Runway is coming for Google 🤯

Good morning. Runway says it's not trying to be a video tool anymore, it's trying to build AI that understands the physical world. Chinese AI labs are now outranking American rivals in video generation quality. And a real bear showed up on a family's Ring camera and their first reaction was "I thought it was AI." Have you been fooled by a fake security cam clip yet? Hit reply and tell me. We cover all three below.

Runway just laid out something bigger than a product update. The company's co-founder Anastasis Germanidis said the next frontier of AI won't come from language models trained on text. It will come from video and world models that learn how the physical world actually works. His argument: language models are trained on the internet, on message boards and textbooks. To get beyond that, you need less biased data. Observational data. Video.

Runway is now valued at $5.3 billion and added $40 million in annual recurring revenue in Q2 2026 alone. It has deals with Lionsgate, AMC Networks, and Adobe. Its tools were used in "Everything Everywhere All At Once." But the real play isn't video generation. In December, Runway launched its first world model, an AI system that simulates environments well enough to predict how they'll behave. It's not alone. Luma, World Labs, and Google's Genie are all chasing the same thing.

For creators, this matters because the tools you're using today are a side effect of a much bigger bet. If Runway's thesis is right, video generation was just the starting point. The end goal is AI that can simulate reality. And whoever gets there first won't just make better clips. They'll reshape everything from filmmaking to robotics to gaming.

Chinese AI companies have moved ahead of American competitors in video generation. ByteDance's Seedance 2.0, Kuaishou's Kling, and newcomer HappyHorse 1.0 are all outranking US models on independent leaderboards based on blind user voting.

The reason comes down to data. ByteDance and Kuaishou operate some of the world's largest short-video platforms. They're training their models on massive proprietary libraries of real video content, plus behavioral data from every like, swipe, and share. That's a training advantage that's extremely hard to replicate. Video, unlike text, can't easily be scraped at scale. Google's Veo 3 is competitive thanks to YouTube footage, but comes with more content restrictions. OpenAI shut down its Sora standalone app in March because the compute costs were too high.

The other edge: Chinese tools are cheaper and have fewer content restrictions, which matters for creators who want flexibility. For the AI video space, this is a significant shift. The US still leads in language models and coding, but in the category that matters most to this audience, video, China is winning on quality, usability, and price.

A Santa Clarita family got a Ring doorbell notification Friday morning and assumed it was a package delivery. It was a black bear casually strolling up to their front door. The daughter's first reaction when her mom sent the clip: "I thought it was like a joke, or like AI, or something."

That reaction tells you everything about where we are right now. AI-generated security camera footage of animals doing absurd things has been flooding TikTok and Instagram for months. Bunnies on trampolines, cats playing instruments on porches, dogs bringing home wild animals. The clips use Ring and doorbell camera aesthetics with timestamps, night vision, low-res compression, and fisheye distortion. One fake bunny trampoline video hit 203 million views on TikTok before anyone called it out.

The format works because security camera footage was the last thing people trusted as "real." No one edits Ring footage. No one stages it. That implicit trust is exactly what makes it the perfect container for AI content. And now the trend has gotten so convincing that when a real bear actually shows up, the first instinct is to assume it's fake. That's the AI video landscape in 2026.