Erik Bernhardsson, CEO of Modal: Modal powers cloud infrastructure for AI

0:00Scaling Europe:Hello, welcome back to the Scaling Europe show. I'm Seb Johnson. Today, we've got a banger. Erik, co-founder and CEO of Modal. How are you doing today? Good. It's great to be here. You're doing crazy things. The company is absolutely flying and I'm excited to get into some of those numbers. But before we do, can you maybe give a quick introduction to what you've built and I guess where you sit in the AI stack? Yeah, so I'm the CEO at a company called Modal. We build AI infrastructure. I think it's the best way to put it, although we sometimes do other things too. We power a lot of inference and sandboxes and training and batch jobs and other things for a lot of customers like Cognition, Suno, Runway, Decagon, DoorDash, and many others.

0:41The foundation is we built a platform, our own infrastructure platform. Fairly general purpose. You can use it for a wide range of things, but inference is one of the biggest use cases. We have a sandbox product, as I just mentioned. And yeah, I've been building it for about five years.

0:56Scaling Europe:And how's it going? I mean, you raised a not small round earlier this year. I think it was$355 million. I think that's the total amount. It's, yeah, it's going super well. I mean, you know, AI, it's growing like, you know, so much demand in the space. And I think we spent the last five years building a platform that, you know, there's just so much demand for helping all these new amazing AI applications, AI companies to build the applications of today and tomorrow. They all need infrastructure, and we help them with that. Yes, amazing. I think at the time you announced, was it$300 million of ARR, which was up 5x in about eight months or something?

1:39Something like that, yeah. And it's a lot higher now. It keeps growing very, very fast. So every month I keep thinking, it's got to slow down a little bit, right? And then it just keeps going. It's been incredible.

1:52Scaling Europe:And how do you forecast that level of growth? I'm sure this is true of so many AI companies today, but with the growth, it's unpredictably fast. Is that a challenge to running the business? It is challenging because we have to go out and procure GPUs a little bit ahead of time. It takes like three to four months to go out and do these commitments, sometimes six months. So it's a challenge, right? Like we, you know, straight line, you know, exponential forecast. We look three months out and project, you know, similar growth. And we pray that that will keep happening. Usually what ends up happening is it grows even more than that.

2:27So far, if anything, we've bought too little capacity. But we're trying to be very aggressive with capacity right now.

2:35Scaling Europe:And how do you really, when you talk about raising big rounds and giving yourself the money to buy these GPUs, how do you think about, I don't know, running the capital allocation? How much do you need to raise and when? To be clear, we don't own the GPUs. We rent GPUs, which is a little bit more capital efficient in the sense that we don't pay the full amount of the GPUs. We pay a little bit of money as a down payment. Often when we do these terms, sometimes nothing. And so that has enabled us to be quite capital efficient. We also work with 20 different cloud vendors, and that helps us get the best price out there.

3:09So we have capacity all over the world, different cloud vendors. I think we work with, you know, 100 different regions in total, maybe 150. Asia, Europe, US, some other places too. We got GPUs in every country almost. No, not quite, but a lot of countries. And so, yeah, we, you know, like we do need money. Like we're not a profitable company, but we've been actually very, very capital efficient. And that's something I think is still quite important for us. We're a software company after all.

3:37Scaling Europe:And where's that growth that you're seeing coming from in terms of both maybe region and type of user or customer? It's, I mean, it's primarily AI natives. So most of our customers are companies that have been started in the last five, seven years. I think, you know, generally that's true across, you know, all of these AI applications. There's just so much demand coming from these companies that fundamentally didn't exist before Gen AI. because like, you know, Gen AI basically like, you know, created a whole business model. So it's a lot of startups. And it's like kind of interesting because like I feel like the sort of conventional advice is always you got to go enterprise.

4:15That's where the money is. But like, we don't see that. Like the money is actually like AI natives. We're starting to see a lot of digital natives waking up and spending a lot of money on this, like kind of rethinking their business models. I do think, you know, give it five, 10 years, we're probably going to see the enterprise companies also ramping up their budgets. But most of the stuff they're doing in AI today is, frankly, more POC type stuff. In terms of regions, I mean, you know, I'm European, you're European. You know, sadly, most of the revenue today is coming from the US, but we are actually investing a lot in Europe.

4:49We started a sales team in Europe very recently, just in the last few weeks. We're going to announce that very soon. Very excited about that. We have an office in Sweden, too. In terms of sectors and use cases, it's all over the place. I would say a lot of generative media, audio, video, music. There's a lot of custom support type companies, a lot of coding models, of course. And then we actually do a fair amount of stuff in computational biotech, which is very exciting to me. So we have a bunch of customers using a model for computational biotech, molecular simulations, and all kinds of cool stuff.

5:18Scaling Europe:It's interesting that you mentioned that most people say to go after enterprises, but it feels like in this age of Gen AI, companies are scaling to enterprise size in very little time. It used to be go after enterprises because they're the ones with billions of dollars. revenue now you've got companies you know that are doing billions of revenue in no time at all yeah yeah and so that kind of begs the question what is even enterprise like i don't know but like i think of it as like later you know later stage companies like again you know sap and bank of america like probably not like our core audience right now but like might be in 10 years do you think that says something more broadly about whether the value of ai is going to really be captured?

5:58I mean, I think there's like two things going on. Like, first of all, there's like AI is like a very new technology. And, you know, it's kind of clear that like startups will be better at like, you know, creating value out of that technology. And so I don't think it's super surprising that you see, you know, all these coding model companies or music generation companies like these are fundamentally like completely new startups, right? Like predicated on AI is like a fundamental business model, just like, you know, 10 years ago, you saw people, companies predicated on the iPhone and the cloud and that was the core thing.

6:28So I think there's a lot of that. The other thing is I think the vendor stack is getting deeper too. We're building infrastructure. So in many cases, we actually have enterprise revenue, but indirectly because we sell to companies that then sell to enterprises.

6:43Scaling Europe:Yeah, that's really interesting. So the companies themselves are not AI native. They're buying, they're paying for the services and the products that have been built by the AI native people who are then being... Exactly. yeah that's what i mean the stack of vendors is kind of getting deeper in a certain way so we were pretty far down in the layers yeah interesting okay i mean can you can you talk about some of the use cases and what you're seeing because i imagine where you sit it's not just the location the type of companies but also some of the uses yeah i mean some of our early customers uh was typically gen ai we initially saw product market fit in gen ai around stable diffusion type use cases so image generation video generation music Suno is an early customer amazing company they do AI generated music very cool that was initially where we saw a lot of product market fit over time like I mentioned we've launched a sandbox product seen a lot of success around that so sandbox is basically means like there's a lot of LLMs generating code and you know you want to execute that code but you don't necessarily trust the LLMs you don't want to execute it on your computer that can be dangerous so we have a sandbox primitive that basically like encapsulates that code and lets you execute it in a safe way.

7:54And that turns out to be an extremely useful primitive for all the coding model companies. We work with companies like Cognition, but also a lot of Neo Labs, building coding models, background agents, work with some Vibe coding platforms. Seen a lot of success with that. We also do LM inference, of course, a very large, very important space with the Chinese models and the open source models. There's just so much demand, especially now that they're becoming very, very good models. You see a lot of companies kind of switching away from the frontier labs to open source models, maybe fine tuning, maybe using reinforcement learning.

8:28So that is a massive space for us. And then there's kind of a long tail of stuff. Like I mentioned, computational biotech is like a really interesting space. We work with a lot of robotics companies, sometimes controlling robotic arms, work with some companies using modal for weather simulation. So at its core, I think modal is good for anything that's like computationally intensive, almost like HPC in a certain way. And we have a lot of the primitives and you can build a lot of different things. But the sort of commonality is like, you need a lot of compute, typically a lot of GPUs.

8:58Scaling Europe:The point you make around the Chinese open models is really interesting because it's something that we're seeing a lot in like the social media and people are talking about it a lot. So it's interesting that you're seeing that. How significant is that switch from companies using the frontier models to more Chinese or open source? Super significant. And I think it's like, the sort of rationale has shifted. Like if you went back like a year ago, it wasn't necessarily that the open source models were as good. But the argument was like, well, if you fine tune them for a specific domain, they can be almost as good and a lot cheaper.

9:32So we saw a lot of demand for fine tuning, like post trading and reinforcement learnings of those types of things. The crazy thing is like in the last few months, like the models are basically as good, like, you know, without any sort of fine tuning. And so now you're seeing a lot of companies just wholesale shifting to open source models because they're cheaper. And so in a weird way, actually, like post-training, I think it's been a little less sort of the topic of the day. I think it's going to come back. I mean, fundamentally, I think I believe in post-training, but it's something, you know, relatively sophisticated customers do.

10:09And there's just so much demand for, you know, Kimi K3 and GLM 5.3 and Quen 3.8 and all these models. like these models are very good. They're Chinese. So there's a geopolitical, maybe scary angle. But I also expect, you know, there's a lot of US labs working on new generation models. I think they're going to be almost as good, if not better than Chinese pretty soon. And, you know, hopefully Europe also has some models. I believe in the Western world. So I hope we can, you know, run open source models that, you know, may not have, you know, biases or who knows at some point.

10:43Scaling Europe:Do you see it really fundamentally presenting a risk to the Frontier Labs and their revenue growth? Because we're seeing this switch to open source Chinese models at the same time as we're seeing OpenAI and Anthropocene's revenue skyrocket. We've seen both things happen simultaneously, it seems. Do you think it is a very serious risk to the longevity of the Frontier Labs? Or do you think both will win? i i think a little bit but i mean you know it's such a big pie and the pie is growing so fast like i i feel like you know it's easy to you know look at like sort of you know kind of you know fixed you know pie mentality like you know what we call a zero-sum you know fight but but but like in reality like you you projected 10 years down i'm sure there's going to be way more consumption of open source models and way more consumptions of proprietary models and way more like on device stuff and way more like cloud stuff and way more all kinds of other stuff so i don't know i don't think we should necessarily frame it as like one versus the other but but of course if i was you know sam or dario like i would look at these models and say like actually like they're pretty cheap and quite good like what are they doing that we're not doing and you know you know some people claim you know distillation and all these things and maybe there's some truth to it maybe not but like these labs are also very good like they have very good people And it is quite remarkable that they've been able to build these models with a fraction of the research budget.

12:09Scaling Europe:Yeah, I completely agree. In 10 years' time, the market size is going to be 50 to 100x, if not more. Can we talk on the, you mentioned the kind of the European expansion, right? You're opening a new office here in London. You've got engineers out of Stockholm already. Why now? Yeah, and I'm actually in the Stockholm office right now. Yeah, we actually, I'm Swedish originally, but betrayed my home country and left to the US 15 years ago. But, you know, I come back every summer. It's an amazing country. And I have a lot of connections here. So when we started hiring people, we hired a number of people in Stockholm.

12:48And so far, that's been our only European presence. We have about 17 people here. By the end of the year, we might be 30 people here. So Stockholm has always been for us, like a very core engineering hub. A lot of very strategic projects. It's our file system initially. Now a lot of sandboxes have been driven out of here. Stockholm is very good for technology. I think for sales, like what we found is like it's a little bit harder. And so also Stockholm is not as big of a market. I mean, there's some local companies, but London is a much bigger market in Europe and has a little bit more of a history of technical sales.

13:19So now we're building up a sales office primarily. We're going to have some engineers there too over time. but we have already I think seven people in London up from zero in the last couple of months so very bullish on London I think there's a lot of incredible local businesses some of them are already customers and I also think London is an exceptional base for sales into the rest of Europe and you know I think Europe long term has the potential to be 20 25 30 percent of our revenue Europe is a fantastic continent there's so many different you know innovators fantastic tech talent i'm a big believer in europe so i think it's important for us uh in fact i actually think of it a little bit as like a first mover advantage like we as a company at modal given that we have a little bit of european roots i think we have an opportunity to go very aggressively into europe and and you know take this continent and build something very large uh so i'm i'm very excited

14:12Scaling Europe:about that and how you think about the overall go-to-market motion in europe you know what has worked so well in america do you think it's going to be do you think you'll be able to translate that and use that same motion in Europe? Probably not. I mean, you should ask me in a year and I see. I have certain assumptions, right? Like, you know, of course it's a different market. I have a feeling like, you know, what works in the US probably works like okay in the Nordics. I actually don't think sales is very different. Like they speak English, you know, and maybe the UK too. There's obviously a little bit more sort of focus on compliance and, you know, zero data retention and gdpr and types of stuff like that but actually surprisingly little uh and i think probably that the reason is we're focusing so much on ai natives anyway they tend to be not so compliance heavy uh but i think you know when you start to look at like germany and and france and these countries like my feeling is like and this is an assumption to be clear i don't know yet but like you probably want like people you know native speakers and you probably need a lot more like sort of focus on the compliance and you know you know you know adapt to how to sell to local markets we'll sort of see so i actually think you know it's a little bit different norweg and the uk i think are more american markets is kind of my feeling germany and france maybe a little less you have to have like kind of more of a unique approach there we'll see yeah they get more more traditionally like continental i guess and how you think about talent you know like who do you who are you looking for to recruit into the london office you've already got some people there now but like what's the characteristics of traits that you look for?

15:42Yeah. Well, you know, you know, Brits, like they have this like extroverted sometimes, like they're very gregarious. You know, I think that's a trait that does well with sales. They have an amazing accent. You know, I think all those traits are the things that we look forward in our sales team. We've actually found a lot of really good salespeople already for some great companies. So, you know, amazing talent so far. then of course on the technology side I think we're going to hire a lot of people in London too I think there's amazing AI talent I think London is really the AI capital of Europe today and so over time I actually hope that we can build up also local engineering talent too I think Sweden has a lot of deep knowledge in systems infrastructure low level like Linux stuff a little bit less on the sort of AI side and I think London is like that's where it really shines Like there's so much AI talent in London versus the other European capitals or the other European cities, I mean.

16:39Scaling Europe:Yeah, absolutely. And we're seeing here in London, like, you know, Anthropic, OpenAI, Curse. And we're seeing a lot of companies, you know, who are trying to get into something. Yeah, it's fine, of course. What is it, Axomorphic? And what's the other? Ineffable Intelligent? Like there's a bunch of them, right? Yeah, yeah. A lot of Neolabs. Absolutely. And, you know, you're a European. You're obviously passionate about Europe. but obviously building in Europe and now building in America, you've had this, I guess, like, opportunity to look at both, right? And what are your views on European tech, right?

17:11Scaling Europe:As a European who's built this amazing company in America, now expanding to Europe, what are we getting right? And what are we getting wrong? I mean, my view is that, like, I hate everything that's, like, building in European X, like, aimed globally. Like, so that to me is like, you know, one sort of pet peeve I have is like, I really think a European company should build for the entire market. I was at Spotify for seven years. And, you know, I think that was so key to Spotify's success is, you know, we started in Europe. And there was some benefits of starting in Europe at that time. Like, we got the record deals we needed because records labels didn't really care about Europe at that time.

17:49But from day one, and Daniel Ek was like very aggressive about this. He's like, we have to go to U.S. we're going to be a global company. US is the biggest market. We have to go there. And, you know, I think that was so core to Spotify's, you know, identity and all these things to have that sort of very large, you know, global approach. And, you know, I think that's something I see in some companies like Ligore and Lovable. There are two companies here in Sweden that almost have this like megalomania that I think Americans often have. You go to the Valley. I spend a lot of time in the Valley and like everyone's like a little bit like like drinking the Kool-Aid and, you know, sometimes a little bit too much.

18:27You know, it drives me nuts sometimes. But you kind of have to have a little bit like that megalomania. And I see that in the sort of next generation of these younger founders. Like I mentioned, the Love Upon the Core. They're a little crazy, but I like this very big, ambitious, like, you know, we're going to build a global company. Like, there's no doubt about it. And that's something I want to see a little bit more in Europe.

18:50Scaling Europe:So we need to think a bit bigger. And I think you're right, especially in the UK, we've suffered from the UK being a big enough market to build a great business in. That if you just live in the UK, you can actually build it. You know, 10 years ago, we were building a lot of great unicorns that were very, almost exclusively focused on the UK. I think that's right. And I think like Stockholm has the advantage of actually, you almost have to go global from day one, because it's quite a small local market. And it's one of the interesting things about London has got great capital flowing through it, great talent, but not that many global champions, because we've been too tempted to focus local.

19:23That's absolutely right. I think you see the same thing in Germany and France. There's a lot of very successful companies that were just successful in Germany and France. And I think you're exactly right. Sweden is not a big enough market. It's like why Swedish music export big. It's kind of the same thing. We just started singing in English because we made a lot more money doing that.

19:44Scaling Europe:You've got to look across the borders, yeah. And to tell me, what was it like working at Spotify? right like you any lessons that you took from your time there that you're now applying you know as you as you build modal yeah spotify was a fantastic company and of course you know like 20 years later whatever it is i started in 2008 so almost 20 years 18 years later a lot of things worked out now it's public company it's in i don't know how many hundred companies 100 countries i i think a lot of things that spotify did right in their early day was like having that megalomania having a big vision we hired the best people we hired some incredibly good people and i think early early spotify had this like very sort of you know strong almost like anarchy internally i mean like in a kind of an extreme thing is like when i joined like no one told me who my manager was which i you know i said i'm like you know i was my first job out of school like i didn't think that was like i just thought that was just normal but like i didn't have a manager for the first year but it didn't matter because like everyone was just like working on cool stuff and like i was just like okay i guess i'll like build a music record initial system and i built that and and so there was this like incredible energy where like a lot of things got done with like and through just like a like self-organization there's like a very like self organizing system in the very early days and that's something i've taken with me is like you actually don't need that much like top-down this top-down planning and strategy and stuff like that like just hire smart people and tell them this is a very large market we're going to build something very large like now you go and figure out how to do it and it sort of worked at spotify that's really interesting is this yeah you kind of mentioned anarchy it's like that lack of structure to like let builders build yeah totally and and i think it worked really well and i uh i try to operate modal in a similar way give people an uncomfortable amount of freedom uh you know and And the right people will step up and enjoy that and really thrive.

21:43Scaling Europe:And you've scaled at quite a remarkable pace. Part of this generation of companies that is hitting milestones faster than the generation before ever did. What's that like in your role? How do you have to manage the internal mechanisms of the business as you scale through at this remarkable pace? Yeah, it's challenging. I think what Alpsom is like I always built, you know, I always been at like very fast growing companies. So I think I'm a little bit used to the chaos. I think you always have to accept a little bit of like chaos and waste and you have to like just accept, you know, live with some of it.

22:19What you can't live with is, you know, people trying to be political or, you know, like dumb, like adversarial stuff or gatekeeping mentality. Like you need to know what's like tolerable waste and anarchy, whereas like what's like not acceptable. So there's certain like non-negotiables like that. I think as a CEO, like, you know, I mean, this goes for every CEO in general, but like, I think you always have to figure out a way to get out of your job, like every six months. Anything I do today, with the exception of maybe doing podcasts, I need to figure out a way to delegate to someone else in the next six months, right?

22:53like you know i i comment on so many notion docs and i'm in so many meetings about you know planning and capacity and finance and conferences and off sites and whatever and all of that stuff my goal is always like in six months i gotta get out of this stuff because there's gonna be something else i need to find people to do that so you know i don't think i've learned so go ahead

23:12Scaling Europe:is there anything that you think you'll always keep holding you know there's certain things like fundraising right that generally stays within the c agreement to an extent talent some ceos very laser focus on talent for you obviously you've got this huge question mark always over capacity and you've got to be thinking strategically a few months down the line about that is there anything like that that you think well firstly that dominates most of your day-to-day and that actually you don't think you're going to be able to outsource well i i mentioned one thing it's doing podcasts right it's hard uh so so that's one thing and and i think you know in this new age of you know building companies i think what i call founder-led marketing i think that's always there's always an element of that that's like hard i mean you are a figurehead and i think fundraising too like you can delegate a lot of behind the scenes yes you know i used to be the one populating the data room and doing all this stuff but i don't have to do that anymore thankfully um you know but but other than that like everything else i think you can get out of of course like you know recruiting is something that i'm always going to have to do but but the recruiting shifts like i stopped interviewing every single person at this point pretty recently though but you know i still spend a lot of time we have a couple of exec roles that we haven't filled so i'm spending a lot of time trying to find a finance leader right now and you know so recruiting you know it doesn't go away you're always gonna have to do recruiting has there been one thing that you've been so happy to delegate i mean everything you know technology finance i recently hired a general counsel and it's like so nice like all the legal garbage like just having a person that can deal with it because i kept getting sucked in you know rsus versus options like what should we do and like the problem with me is like you know i'll go to chat gpt and like three days later i know everything about every fucking like tax law because like i just like go into like rabbit holes because i i know you know so now i know a lot about the stuff but like it's also something i kind of have to let go and like hire people and you know i go a lot ask a lot of questions instead and i think that's like the right model yeah okay so then you just let other people manage it yeah you become the strategic prober or something you know make sure that people i think so and i call it sometimes like management by random sampling it's like how do you how do you create the right like org structure with the right incentive structure and i think the right one of the things i've sort of arrived to it's like you're never going to be able to check everything but like you know you can go around and kind of like ask a question about some random button and you know not like you know silly things but like i'm a very curious person so i go around and ask a lot of random questions and you know you kind of very quickly sniff out if people have they have a thoughtful answer there's a certain level of like you know they thought through it and they understand why versus like you know i don't know like both like some and i think that's way to kind of know who you can trust yeah that's interesting is there any key learnings you've made and now you've built this large org is there anything that you've learned that it's like a really clear telltale sign that someone's going to have you know a lot of autonomy and kind of get shit done anything that you've learned is like okay this is a way of identifying a top top person i yeah i don't know it's a hard one i i i actually don't have a good answer i i know it when i see it you know it's like if it's a designer like you know i i see it in the yeah you know like and everything else like you know let's talk about design for I care a lot about design and I think a good designer, like, first of all, this stuff looks good.

26:43But also, like, it's like I ask these questions, right? Like, and I can tell, like, there's a level of, like, meaning to everything they do. Like, intentionality. Like, there's intentionality to everything. And there's, like, a level of consistency and a level of, like, ability to explain how they were thinking about it. and that's like how i know you know that okay this is like a designer that did good job you know and i think there's stuff like that for like almost everything right like certain telltale

27:12Scaling Europe:signs yeah interesting yeah okay we've got a few minutes left i want to ask you some final questions you've just raised this big round right how are you thinking about uh deploying it are there milestones that you've got in mind that you need to achieve before you take it to the next stage I think one of the super exciting things about modal is like we always had this vision of being a very general purpose company and a platform and a toolbox of a lot of different tools. And that wasn't necessarily true to the outside world. It wasn't really clear until last year when we now, when sandboxes started really ripping.

27:47And now we have two products out there in the market. Sandbox is about a third of our revenue and the serverless function, which is practically in practice inference. It's about one third. Sorry, sandbox is one third and inference is about two thirds. So evolving into this multi-product company, I think to me, that's like the next evolution of the company. We're working on a training product that we hope to launch very soon. We have a lot of ideas of the next stuff. So if we catch up in a year, I hope that there's not just two products, but three or four or five products. And that takes, of course, people.