ft.com web signal

Kling, Seedance outpace US rivals in AI video

Key insights

  • Kling has reached a $500M annualised revenue run rate, a commercial threshold no US AI video model has matched.
  • Chinese video models generate content at roughly $0.04 per second, substantially undercutting US competitor pricing.
  • ByteDance and Kuaishou's short-form platform ownership creates proprietary data feedback loops unavailable to US-only labs.

Why this matters

US AI labs building generative video products face a structural disadvantage that pricing or model quality alone cannot close: Chinese competitors have vertically integrated distribution, monetization, and training data pipelines through platforms like Douyin and Kuaishou, creating compounding advantages that widen each quarter. The $500M Kling revenue figure signals that the commercial video generation market is already consolidating around incumbents with platform leverage, not necessarily best-in-class models. Founders and investors treating video generation as a purely technical competition risk underestimating how deeply the flywheel dynamic has already shaped which players can afford to iterate fastest.

Summary

Kuaishou's Kling and ByteDance's Seedance have moved decisively ahead of US competitors in commercial AI video generation, with Kling alone posting a $500M annualised revenue run rate while OpenAI's Sora struggles to generate meaningful revenue and faces SAG-AFTRA litigation over training data. The structural advantage isn't pricing alone, though that's stark: Chinese models generate video at roughly $0.04 per second, with 4K native output and multilingual support as baseline features. The deeper edge is ecosystem integration. Both Kuaishou and ByteDance operate massive short-form video platforms that create data feedback loops at a scale simply unavailable to US labs, which sit outside those distribution and monetization infrastructures entirely. Essentially: (Kling, Seedance) have platform flywheels baked in; (Sora, Runway) are selling into markets they don't control. - Kling's $500M annualised run rate is a commercial milestone no US video model has matched - Chinese models benefit from direct integration into existing creator monetization pipelines, accelerating adoption and generating proprietary training signal - Sora faces both a revenue gap and legal exposure from SAG-AFTRA's training data lawsuit simultaneously The video generation race is increasingly a distribution story, and Chinese labs entered it already owning the distribution.

Potential risks and opportunities

Risks

  • OpenAI faces simultaneous revenue underperformance on Sora and active SAG-AFTRA litigation, which could result in injunctive relief or licensing costs that further widen the competitive gap with Chinese rivals by end of 2026
  • US-based video generation startups (Runway, Pika, Luma) risk margin compression and customer churn if Chinese models at $0.04/sec pricing become accessible to Western buyers through API resellers or VPN-adjacent distribution
  • Hollywood studios and production companies that signed early deals with US video AI vendors may face contract renegotiation pressure as Chinese model capabilities reach parity or better at a fraction of the cost

Opportunities

  • Western short-form platforms (Snapchat, YouTube Shorts, Instagram Reels) could license or acquire US video generation capabilities to create their own data flywheels before Chinese models expand distribution access
  • Enterprise video production buyers (advertising agencies, game studios) gain near-term leverage to renegotiate US vendor pricing using Chinese model benchmarks as credible alternatives
  • Inference optimization specialists and cloud providers (Together AI, Fireworks AI, CoreWeave) can position as cost-reduction partners for US video AI labs seeking to close the per-second pricing gap against Kling and Seedance

What we don't know yet

  • Whether Kling or Seedance models trained on short-form platform data are subject to the same creator IP litigation exposure now targeting Sora in the US
  • How quickly Chinese video generation pricing ($0.04/sec) can be matched by US labs, and which specific cost components (compute, inference optimization, data acquisition) account for the gap
  • Whether ByteDance's US regulatory exposure (ongoing TikTok proceedings) creates any near-term constraint on Seedance's expansion into Western commercial markets