Runway raises $315M Series E to scale video generation compute and hit millions of media and entertainment customers

at this scale.

Well, we'll have to dig into it. Uh let's start the lambda lightning round. We already have the mallet down, but we have Chrisal Valenuela [music] returning to the show. He's the co-founder and CEO of Runway, and here he is in the TV again, Ultradome. Good to see you. How are you doing?

Hey, good to see you guys. How are you?

Congratulations. Give us the news. What happened today?

Yeah, a lot of news. We just raised our series and so we're announcing it.

How much?

Uh 315. That's five.

Big big numbers. I love this product. You know, I'm an OG,

Jordy.

Yeah. You weren't one of the first.

Yeah. Yeah. Back when it was like a rotoscoping tool almost. Not to sell it short, there were other features, but the roto was very very cool.

You invested [laughter] you had this a running joke. John John's like an early user of like every every you know to hundred billion dollar company. And uh no uh break break down what's been going on the last kind of few months. talk about traction, uh, all that good stuff, and then I want to talk about where this money's going toward,

of course. Yeah, there's a lot going on. And so, uh, obviously closed around late last year. I think we just like, uh, figure out now we should announce it. There's a lot going on just in the company model release. Um, we released 4.5, which uh, like is one of the best video models in the world, and it's been getting used uh, a lot, I think, a lot more than what we usually like plan for. And so, one of the biggest thing in like research and both inference is capacity. how do you manage like the amount of volume of users and like inference that we're seeing? And so scaling compute has been critical for us.

Um reminded like we're we're one of the perhaps like smaller uh frontier labs out there. Um and we're having and creating some of the most powerful like models and I think the way we do it is we're very like efficient you know with how we do things and the research itself. Mhm.

Um and so a lot of our racing and we're putting our funding towards is scaling that like machine we built to to get more training runs to get more pre-training of the next tier models

and then supporting inference like we have millions of customers who just require more runway and like supporting them is like

the biggest thing and the biggest obsession for us right now.

Yeah.

So yeah putting putting funds into that and then talent we're sure

um we're hiring across the board. So from from from research, from engineering, from sales, uh for pretty much everything at runway, which is which is great.

Yeah. Talk about uh going all in on on video generation, training models, uh the actual video generation from start to finish versus tools. I was watching Corridor Crew, you probably know those guys, and they were like, I still don't have a good roto tool, you know, and they were like sort of vibe coding and stuff. And it feels like there's that we're in the centaur period of video uh editing and I see vibriels on Instagram all the time that are like no video model could do exactly what they're doing because they're editing so fast and in such bizarre ways. Uh at the same time the models just keep getting better. That's sort of where you want to be. How do you think about the trade-off between uh building tools and harnesses and and tool use within video generation versus just like just all in on the video generation will solve every problem.

So so I think the reality these days that you have to do both. Um the way we call it at runway is we there's that the exploration mindset and exploitation like mindset. You need to be able to explore and create new models, pre-train them, build net new products, net new things. I think we've been relatively good at that. with first to market with video, first to market with role models with real time Jan like the surface area of what you can do there to help either filmmakers like the cor crew guys or other folks is just so big. At the same time like the gaps you can fill by building the right workflow on top of the right model or chaining models together to get to where you need to go is the way you educate the market to to like that next frontier of sorts. And so the best companies I would say are the ones that can do both. Yeah. If you only focus on one, you're detached from your user base, from the reality, from use cases. If you focus primarily on like uh workflows, you're going to get leaprogged. And this is the beard lesson. I think we have seen over the last we've been working around for seven, eight years now.

The beard lesson of like AI startups is that they eventually get leaprogged by a better like bigger much more capable model. Yeah. And so, but there's definitely a market opportunity while you're getting leaprog. And so, being good at like being ambidextrous of like you need to do your you need to use your left hand and your right hand all the time to like winer.

Yeah. Talk a little bit about tool use. It feels like nano banana definitely has access to some tools just to like cut out an image, overlay it. When I asked chatt to multiply two big numbers together, it doesn't just try and, you know, inference that. It actually writes some Python, executes it in a ripple. If I go to runway and I say generate me a video of, you know, beautiful mountains, it's going to do that flawlessly. If I say now make it black and white, is it going to know just to drop out the colors using like a traditional workflow or will it regenerate the footage because that feels like an extra step, wasteful, and then if I like the way those mountains look, the new mountains might be slightly off. So, how do you think about that that workflow and tool use coming to these models? Yeah, that's an interesting point. I think two two use might be like um like I wouldn't say like the best way to necessarily think about it. I think the the way the best way to think about I would say what video models are capable of is that in some way they have all these emerging properties that if you like tune them well into like we there's there's examples and research and I think others and we have proven that if you want to train a great rotoscoper a great scoping machine you can just take a baseline model and show it a couple of examples of rotoscoping and the model has and this is right where the world model like approach comes from it's like the model has

innate understanding of the world and so with the right examples then the model can learn that particular task without necessarily having to be like pre-trained for it, without necessarily having to see enough examples of it.

And so the most unique opportunity around here is that these are eventually simulation machines. That's that's what data models are. They're simulating the world and the way they simulate the world is by watching the world. Therefore, you can ask it for almost any task like remove things or uh add things or change how things are are seen, create a new novel camera angle for for example.

Yeah.

Um and and when you think about it that way, then simulation becomes extremely useful in other domains as well like robotics or self-driving cars or many other opportunities where you're not showing videos to humans, but you're showing videos to robots and robots are learning from those videos. Yeah.

And I think that video as a world simulation engine is perhaps the most impactful thing that AI that we can work in these days.

Yeah.

Uh how did you react to the Super Bowl? There was like the first AI Super Bowl ad. I have to imagine a lot of the other ads were using AI like in some way or another.

It's kind of hard to clock. They're 15-second ads. You kind of want to keep up with the program. So not necessarily pausing and looking at each pixel. I see six fingers.

Yeah.

Yeah. I think a lot of people I think more more ads were using I than I think people realize. But but it's hard to tell. And maybe that's the trick. Like you can tell.

Yeah.

Like uh I like the idea that you're making an ad and you're putting it out because it's good, not because it was made with AI,

but I'm pretty sure there are a lot of fats out there that were made with AI.

Yeah.

Um also first time I watched like a Super Bowl. I think I'm not a super like sports fan, but um we had I'm from Chile. We had a lot of like great Chilean representation there. Cool. Pascal was there. So biggest Chilean representation we've ever had.

Yeah. Uh who's uh who are the biggest customer segments? Uh some some of these apps have people that just you the video generation just for themselves. I was talking to Sam Alman about this. Like when I use Sora, it's just for like an in joke between me and Jordy. It goes in like a text chat. Then you talk to other companies and they'll be like we are puppet. We we are used by social media advertising agencies like crazy. Then there's other people that might be in Hollywood and they're like, "Hey, we just need to do background shot or stock footage." Uh, what's the biggest use case right now?

Uh, I mean biggest cases is just media all around like marketing, entertainment, film, pre-production, post-production. We've signed with all the studios out there. We've signed with a lot of agencies and brands, marketing teams. I think the default way you make things will be with AI and AI first. like it's just such a and you started to see this already where like studios are now hiring like AI like chiefs or like they're organizing their entire companies around like AI native workflows. I think that's the norm of what you'd expect over the next couple of years. U because the trade-offs are just so um are so big like you start to realize you can do things in days instead of months and so for us media entertainment just content all across from like social to much more professional will continue to be a source of growth for sure.

Yeah, it was eye opening yesterday. We we were on uh Fox Business for like a total of 3 minutes, but while we were in the waiting room to go on, we we watched like 5 minutes of ads. And every single one of the ads I was like, I feel like Runway could like not one shot this necessarily, but like it's,

you know, five different kind of scenes stitched together. And I don't know why you would in even in 2026 do this as a IRL production.

Yeah. No, I mean you don't have to. And I think that's what most agencies have realized, but also the marketing teams. We've signed with PayPal, with All State, we've signed with agencies across the board, with AMC, with Lions, like all of them,

all of those companies and brands and teams have realized like why would you spend six months hire an agency to do one work when you could just like do it internally as you were saying, not entirely oneshot it, but like kind of close to it to be honest.

That's amazing. Well, congratulations on all the success. Thank you so much for

sure you'll be back in this year. you've got a habit of, you know, you know, raising money every two months. So,

such an exciting company.

Yeah, maybe in two months we'll see you again.

Yeah, we'll see you soon.

Good stuff. Congrats to the team.

Good seeing you.

Thank you.

Bye. Let me tell you about Reream. One live stream, 30 plus destinations. If