costs range from $0.5–$1), payment conversion will suffer.

If pricing is too low, computing costs will erode gross margins, with scaling deepening losses.

The solution depends on three variables: First, whether model efficiency optimizations can reduce single-generation costs below $0.1; second, whether ARPU (average revenue per user) from professional creators and enterprise clients can Continuously improving to cover costs ; third, whether computing hardware prices continue to decline.

Leading securities firm research reports generally estimate that if inference chip costs drop 40% annually and model compression technologies keep iterating, video generation unit costs could decline by approximately 60% over the next 18–24 months.

However, all three paths currently carry significant uncertainty.

OpenAI's shutdown of Sora has demonstrated that video generation projects relying solely on technological breakthroughs without viable commercialization logic struggle to succeed.

OpenAI's Sora Service Termination Notice

Leveraging Kuaishou's content ecosystem and commercialization team, Kling holds inherent advantages in customer acquisition, conversion, and retention. However, whether this "short-video DNA transplantation" logic can persist post-independence remains to be validated.

Notably, if Kling ultimately succeeds in an independent listing, it will chart a viable path for China's internet industry to capitalize high-quality AI assets from tech giants independently.

Coincidentally, competitors are making similar preparations: ByteDance's Doubao has completed institutional groundwork through an independent equity incentive plan, Alibaba has organized for independence via the Token Foundry Division's establishment, and Baidu's Kunlunxin has entered Tutoring stage for the STAR Market...

Yet the true test lies not in how high pre-money valuations climb, but whether sustained market pricing logic aligns with companies' actual operating performance post-funding rounds and listing.

After all, as "Klings" embark on independent financing, the market will no longer ask only about ARR growth—it will relentlessly pursue a harder question: When will they turn a profit?