Runway Raises $315M for AI World Models

0:00Welcome to the podcast. I'm your host, Jaden Schaefer. Today, I want to talk about Runway. This is a company that I feel like I haven't talked about in a little while. It's an AI video company. It's one of the number one. It was kind of out before any of the other AI video companies were out. So Runway has kind of been leading the way. But sometimes it feels like they fall a little bit behind when there's a lot of hype from Google's VO3 model or even OpenAI's Sora. It is a really solid company and there's some interesting stuff going on with it. They just raised$315 million in a Series E and it's about double their valuation.

0:31and now they have a 5.3 billion dollar valuation and all of this new money that they've just raised is basically going to be going towards what they say is the next frontier in AI which is these world models. So today on the podcast I want to talk about what Runway is doing with world models there's also a lot of new regulation coming out of countries like India on deep fakes so I'll talk about the state of AI video in general and where kind of Sora and Google's VO3 fit into this and this is all technology that I personally have been using a ton lately so I'm pretty excited about this Before we get into this, I have a super exciting announcement, which you've probably heard me say for the last couple episodes, which is that my own startup, AIbox.ai, you've probably heard me talk about it 100 times, it essentially gives you access to over 40 of the top AI models in one place.

1:16We just released new pricing tiers for it, so you can actually get access to all 40 of those AI models for$8.99 a month. This should save you a ton of money. you get access to open AI, 11 labs, Anthropic, Google, Gemini, Grok, like every single AI model for$9 a month in one place. It's like chatting with ChatGPT, but you just switch the model anytime because they all have different strengths and weaknesses and you get a bunch of image and audio models in there. And if that wasn't good enough, you also have the ability to, if you pay for yearly, you get to save 20 % on it. So for the standard plan, if you're already paying 20 bucks a month for it, which I know a lot of listeners are, you can get that 20 % cheaper if you pay for annual and you get that on all we've got a bunch of new tiers so go check it out there's a new 899 tier and let's get into the episode today so there was a big blog post that just came out which came from runway and they were essentially kind of announcing this big raise this 315 million dollars and they specifically kind of outlined what they plan on using the money for so it's going to be used to quote-unquote pre-train the next generation of world models and bring them to new products and industries.

2:22So world models essentially are systems that build internal representations of environments so they can simulate outcomes and plan ahead. This is, I'm reading that there, but basically what's happening is there's a lot of researchers that think that these are necessary, right? They're basically building these simulated worlds, which it's insane to think about, but they use a lot of 3D technology and simulators and stuff. But when you're making AI generated in video, it's so much more complicated than just, you know, with chat GPT or these LLMs where they're predicting the next character or the next token in a sentence when you're doing video.

2:57And you know, someone's like, for example, like I've been using Google VO three a lot lately. And I'll say like, Hey, like I want, you know, uh, an image of this certain shoe with a certain logo branding on the shoe. And I want the camera to pan around the shoe. And I want the shoe to be spinning and then lightning to strike. Like it's so creative, but also as the camera pans and changes the angle, the video AI model has to understand what everything around that shoe looks like otherwise when the camera pans around it's going to be glitchy and kind of weird so what they're doing in order to make better and better AI video models is basically to make 3D simulated environments and so if you're the you know the user telling the camera to point in any which direction the AI has already created the whole environment and that's easy you can just kind of maneuver the camera in any direction.

3:41So basically, this is kind of the future of AI in video, you have to do a lot of interesting things with this, you have to be able to do reasoning, you have to simulate space, you have to simulate physics and time, there's a lot that goes into AI video. So I think it's just one of the most fascinating technologies. Runway first became really popular because they had a physics aware video generation tool, which I think a lot of people in media and entertainment and advertising jumped on board with and they recently released their very first world model they now see that technology as kind of the foundation so not just for creative work but also for things like robotics climate modeling energy systems and even medicine so it's interesting because this technology was originally built just for their video models and it's very quickly getting rolled out into a lot of different areas so they have a lot of partnerships they're doing collaborations with Adobe right now to help in some of their creative, you know, sweet runway says that they're increasingly seeing a lot of traction in gaming and robotics.

4:41You can imagine these simulated worlds for gaming, that makes a lot of sense. And I think gaming studios are going to be using this a lot in the future, where you can have AI maybe create custom game environments per player, which I think is a fascinating concept. Robotics, obviously, they need to be able to understand the whole world around them, even for self driving cars, having this type of technology, you can create synthetic data of you know a road or or you know a city that the car can go learn how to drive around so i think there's uh runways kind of has a much bigger ambitions than just video but they are doing video quite well um this also is coming right after the launch of gen 4.5 which is runway's latest video model that update in particular basically let them do hd video generation from text prompts uh they do um a bunch of native audio that also got added and these are things that you Sora and VO3 had.

5:33So it felt like some people thought they were kind of catching up, but Runway really has been at the forefront for a long time. They also are now supporting long form multi-shot storytelling. And this is something that Google VO is not doing because they mostly will just do individual clips and you got to go link them together yourself. So this is really a fascinating step in my opinion. Their Gen 4.5 has apparently outperformed a bunch of competing AI models. They say that they're better than OpenAI's Sora and also Google's VO3. And so, I mean, obviously when you come up with a new model, you don't want it to be worse than the competitors.

6:07But like OpenAI and Google are pretty big competitors. And so being able to beat them on the benchmarks, being able to make AI models that are outperforming them on that specific area of video, I think is fantastic. And that's kind of what I like to see. At the same time, I think Runway has taken a much more cautious stance than some of their competitors. when it comes to generating realistic videos of people. So Runway right now, I think, historically has emphasized putting guardrails around impersonation and consent and identity misuse. They've limited the ability to create any sort of convincing deepfakes of real people without safeguards.

6:45That approach, I think, is, to be honest, it's kind of a big contrast to what we're seeing across the AI industry in general. You have OpenAI who, you know, their whole Sora model was built on being able to make deep fakes of people. You can just basically sign up your likeness and image on Sora for anyone to make deep fakes of you. Google also has, you know, the capabilities for you to just upload a picture of any person and have them saying and doing anything you want. And likewise, you can do that with X's Grok video generator, which landed them in a ton of hot water because they have, you know, you're able to do that.

7:18and also they have quote unquote spicy mode. So it's basically a recipe for disaster on that front. So I think they have done a decent job on layering in a lot of restrictions, watermarking systems. And I think a lot of regulators are scrutinizing this synthetic kind of these AI generated videos a lot more. And because Runway has kind of put an emphasis on this, I think it's gonna be important for them, specifically working with enterprise, right? And like, obviously there's gonna be open source models and video models out there that anyone can use for anything, for better or for worse. But at the end of the day, I think if you're a serious company and you're trying to attract enterprise and you're trying to like think it's a real business, we see that, you know, a lot of the drama and a lot of the pushback that happened to XAI's Grok video generation model.

8:05I think a lot of companies want to avoid that, especially if, you know, you're trying to have a reputation and Walmart wants to make ads on you. They don't want to make ads on a platform that has a ton of bad PR and, you know, is being used for sketchy stuff. So I think right now, as far as the world models competition is going, there is some competition. Fifi Lee created World Labs and Google Mind. Both of those have kind of released some research in public-facing systems in the last couple of months that are competing here. I think that basically this is showing that this kind of simulation-based AI is becoming the biggest new battleground.

8:39I think for a lot of labs, the long-term goal isn't just generating content, but it's basically building a system that can understand and interact with these really complex environments. And they're trying to do it in a more human-like way. So beyond just making new models, Runway is also scaling their infrastructure. They recently signed a compute agreement with CoreWeave to expand their capacity. I think this is a big move right now when, you know, training models is just using so much power. And it's sometimes it's actually hard to get the contracts for these AI companies. So investors right now, I think, are a little bit more assured that Runway is going to have enough compute resources to build these new models, right?

9:19It's like one thing to give them a ton of money, but it's another thing to say, like, do you have enough, you know, do you have enough access to compute? That was like something I remember reading a memo maybe a year ago that was kind of shocking to me where Sam Altman was saying he was concerned that XAI, Elon's company, was going to get access to more compute than him because he was kind of moving around and sharing chips between SpaceX and Tesla. And he was like, you know, if he does, he's gonna be able to make better models than OpenAI has. So it's definitely something that's top of mind. So in this whole round that just happened for Runway, though, it's their Series E, I'll say it was led by General Atlantic, they had Nvidia participating.

9:53I mean, it would be shocking if Nvidia didn't participate in this, frankly. Fidelity Management, Adobe Ventures, Emphatic Capital, AMD Ventures. There's a lot of big people that are playing into this, but Runway basically is going to be using this money to expand. They have about 140 people right now. That's across a few different teams. They have research engineering and kind of their go-to-market, their marketing teams. And so they're going to expand their teams with this money. And they're also doing this right as I think an interesting moment is happening in AI and video, which is that governments right now are really focusing more on these AI generated deepfakes.

10:30In India, there's a new law to the country's IT rules, which is basically bringing deepfakes under a new regulatory framework. And they're mandating that all deepfakes are labeled, traceable, and then they're rapidly taken down. If there's any sort of synthetic audio visual content that is, you know, a deepfake of some actual person, it's got to be removed quickly. So I think because of that, like most of this pressure, I don't think necessarily falls directly on someone like runway, it's going to be more falling on Google, it's going to be more falling on sorry, Facebook and Instagram and TikTok and a lot of those types of I think TikTok's banned in India if I remember correctly, but a lot of those kind of social media platforms.

11:11But I think there is, you know, billions of internet users, India is one of the world's largest digital markets. So being compliant there and some of their compliance rules that they make there do influence global moderation practices, especially, you know, as platforms are required to label synthetic media. and you know it'd be kind of weird if they only labeled it in India and other places although they very well may do that but it'll be interesting to see what happens so anyways there's a lot of there's a lot going on there I think if you kind of take all these things together basically these forces right now that we're seeing in kind of the AI video space show there's a lot of tension in the next phase of AI video right we have it's kind of rapid technical progress on the one hand and on the other hand we have a lot of new regulatory and kind of ethical pressure on the other runway Runway, I think right now is doing a really good job of kind of getting into both of those worlds.

12:03They have a very sophisticated model and a framework for how they build their models. They're positioned around being kind of like responsible and working with enterprises to create real video and working with, you know, media production companies. So I think this is going to be interesting. Overall, a great company to keep your eye on. I would not count Runway out. And yeah, this is a great, great news for them to be able to raise$315 million more in this latest round of funding. Thanks so much for tuning into the podcast. If you enjoyed the episode, make sure to leave a rating and review wherever you get your podcasts.

12:36And as always, make sure to go check out AIbox.ai. You can get the brand new$8.99 a month subscription and get access to over 40 of the top AI models. So you can cancel a ton of your subscriptions, save a lot of money. But more importantly than say money, I think it's just really nice to consolidate all of your AI chats and content into one platform. The other day I was remembering like a really long and well thought out chat I had and I'm like, I'm like, dang, was that on AI box? Or was that on chat GPT? Or did I do that on Anthropic? Because I have to have subscriptions to every platform whenever there's a new feature, I got to test it out.

13:09Anyways, it was a pain in the butt. And I'm like, dang, I really just need to only ever use AI box. So that it's all in one place. And that's hopefully what the goal of this is. So you can go check it out. It is AIbox.ai. I'll leave a link in the description. All right. Catch you guys in the next episode.