Earlier this year, OpenAI quietly shut down Sora.

Not downsized. Not restructured. Shut down. The AI video model that launched at the Super Bowl, generated the most press coverage of any AI product in 2024, and caused Hollywood executives to lose sleep — gone.

The reason was simple, even if no one at OpenAI said it directly: Sora could not compete. Not on price. Not on quality. Not on speed. Not on adoption.

While OpenAI was charging $20 a month just for access to Sora's basic features, a Chinese company called Kuaishou had spent 19 months building something that 60 million creators were using for free or on affordable paid tiers — and generating 600 million videos in the process.

What happened to AI video is the same story that happened to Chinese open-source AI language models. It just happened faster, more completely, and with even less coverage in Western tech media.

Seven of the top eight Chinese AI video models in the world are now Chinese. And the eighth — Runway, the only American entry — is a scrappy 100-person team that outbuilt trillion-dollar giants on sheer focus.

This is not the future. This already happened.

The Numbers Behind Kling's Takeover

The lead story in Chinese AI video is Kling, built by Kuaishou — the same company that runs China's second-largest short video platform after TikTok.

Kling launched in June 2024. By April 2025 — just 10 months in — it crossed a $100 million annualized revenue run rate. That milestone, which takes most SaaS businesses five to eight years, happened in under a year.

By December 2025, Kling's annualized revenue run rate had more than doubled to $240 million. Monthly revenue exceeded $20 million. The platform had over 60 million registered creators worldwide. More than 600 million videos had been generated.

In Q1 2026, Kuaishou reported that Kling's revenue surged more than 300% year-over-year to 650 million yuan ($90 million) in a single quarter. The platform held the top position in the App Store across 42 markets, including Brazil and Germany. Kuaishou CEO Cheng Yixiao told investors the company expects Kling revenue to more than double again across full-year 2026.

Kling is not a niche product for Chinese users. It is the global default for AI video generation — and it got there in 19 months.

The Competitive Landscape: What You Are Actually Choosing Between

The AI video market in 2026 looks nothing like the market Western media describes. The full ranking from Artificial Analysis benchmarks, as of May 2026, shows Chinese dominance that is difficult to overstate.

At the top tier in quality: Runway Gen-4.5 (USA) holds the number one slot. Below it: Seedance 2.0 (ByteDance), Vidu Q3 Pro (Shengshu AI), Kling 3.0 (Kuaishou), and Wan 2.6 (Alibaba WanX) occupy positions 2 through 5. Google's Veo 3.1 and the final remnants of Sora 2 appear further down the list.

Seven of the top eight. All Chinese, except Runway.

And the cost differential is the same story that plays out across every AI pricing model.

Sora 2, before it was shut down, required a $20/month ChatGPT Plus subscription for basic access. Kling 3.0 costs $0.153 per second of generated video through its API. Wan 2.6 — the most open-source-friendly of the major Chinese video models, with less content filtering — runs at $0.018 per second. Vidu Q3 Pro comes in at $0.06 per second.

For a production team generating 10 minutes of video per month for marketing or content creation, the cost difference between Sora-class Western tools and Wan 2.6 is roughly $300 versus $11. Same order of magnitude gap as LLM tokens.

Why This Matters Beyond Content Creation

Most coverage of AI video treats this as a story about filmmakers and YouTubers. That is the wrong frame.

The real story is about marketing at scale, advertising production, product visualization, sales enablement, and synthetic media for business — and the fact that Chinese models just made all of it 10 to 30 times cheaper than it was 18 months ago.

Kuaishou's own earnings data tells this story directly. The company reported that AI-generated content is now being used by merchants on its e-commerce platform to automatically produce product highlight clips from livestreams. Daily gross merchandise volume from AI-generated livestream clips increased more than 300% year-over-year in Q1 2025. By Q1 2026, Kling's commercial clients span marketing, film and television, short drama, and games — with what Kuaishou calls a "high enterprise renewal rate."

The use cases are not creative experiments. They are production pipelines.

A marketing team that previously needed three days and a production budget to create a 30-second ad can now generate it in under an hour at near-zero marginal cost. A SaaS company can produce personalized onboarding videos for enterprise customers. A real estate firm can create property walk-throughs without a camera crew.

These are not hypotheticals. These are live use cases running on Kling's API today.

ByteDance's Entry: Seedance and the Race for Technical Leadership

If Kling is the commercial leader, Seedance from ByteDance is the technical challenger — and its January 2026 release of Seedance 2.0 may represent the most significant model architecture advancement in AI video since the category was invented.

Where most AI video models treat generation as a single-pass task — here is your text prompt, here is your video — Seedance 2.0 introduces what industry analysts at Global Semi Research called "a critical shift": video generation moving from capability demonstration toward becoming a production tool for repeated, practical use.

Seedance 2.0's architecture integrates multiple input formats simultaneously: text, images, video clips, sound, and motion data. The MVL (multi-modal visual language) framework means a creator can provide a reference image, describe motion, include an audio cue, and specify camera movement — all in a single generation pass. The output reflects all of those inputs in a way that single-modal prompting simply cannot match.

ByteDance simultaneously released Seedream 5.0 Preview for image generation and Doubao 2.0, its omni-modal foundation model, in the same week. The pace of concurrent multimodal releases from a single Chinese company in a single week was, by any measure, without precedent in the industry.

The Artificial Analysis intelligence index ranks multiple ByteDance and Kuaishou models in the top 8 globally. No other country has two companies in that tier simultaneously.

The Open-Source Angle: Wan 2.6 and What It Means for Builders

For builders, founders, and developers — the most interesting player in the AI video landscape may not be the most obvious one.

Wan (formerly WanX) from Alibaba is the most permissively licensed high-quality video model available anywhere. The 2.6 version, available through Wan's API and also available for self-hosting, delivers benchmark performance that rivals Kling and Vidu at $0.018 per second — the lowest cost of any top-tier model.

More importantly for builders: Wan 2.6 has a comparatively relaxed content filtering approach, which makes it the preferred choice for developers who need reliable prompt adherence across a wider range of creative content types.

Wan follows the same playbook that made DeepSeek and Qwen dominant in language models: build at frontier quality, license openly, compete aggressively on price, and capture the developer ecosystem before anyone else does.

The self-hosting case for Wan is less developed than for language models — video generation requires significantly more GPU compute per output than text — but the API pricing is already so low that for most production workloads, the API is the right answer before you even get to the self-hosting question.

Stanford's 2026 AI Index documented that AI video capabilities have advanced dramatically: Google's Veo 3, tested across more than 18,000 generated videos, demonstrated abilities like simulating physics without being explicitly trained on those tasks. Chinese models at equivalent capability levels are available at a fraction of the cost. The performance convergence that characterized the LLM market has arrived in video — and the pricing divergence has arrived with it.

What the Sora Shutdown Actually Signals

The discontinuation of Sora deserves a sentence on its own, because it is the clearest single data point about where this market has moved.

OpenAI built Sora with enormous resources, launched it with enormous marketing, and had the brand recognition of the most famous AI company in the world behind it. None of that was enough.

The market did not move away from Sora because of any technical failure. Sora's quality was competitive when it launched. The market moved because Kling, Seedance, Wan, and Vidu were generating better results at lower costs with more accessible pricing tiers and faster iteration cycles. In Q1 2026 Reddit threads on Sora alternatives, the developer community's consensus was clear: Kling for quality and API reliability, Wan for open access and cost, Seedance for cinematic production work, Vidu for premium aesthetics.

OpenAI's premium-priced, closed-access model lost to an ecosystem of cheaper, faster, more accessible Chinese alternatives. The same story. A different modality.

What Builders and Business Owners Should Do Right Now

The practical takeaway from this market shift is simpler than it might appear.

If you are creating marketing content, product demos, or social media video at any scale, you should benchmark Kling 3.0 against whatever you are currently using. Its API is live, globally accessible, and its $0.153/second pricing means a full minute of generated video costs less than $10.

If you are building a product that includes video generation, Wan 2.6's API and its open-weight release gives you the lowest-cost, most permissive option in the market. Kling's API gives you the highest commercial reliability at still-aggressive pricing.

If you are in marketing automation, ad-tech, or e-commerce, the use case data from Kuaishou's own platform should be your proof of concept. AI-generated product clips running at 300% year-over-year GMV growth is not a test result. It is a production outcome.

The AI video revolution that Western tech media expected to be led by OpenAI and Google is being led by Kuaishou and ByteDance. The economics are the same as the language model story. The playbook is the same. The outcome, so far, is tracking the same.

The question for every business using video — for marketing, sales, training, or product — is the same one that every business using AI language models had to ask 18 months ago:

Are you still paying a premium for a product that has already been commoditized at a fraction of the price?