Kling AI Just Hit $147M — And Sora Isn't Even Close

Reports this week claim Kling alone pulled $147M while OpenAI and Google dominated the headlines. Here's what the revenue numbers actually tell us about who's winning the AI video race.

Editorial disclosure: This article is published by VO3 AI (vo3ai.com), an AI video generation platform built on Veo 3. We cover the broader AI video industry — including direct and indirect competitors — because our readers are creators, marketers and founders who need an honest map of the landscape. Quotes and figures below are attributed to their original sources.

The Headline Nobody Wanted to Print

For the past six months, English-language tech coverage of AI video has revolved around three names: Sora, Veo, and Runway. Demo reels go viral. CEOs do podcast tours. Analysts predict who will "win Hollywood."

Meanwhile, a very different number quietly surfaced on X this week.

Chinese AI firms including ByteDance, Kuaishou, and Alibaba are reportedly outpacing US rivals in video generation, with Kling alone generating $147M.

— @TheAgentTimes, May 18 2026

That figure — sourced to The Agent Times' reporting and circulated widely on Chinese-language tech accounts — is striking for one reason: it is a revenue number, not a hype number. Sora 2 has not disclosed paid revenue. Veo 3 is bundled inside Gemini Advanced and Vertex AI subscriptions. Runway publishes neither. Yet Kuaishou's standalone video product, sold mostly to Chinese SMBs and creator agencies, appears to be the first AI video tool to publicly cross the nine-figure annualized mark.

If the number holds up under audit — and it has not been formally confirmed by Kuaishou Investor Relations as of this writing — it would reframe the entire industry narrative.

The Quiet Operators Versus The Loud Ones

Here is what the public scoreboard actually looks like in May 2026, based on observable signals (paid case studies, public commerce, third-party usage reports):

What the numbers suggest is something most Western coverage misses: commercial AI video is not being won by the lab with the best demo. It's being won by the company with the deepest distribution into actual paying customers.

Kuaishou ships Kling to roughly 700 million monthly users via its existing short-video app. That's a built-in audience an order of magnitude larger than the seat count of any Western generative video product. ByteDance's Seedance is wired directly into CapCut, the world's most popular video editor.

What Creators Are Actually Doing With This

The revenue figure isn't theoretical. Working creators are quietly shipping production work with Chinese video models right now.

Frank Larri — a creator with a sizable audience on X — describes producing a full client product video using Kling AI with "no camera. No crew. No studio. No travel. Just a clear brief and a 40-minute tour." That workflow used to require a five-figure budget. He doesn't disclose the client's fee, but agency rate cards suggest the savings on production alone routinely exceed $8,000–$15,000 per video.

Meanwhile, Higgsfield — the dark-horse Western entry — keeps showing up in founder polls as the tool of choice for marketing and brand work:

The author of that post used Seedance 2.0 inside Higgsfield — interestingly, a Western platform serving a Chinese model — to produce a generation he describes as surprising for both quality and speed. We have not independently benchmarked Seedance 2.0 against Sora 2 or Veo 3, so we are not making any claim here about which is "faster" or "better." But the fact that creators are reaching for the Chinese model inside a Western interface tells you something about the product-market fit conversation.

Why The Western Narrative Lagged

There are three reasons US-based tech press has been slow to catch this:

1. Demo coverage favors familiar names. A new Sora 2 clip drives more impressions than a Kuaishou earnings note, so editors prioritize the former.
2. The Chinese tools mostly ship in Mandarin first. Kling and Seedance have English interfaces, but their documentation, community tutorials and case studies are heavily Chinese-language.
3. Revenue from SMB customers doesn't generate the same press cycle as VC funding rounds. A $147M ARR claim from a profitable mature business gets less coverage than a $30M Series A.

That doesn't mean Western tools have lost. It means the score is closer than the headlines suggest — and the commercial winners may not be the same as the demo-reel winners.

What This Means For Your Next Production

If you make videos for a living — or for clients — here is the practical takeaway from this week's data:

• Stop picking tools based on Twitter hype. Pick the tool whose strengths match your specific brief. Product shots? Cinematic narrative? Testimonials? Each model has tradeoffs.
• Treat AI video as a portfolio, not a pick-one decision. Most working creators we talk to keep accounts on 3–4 generators and route briefs to whichever model handles each style best.
• Watch Google I/O this week. Multiple credible accounts are flagging a possible Veo 4 reveal, which would be Google's response to Kling's commercial traction:

The creator workflow is converging on a pattern: write the brief in Claude or ChatGPT, generate a contact sheet of stills, then pass selected frames into a video generator with a tightly specified prompt. The model you choose at the final step matters less than the prompt discipline that came before it.

A Concrete Example From Our Own Stack

Here are two recent generations from VO3 AI, both produced with single-prompt runs on Veo 3. We're including them so you can judge the output quality yourself — not as a sales pitch, but as a benchmark you can compare against your own results on Kling, Higgsfield, or Sora.

Generated with VO3 AI — testimonial-style spot for a defense attorney, single prompt, no manual editing.

Generated with VO3 AI — phone-shot food truck close-up, also a single-prompt run.

A fair-minded reader will look at those and conclude what we conclude internally: Veo 3 is excellent at certain styles (cinematic realism, controlled close-ups) and Kling appears to be excellent at others (motion-heavy product shots, longer durations). The honest answer to "which is best?" is "it depends what you're shooting."

Try It Yourself

If you'd like to test where Veo 3 falls on the scoreboard for your specific use case, you can run a free generation at vo3ai.com. Bring a brief that's been giving you trouble on another platform and see how it handles. The point is not to convince you Veo 3 wins — the point is to give you another data point in a market where the leader changes quarter by quarter.

Kling's reported $147M is a wake-up call for Western creators who assumed this race was already decided. It isn't. Run your own tests, follow the money, and ignore the demos.