AI Video, VC Metrics, and Crusoe's Cloud Ambition | July 23, 2025

0:13Welcome everyone to the Informations TI TV. My name is Akash Basricha. It is July 23rd. Today on the show, we are talking to the CEO of AI video company Runway about the company's enterprise sales approach. We are also talking to Tomas Tungus about the most interesting metrics that he has to assess AI startups that always promises to be a data heavy discussion. I'm excited for that one. And finally, we're going to end by talking to our cloud reporter about Crusoe, a data center company that is now making a much bigger play at competing with major cloud providers. That was a story that was published in the information this morning.

0:50Very excited for that conversation. Before we get started, I want to go through a couple headlines that are making news this morning. Alibaba has launched an open source AI coding model. They are basically going after Anthropik with this play. We, of course, know that tons of companies in China, including ByteDance, are working on AI coding models. And so you can add yet another company to the list. Here's an interesting story. Amazon bought Bee, an AI hardware startup, that the company sells a$50 wristband. It's kind of like a Fitbit. The only thing is that it transcribes your conversations because who doesn't want their trip to Starbucks on paper transcribed?

1:30I was looking at this headline and I couldn't help but recall Amazon's last play into wearables with its Halo Fitbit-like wristband. That obviously didn't go the way they wanted. They pulled out a couple years ago, but we are giving them all of our best wishes for this new play. AI wearables is certainly the thing to be in on these days. Last but not least, I want to go through some earnings that we've got tonight, Google and Tesla. I asked our Google reporter, Erin Wu, what she's watching for on Google tonight. She told me she is looking to see if chatbot competition eats into Google's search revenue.

2:07We, of course, know that search revenue slowed in the last quarter. And as it relates to Tesla, I asked our electric vehicles reporter, Steve Levine, what he's looking for. He told me two things. Number one, any update on removing safety drivers from the robo taxis that are roaming around Austin. And number two, he's also looking at the extent to which lower prices and lower deliveries are hitting the company's bottom line. Okay, let's get to our first guest. AI video has been all the rave lately, and few companies have been as central to that conversation as Runway. The company was last valued at$3 billion, and I want to bring on their co-founder and CEO, Chris Valenzuela, to talk about this moment for the company.

2:51Chris, thank you so much for being here and welcome to TITV. Yeah, of course. Thank you for having me. So I want to talk about your company's enterprise sales and product strategy. Help us understand what products you have that are targeted towards enterprise and also how you're pitching them to businesses. Sure. We work a lot with different kinds of enterprises, which is kind of interesting because you might think of Runway as a company that's mostly working with media, with Hollywood, with videos, which is true. But the thing is that many of the underlying capabilities and features of the models themselves are also very useful and applicable in wide domains of industries.

3:31We sell a lot to architects. We work with brands and with the e-commerce sites. And really, our enterprise solutions are not too different from what consumers or consumers can access. I would say the beauty about good generalizable models is that the use cases are broad. And so if you have enough of a good set of references, for example, that you can provide the model, then you can use them in a wide set of spectrums and kind of objectives. And so we're selling right now to all of the studios, media companies, and the offerings are just the best product. It's the best way of creating media, creating pixels, creating images, creating videos in a way that's secure, in a way that's fast, in a way that's reliable.

4:13We have a way of managing credits and managing subscriptions for your accounts and your teams, things that companies really care about. And all, of course, in a very safe environment as well. So I think it's a combination of different things that starts with a really good product and then happens kind of like across the board. So for hopefully every single task that they have when it comes to media creation. So when you break down the company's revenue right now, how much of it is coming from enterprises and how much of it is consumers? We see a growing trend from enterprises. I think it's a good sign because it tells and it means that companies are more ready than ever before.

4:51Look, selling to Hollywood two or three years ago was very tough, was very hard. People had a lot of concerns and considerations. And there was a lack of understanding what the models do, did, and how they could be useful. Now these days, I think everyone is on board. And so selling to them is way more easier. It's just a better understanding. and a lot of our growth, it's coming from those kind of inflection points. We've seen that. We're talking like 10 % of revenue, 20%, 50%. I mean, how much is coming from enterprises, right? It's growing a lot. We don't disclose numbers specifically, but it's a growing kind of like part of the business when it comes to growth.

5:32Like this is where we're spending perhaps most of our time these days. Correct. Because the demand is there. People want to use it. We get way more influx of requests that we can handle at any given point. And a lot of it has to do with helping people on board. People can just buy and sell, serve themselves with an enterprise plan, but we have the special offering. We were working on this idea for quite an awesome time where we deployed creatives inside the companies. And so you can think about it as kind of like what Palantir does with forward deploy engineers. We do forward deploy technical artists.

6:06Also, we have artists that go inside your studio, your company, your advertising agency, whatever form of function you have, and they help you drive adoption within the company itself. And the reason for that is most of what you can do with Runway is non-traditional. It kind of breaks the mold on how you've done content before. And so if your whole company has been working with Adobe software for 20 years, you can do the same thing for a fraction of the time of the cost with Runway. we just need to switch your mental models for the first time we need to understand this different way of working and so we get people on board we deploy them into those companies and that helps drive adoption significantly So these are like your own runway consultants inside your enterprise companies showing them how to use them and helping them transition I think for all of these companies this is a transformation I think the metaphor the equivalent I heard a lot of times inside these companies themselves, this feels like a cloud transition, you know?

7:08Where you were... You had to realize that the cloud was how you're going to do business like 10, 20 years ago. Do you charge them more for that? Yeah, there's a premium that we charge. We're deploying our employees to your company. So you get to work with them directly and also with the rest of the team. And I think that the value there is that once you get there, we have case studies or companies who start with 20, 40 licenses, and now they're gone to like a thousand licenses or so. And I think part of it is just you start understanding more of the workflow, you start understanding more of the use case, you fine tune your models for that, you keep improving.

7:44And so then it's like an obvious like trade-off. Talk to me about, you know, as it relates to what has been interesting for enterprises. I mean, are these mostly enterprise subscriptions? Is it access to the API? Where is your focus right now? It's a combination of both. on the API side we produced a couple new models most recently we did Act 2 which is this state of the art performance motion capture model that's been incredibly well received, it's one of our best products we have the API for that so there's a lot of use cases around custom workflows that people have specifically within high production and media and advertising that people are leveraging Runway and Act 2 for we have references, so references is an interesting product because references allows you to kind of customize or edit an existing image And so let's say you're an e-commerce site.

8:32We're selling to brands. We just sign a couple of some of the biggest luxury brands. And the way they could use it is you could take an asset that's either in your website or it's part of a campaign, and you can iterate and edit it and modify it either for specific customer needs. So you go into one of these websites, one of these companies, e-commerce sites, and you can put yourself in the clothes or the brands or the things you're buying. just by hitting the API once. Or you can think about a major e-commerce site that has millions of assets and they want to either remake some of those visuals so they're more complying.

9:12You can use the API for that. And so you have this programmatic, massive at scale editing and generation of content that people are using Granblue for. And then for more professional and specific and campaign-specific kind of workflows or movies or short films or ads, people pay for the subscription. So you pay for runway, you subscribe, you get the credits and use it within the interface. You have this team account so you can show your ass to other people. You have a manager. You have all the things that software professional enterprise software needs these days. So yeah, it depends on the need.

9:44And I think we're solving for both. You guys have gotten a lot of traction in Hollywood with your product. You're one of the few companies that Hollywood is actually really eager to work with. Are you working with Netflix? Alex? We're working with a wide spectrum of studios and companies. I think these days it's safe to assume that many of them are already exploring AI or using AI. We don't disclose our customers publicly, but it's safe to assume that I would say every single studio there and every single media company has a need and it's already aware that AI is perhaps the most important technology of a lifetime.

10:23I'm not exaggerating. This is something they have said publicly as well. And the reason for that is that content decreases, the cost of content decreases, the volume of content increases. And if you're in the business of media, that changes dramatically the way you're doing business. And so you need to be aware of it. And look, I think media companies and Hollywood studios and streaming companies and production companies have been thinking about this for quite some time. I think everyone, most of them are already fully on board on what AI means. And I think now we're transitioning to other industries.

10:55Now gaming companies are becoming very aware of it and understanding the affordances of what it means. So we want to make sure we can also help and serve them as well. I want to shift a bit to talk about sort of the cloud provider angle to AI. I mean, video is a very compute-intensive operation. You guys work with Microsoft and Amazon, am I right, as it relates to cloud providers? We probably have relationships with AWS and GCP. And tell me about just, you know, how do you think about the big three insofar as, you know, who's good for what? And how do you sort of delineate between them? I mean, they're all in their own way.

11:39I think it depends on what you're building. I think some companies like us that are mostly doing core research or frontier research, you just want to have access to the best infrastructure to do like research at scale. And so you want to partner with someone who can give you good prices, someone that can help you debug. There's a lot of new and upcoming challenges on managing the size of infrastructure that these companies are building. I don't think there's one or other that are better. I think they're different and different things. And so most people tend to have different multi-cloud provider options just so you can switch as you go.

12:12I mean, if we zoom out a bit, I wanted to ask about compute a little more broadly. I mean, you know, if you take sort of, you know, your company's sort of demand for compute over the past year, talk to me about how that's changed over the past year. It's increased. I think that's a implication. And it will continue to increase. I think that's a reality of both training models at scale and inference as well. Have you been able to lower your cost of compute at all? Yeah, of course. I mean, so I don't think there's those two things go like together. I think you're able to lower the cost, mostly because you're able to optimize the deployment of the models and inference workloads.

12:52And also, chips become cheaper. Like, innovation keeps on happening. And so you benefit from your real macro of where the industry is heading. And you should expect inference and training costs to continue to go down. Now, that's happening at the same time that you need to spend more time and more resources to train larger, bigger models. And so you spend more on compute. So you can see both things happening at the same time, perhaps at different ratios, because better chips and better GPUs are also constrained with logistics and just physical, the atoms are always harder. But these are something that engineers really love, is optimizing.

13:29So if you have a model, distilling it and finding ways of optimizing it so it's cheaper to run is something I think every company is working on these days. When you think about technical challenges that Runway is facing, as you continue to develop your products. What is the biggest technical challenge that your team is grappling with right now? Well, there are a few. I mean, from the one end, making sure these models are really controllable for the experiences that we want for our customers to have remains a big challenge. And that's what we've done and we saw with references. References is an image and video model that allows you to manipulate pixels in ways you couldn't do before.

14:07And a lot of it has to do with the insight that we get from our customers on what they really want. And so a challenge is how do you train models to do that? There's another challenge on like speed. So how do you make those models fast, really reliable, really like with the feedback loop in creation becomes like instantaneous. And I think that comes with like a list of engineering research problems as well. And a lot of it has also to do with helping people transition from this very obsolete, outdated workflows they've been relying for many years to a completely different way of making things.

14:38and that also tells has a just a market challenge as well you need to help people understand kind of how this works um and i think we're working all of them i think we have a very unique perspective we've been working on learning for quite some years now like almost almost seven years and so we've getting we've gotten a lot of experience i would say and intuition on what works and i think one of the things that is perhaps underestimated these days in research people think about compute and and you know and and data and like talent but i think intuition and taste and direction of where you want to go, you could technically go anywhere research-wise.

15:10Peeking the right vertical or the right direction becomes incredibly powerful. And you can see some companies, like major big companies, who've chosen the wrong path, and it's cost them a lot. And so intuition for us has become critical in the way of solving some of those challenges. I want to pick up on that. I mean, one of the things that we've reported on the information is we know that Meta is beefing up their research team. That's the approach that they've taken. We also wrote a story this week about how Apple, a lot of the power at Apple with respect to AI now sits with the product team, and that's been sort of a source of tension at that company.

15:50I mean, as it relates to research versus product, I mean, how do you think about that at Runway? Do you think about it yourself as a product-first company or research first? Yeah, it's a great question. I actually wrote about this a couple of days ago and just to go like, I'd be more in depth into my thoughts about this. But my summary and my realization over time is that these days there's no distinction between the two. I think it used to be the case that diffusion of technology was slower. You had research, R &D, innovation, and that took like years to diffuse itself to customers and everyone who would like access it.

16:30And within that transition period, you change the research, you improve the product, you get feedback. And it's a long, it was years to make. These days, for a combination of different factors, including just much better distribution networks, social media, AI being so valuable so quickly, the diffusion of technology from a core research model, something the R &D team has been working on, to a product is like that. It's instantaneous. It just literally is one line of code that you can just have. So I know you don't see any, you know, you sort of see them as the same, but, you know, don't researchers want to just focus on research?

17:08I mean, I think that's true. Researchers might want to just focus. Some researchers might want to focus just on research. And if you're a research organization who is not thinking of our product, that's fine. Like you could do that. And I think a part of the companies you were mentioning, perhaps, were too much focused on that and missed the train or understanding that you need to be able to deploy these things in real case scenarios. You need to be able to learn really quickly. If you don't, you're going to leave behind. And look, you can have great research papers. You can have new methods, all great.

17:36But at the end of the day, if you're focusing on building a business, you need to build stuff that people care about. And so my realization is the best, and this is comes back to some things like perhaps hiring and who you hire. But in my experience, the best people are the ones who are researchers who understand the product feedback loop. If you're just focused on research for the sake of research, that's great. but in this world and product world, that might actually backfire in the long term. Right. Last question before I let you go. There were reports last month that Meta was talking to you about a potential acquisition.

18:06Is that true? I can't really speak about that. He knows. All right. Well, thank you so much, Chris, for someone on the show. We really appreciate it. It really is fascinating to hear your perspective on research first product. And we'll have you on the show the next time we hear that runways used in a feature fill. I'm excited about that. Awesome. Thank you. That was Chris Valenzuela from Runway. Our next guest is one of the most data savvy venture capitalists that I've ever spoken to. Tomas Tungus is a general partner at Theory Ventures. The company's most recent fund is focused on investing in three categories, data, AI, and decentralization.

18:44He also has a great newsletter that goes out every night. And he wrote a great opinion piece for the information a little while ago. We will link it in the show notes. Tomas, thank you so much for being with us. We're so excited to have you here on TITV. Oh, thrilled to be here. Thanks for welcoming me on. I appreciate it. So, look, you're a numbers guy. We talk about numbers a lot together. I'm curious, what metric have you been fixated on in 2025 as it relates to assessing AI startups? Net dollar retention. Same as usual. Yeah, well, I think it's ultimately the core health of a business. And we're in this market where you see these astronomical growth rates of pretty phenomenal companies, and we have very little longitudinal data on ultimate retention.

19:34And ultimately, margin structure and net dollar retention are what make company growth, obviously, but what make companies valuable in the long term. And we don't have a lot of that data yet. Tell me a little bit about the ways that you see startups putting lipstick on their numbers. as it relates to revenue and ARR and annualized revenue. Walk us through what you see. Right. Well, I mean, I think the first thing that you see is you see transactional revenue or consumption revenue that's annualized. And Snowflake does this in the public markets. Twilio has done it for a long time. It's very reasonable to do at large scale, right?

20:10If you're a publicly traded company and you have four or five or six or seven years of usage data, you can create forward models with even basically linear regressions that get you to a place. You can be confident that you can annualize a number. But if you're a company that's nine months old, much harder to do. We still call it ARR. We accept it as an industry that it's annual recurring revenue, even though it's transactional usage based. The second dynamic that we see is 12-month contracts with three-month opt-outs. So I signed the information. I started an AI company. Gosh, I sell you some fancy piece of software.

20:42Right. And you have 90 days to opt out. startups are annualizing those first three months and saying this is our ARR without any conversion information or any conversion information at large scale and so there again you at if you're big and you have a lot of historical data it's fair to annualize a more conservative view might be it's it's a bit aggressive yeah and then the very last is monthly net dollar retention or monthly the expansion. We really need to look at those numbers on an annual basis. So all of these things are kind of borne out or the byproduct of just very high growth companies, growth rates that we haven't seen before, trying to fit metrics that worked in a previous generation.

21:27I'm curious, how much do you look at the AI companies now? You've got teams of just one, two, three dozen employees that are posting huge top-line profiles with very slim teams. I mean, And there is a question about how sustainable the company will be, whether they have headcount to support that. I mean, do you look at any numbers there with respect to like, this is still a small team, we got to be careful here? Yeah, I mean, I think ARR or annual recurring revenue per employee was a metric we've been tracking. I've been tracking for a long time and startups kind of start around like 50 to 60 ,000.

22:04And then publicly traded companies in the previous regime were around 150 to 250K, topping out like 400. And now we're starting to see startups with 1, 2, 3, 4, 5, 10 million of ARR per employee. It's pretty sensational. I will make a distinction here, which is the companies with the highest ratios tend to be product-led growth. They tend to be companies that are adopted by individuals, grow through word of mouth. There's not a significant, say, enterprise business yet. And I think as the companies grow, the importance of customer success, onboarding, solution architects, all those kinds of enterprise relationship management functions will become more important.

22:48I mean, to that end, you actually wrote a blog post about how Figma, coming up in the public market soon, they are an example from your view of product-led growth and just how sensational that could be. It's phenomenal. I mean, they have a sales efficiency of one. A sales efficiency is how many dollars of sales and marketing investment do I need to invest to produce$1 of gross profit in the subsequent period? And I think there's only one company that's ever come close, which is Shopify. And we can debate whether or not that's a pure software company because there's a huge amount of transactional revenue there.

23:17But it just tells you how efficient that business is and why I think it'll trade really nice and up. This is not an investment device, but I think it'll trade really well in the public markets because it's an incredibly efficient, high-growth business. You know, I was listening to a podcast that you did earlier this week. You were talking about building your firm as an investing corporation. And, you know, it sounds like an investment firm to me, but I think you have a very specific view on what an investing corporation is. So, I mean, walk us through your methodology there. What do you mean by that?

23:51Yeah, so venture capital started as a little industry with six or seven people around the lunch table on Tuesdays. And Akash, if you and I were in that seven, we'd get together Tuesdays and I'd say, I found this investment opportunity. I'm willing to invest 100 ,000. The round is a million. Everybody else would write a check. And that's how the venture capital industry started. And then that lunch group formalized itself into a partnership. And we've been running with that model for a very long time. I think now we're starting to see the evolution of venture firms where they are financial companies Right.

24:25They look like a Fidelity or a T-Row. They look like the hedge funds that are like a Bridgewater, Citadel, where they're corporations, right? There's a CEO. There's somebody who's running the firm every day. They think about the product that they sell. There's sales and marketing, all the classic departments. There's R &D. There's AI. And so that's a pretty fundamental shift, right? If I think about when I started in the industry, most of the firms, it was a group of investing partners and some associates and some principals and then a finance team and some support staff. And that's the way that if you look at the industry now, it's radically different.

25:02We have platforms. We have all these different functions. And so they end up looking like companies. We sell financial products at the end of the day. And we do it to two different markets. First, the startups and to our investors, our LPs. and that formalization parallels what's happened in the private equity industry and the hedge fund industry that's coming for venture. Right. I mean, I can certainly see how having all that support from a startup's perspective is great. You get a VC that invests in you, you've got all the support around you. Is there a downside to this shift at all? I don't think so.