Modal raises at $15.75B valuation with Cognition, Ramp, Suno, Runway, and DoorDash as customers

Speaker 3: Have a good one.

Speaker 7: Great to

Speaker 3: see you. Bye bye. Up next, we have Eric Modell, the co founder and CEO. I've got a... I'm glad the Gong's warmed up. You know, I warmed up the Gong earlier and I'm glad it's warmed up

Speaker 2: because he's ready for the...

Speaker 3: Here he is. Welcome back. How are doing?

Speaker 2: Good. How are you? The man. Fantastic. Legend.

Speaker 3: Tell us what happened.

Speaker 2: Is this your third is Third time.

Speaker 3: Your third time on the show?

Speaker 2: Third time.

Speaker 7: Was it

Speaker 2: the Did we Repeat. Did we get the... Did you do the series a with us or was it the series b? Was that the first one? Trying to remember.

Speaker 7: B c and d, I think.

Speaker 2: I forgot c and d. I'm I'm upset we didn't get the a. I would've liked to start earlier, but you've been on absolute tear. And John really wants

Speaker 3: to How get this much did you raise? Go

Speaker 7: ahead. We raised a

Speaker 2: bunch of money. Boom.

Speaker 3: A bunch of $750. $15,750,000,000 valuation. Congratulations.

Speaker 2: Amazing progress.

Speaker 3: What's working other than...

Speaker 7: We see so much demand from all these incredible AI applications out there, you know, Cognition, Ramp, Suno, Runway, Decagon, DoorDash, and Mhmm. There's so many companies out there building incredible stuff running on model because they like the building blocks that we give them, inference and training and sandboxes and and all kinds of other stuff.

Speaker 3: Yeah. What what... What's then been your... Other than capital, like, what's been your binding constraint? Is it power, chips? Like, where

Speaker 7: is We your run data centers. Run capacity today. It might change in the future. But, yeah, capacity has certainly been constrained where a lot of the capital we're going to use towards ramping up capacity a lot to serving all the demand out

Speaker 3: And so what's been the secret between balancing? You're basically load balancing between all the neo clouds, bringing your software to bear, your techniques, and then also balancing against all of your customers. But what has been the secret to actually being able to reliably deploy across different clouds? I mean, follow the semi analysis cluster max 3.0. It feels like there are some differences between these different platforms and these different neo clouds. What's your experience been like?

Speaker 7: Yeah. There's a lot and we have our own internal modal max rating. I mean, in general, like, we aggregate capacity from 20, I think maybe 25 clouds today and Yeah. It's painful, right, and they're all somewhat different. But I think that's also the value prop of using modal for our customers is like you don't have to deal with that nonsense. And then we add like a pretty rich software layer on top of it to to write a few lines of Python and deploy incredible applications and run sandboxes in a way we don't have to think about all this Yeah. Cloud idiosyncrasies.

Speaker 3: What's the state of load balancing across time and geolocation? Is that becoming a priority? I remember hearing early on that mid journey had some inference capacity in Asia because there were people that were generating mid journey images in America during daytime. As we get into more agentic workflows, I'm thinking, oh, this is something that ran in the background that I wouldn't mind if it ran at 3AM local time. Totally.

Speaker 7: So... This is something we think a lot about. As we aggregate and and run larger and larger percentage of the world's compute, still pretty small, but hopefully growing over Those those things become like really optimization problem. Can we run things overnight? Can we run things on weekend? Can we like arb the geo stuff? Run things in Asia when they're sleeping and stuff like that. There's a lot of opportunities to do that. And we can do that transparently because we aggregate all that capacity with Cheetah as one big pool. Sure. Our customers don't have to think about

Speaker 3: Yeah.

Speaker 2: Yeah. Sorry. How how are you thinking about diffusion timelines? Like, I imagine your customers, like the ramps and the and the DoorDashes of the world are doing things that, like, the Fortune 500 might be getting pitched these ideas from various like AI, like consulting companies and things Yeah. Like that. But they're probably not like at the frontier of whatever that new capability is. So how are you how are... Yeah. How how does like what is what is the current state of diffusion from the sort of like scale up, you know, sort of tech tech companies Yeah. That still have a bunch of like really AI pill talent versus the rest of the world where they have smart talented people but they're not necessarily like watching... Listening to Dwarf Cash or or watching TBPN. TBPN or whatever it is. Right? They're just like Yeah. They don't really care about AI. They're like, I'll I'll adopt it if if that's what we're gonna do, but I'm not spending time on the weekend.

Speaker 7: Yeah. I mean, I think this is like a massive shift. We basically opened up a whole new set of capabilities for software. It's... You know, I've been programming since early nineties, seen a lot of these things in the past, you know, whether it's cloud or big data, mobile or a lot of these shifts in the past. And it takes time to transpire and, like, you know, to reach the enterprise customers. And I think we're still very early, very early in that adoption. But, you know, we're starting to see... Initially, we had a lot of AI natives adopting this, you know, all the companies I mentioned kind of building things from scratch. Starting to see slightly more mature companies. I mentioned DoorDash as one of our customers. And over time, I think it's gonna be, you know, Deutsche Bank or whatever. Like, eventually Yeah. We'll get there.

Speaker 2: Yeah. But but it's... But it doesn't feel like there's still like a large gap between this sort of AI pilled scale ups.

Speaker 7: Massive. Massive. Massive. And and I think a lot of it is like it's harder to work with these things. And so that's a big problem that we're obviously very focused on. It's like, with this new technology, you need a new set of tools. Like, you need to kind of build the right building blocks and, you know, what are those? I mean, we think they're, you know, the things I mentioned like inference and training and sandboxes, reinforcement learning and post training and a bunch of other stuff. Like, kind of packaging that up and making it easier to take things from research to to production and scale it up. Like, that's really what we enable a lot of these companies to move fast and get value into production.

Speaker 2: Yeah. And then

Speaker 7: we don't have to think infrastructure.

Speaker 2: Yeah. But because we're so early, it's like you guys can only do so much and then you're sort of relying on your own customers' creativity to, like, fully be able to apply this set of tools Exactly. Their

Speaker 7: Yeah. And that's the beauty of software. It's all about creativity. Like, we give them the building blocks, They can kind of combine them in an infinite way and Mhmm. You know, that they can build software.

Speaker 3: How much demand are you seeing or how many CEOs are talking to you about the potential of bringing an ultra fast product to their experience? I'm thinking about Suno. If I install that app, they got me to like get over the hump to do the install. That's important. And if they were able to inference the first song I generate, I mean it's already really, really fast. But if it was like milliseconds, that might get me over like the conversion hump a little bit faster because I don't even close the app while it's queuing for five seconds. And you can imagine developers wanting an ultrafast product, something on Cerebras or Gocker. Like there's whole bunch of chips that are out there. Yeah. And the labs are... Have their strategy, but you have the opportunity to sort of bring that to to everyone. And I'm wondering how much demand you're hearing from that and if you're planning on solving it.

Speaker 7: So I I I think the beauty of Monol is is that it's really a general purpose product and you can build almost anything on it, and that includes things that that need to run very fast. Mean, we're very focused on real time voice, real time audio, things like that that need streaming. We're also looking into alternative chips, like maybe Cerebras, like things like that. Like Yeah. It's still kind of early, like, we'll see what happens, but I think there's some really exciting stuff. And I think that's like the sort of general purpose nature of Modal is that we can cater to all these things, like batch jobs on one side of the latency spectrum, really low latency inference on the other side of the spectrum. We're spending a lot of time optimizing LM inference as well and getting some really good performance. So there's a lot of that for sure.

Speaker 2: Yeah. That's exciting. When do you feel like inference platforms are going to be are they starting to diverge like, you know, the other companies in your space? Are they starting are are we in a period of convergence or divergence in the sense of like, Claude Code and Codex have been sort of like looking more and more like each other in in many ways, and then there's a bunch of other applications that end up looking more and more like each other. But I can imagine, like, I can imagine there's... There there becomes more separation just because there's so there's so We much to

Speaker 7: we certainly think that we're very unique in the sense that we're more general purpose than than pretty much all the other players in the space. I I do think just because the market turned out to be, like, 10 x or a 100 x or a thousand x larger than anyone thought, including myself, by the way, I think by just virtue of the market being large, like, we're... I think we're gonna see a lot of specialization. The market's gonna fragment a lot of stuff. And and and I think a lot... Our role there is to to cater to a lot of those different needs, like, bring a lot of the products. Like, Inference is itself, just vanilla straight down in the middle, you just want Kimi k three. I I think it's gonna be a very competitive space. But if you do all the stuff around it, you know, sandboxes, agentic stuff, like reinforcement learning, I think you can build a really really strong generational company. Something that can grow and, you know, grow and grow and grow for many many years.

Speaker 3: Yeah. No. I mean, we're seeing it every day. Yeah. It's awesome.

Speaker 2: Very cool. Well, congratulations on the milestone.

Speaker 3: Thank you so much.

Speaker 5: Thank you

Speaker 4: so much. Looking forward

Speaker 2: to the next one.

Speaker 3: Hanging with us. We'll talk to you soon. Thanks.

Speaker 2: Cheers, Eric. Goodbye.

Speaker 3: Up next, we have Josh from The Atlantic, a writer who talks about... He wrote the November cover arguing that Elon Musk and Jeff Bezos are racing to vertically integrate the space economy. So we'll get to the bottom of it. Who you got? Musk?