Stephen Balaban on 14 Years of Lambda and the Future of AI

There's so much disinformation out there with regards to what a data center is, what it isn't. There's tons of evidence that there's been nation-state actors that have spread disinformation in social media. There's no question about it.

I think there's the water thing, and I'm like, you know, why are they guzzling water? It doesn't make any sense.

No. And the water thing doesn't make any sense. It makes no sense because These systems are literally closed-loop dry coolers. It's no different than your AC. Your AC has refrigerant or a liquid that runs through, and it doesn't leave. Loudoun County, the county in Virginia which has the most data centers in the United States, has a lower average dollar per kilowatt hour than— it's a slightly decreased one than most of the rest of the United States. And so these data centers have been bringing power, one, two, bringing tax revenue. I think Loudoun County is like a billion dollars or more. There's just a lot of misinformation about it.

Hey everyone, welcome to the Founders in Arms podcast with me, Imad Akhund, co-founder and CEO of Mercury.

And I'm Raj Suri, co-founder of Lehman Tribe. And today we have with us Stephen Balaban, uh, co-founder and CTO of Lambda. Stephen, I haven't seen you in so long. Uh, welcome to the podcast.

Thank you so much for having me. It's, it's a, it's an absolute pleasure to be here. It's good to see you again.

You haven't aged a bit. You know, you look like maybe younger than, like, I probably saw you a decade ago. You haven't aged at all.

Yeah. And I was looking up where we had first met and it was at the Hacky Happy Hack block party in Palo Alto, um, in 2012. And, you know, it was a, like, sort of a golden age of startups in Palo Alto.

Oh yeah, absolutely. And we're trying to bring that back now, you know, with the, with the, I'm trying to do this like South Bay Renaissance thing of like bringing back startups to the South Bay. And so, yeah, I remember those days.

University Cafe.

Yeah, yeah, that exactly. And I probably tried to recruit you 'cause I used to go to hackathons to recruit people at that time. And, uh, yeah, yeah. And so, uh, but, but I mean, I remember you back then as like, you know, a, You know, a guy who just had like a dream and like the dream seemed like completely impossible to me. It was like not going to happen. And but I was like really impressed that you just had this, this like dream of being a founder and like, you know, because I've seen so many people fail at it as well, right? You know, I've seen so many people who just never, who always wanted to do it and never did it. But you actually were like super optimistic. I remember that was one thing that really stuck out to me back then. And that's one reason I wanted to always stay in touch is like, you know, those people inspire me as well, right? The people who are just always optimistic, always have a positive view, no matter how much they get kicked in the face, they're like somehow still believe. Um, so tell, tell me, tell us about your stories from like 2012 Palo Alto. How old were you then?

Oh, I was, I was probably about, I was 23, uh, or just turning 23 at that point in time. Um, And I had really just started working on Lambda. I had started a company previously called Babo Labs that was sort of an ebook marketplace for sort of digital goods, kind of actually kind of competitive with Gumroad. And so I actually knew Sahil and the Stripe guys because just, you know, overlapping industries. And yeah, I, I, um, you know, I, I was really into at the time sort of face recognition and traditional computer vision and AI. And, um, at the time I think I was building our very first product, which was called, um, it, it was called Heads Up, and it was a facial recognition-powered contact book for iOS. So it was like, you know, almost, you know, very timely or otherwise like kind of, you know, trying to shove an AI thing into the— how startups— you thought startups were supposed to be around, which is like a mobile-first company at the time.

Yeah, I mean, it was one of the first like gen— even not maybe generative, but you created these amazing like, uh, you know, um, um, images, like almost like AI skins today. Um, how did you make those? And, uh, were you trying to build a consumer company?

Yeah, well, so the sort of like the chain of how I even got to— so that part, the image, you know, using neural networks for generating images, that product was called DreamScope. And the way that we got there, so that came out in around 2015, and that was with the Google DeepDream sort of paper. And the way that I kind of got there was that starting on this facial recognition pathway. Um, the ImageNet paper came out in 2012, um, and I started training neural networks on some NVIDIA GPUs. And then eventually this sort of new way of generating images instead of just recognizing them, um, you know, be kind of became possible. And we ended up building this— it, it was, it was a consumer, you know, kind of AI filter application is kind of how we would discuss it with investors. And that's, that's how, that's how we really got some exposure to having to ramp up the compute. And so the idea was actually kind of, okay, well, instead of having to handcraft image filters, which is like the way that Instagram worked or something using traditional, you know, image processing, computer vision, that maybe you could create filters using neural networks and Um, that is how we kind of got started in, in, in seeing, um, you know, actually some initial traction for the company.

So how many pivots was it before you got to kind of where you are today? Because it seems pretty different.

Like legitimately, I think if you were to enumerate them, the company started off, I really wanted to do something in the augmented reality space and that's how I got into face— facial recognition. I really was interested in, you know, I'd seen a couple of animes like Dennou Coil and a few different ones that talk about augmented reality as being kind of this future. And, you know, I realized that sort of the, the image recognition part of augmented reality, for it to work, to do sort of both 3D mapping of a world as well as recognize objects within it, that that was like an unsolved problem. So I got into face recognition. So that's like pivot number 1, made this face recognition contact book that didn't really work out very well. There wasn't a lot of demand for this because it's not a very good idea. And then Face.com got acquired by Facebook. And so I said, hey, well, let's try to throw up a face recognition API. So that was the 4th pivot. So from augmented reality to face recognition contact book to, to, to Um, um, so that's pivot 1, 2, 3.

This is insane.

Then DreamScope was created, which was, um, which was the, um, the, the, the sort of image filter thing. And then we ended up building a lot of compute for DreamScope, and that didn't work out. Um, you know, the DreamScope didn't work out, but so we started selling the workstations and servers, I guess maybe pivot 4. And then that grew, and then we introduced eventually our cloud product, and that's pivot 5. So probably about 5 pivots to get to something that started to make money.

And this is over 14 years, right? When was the last pivot, like when you started doing cloud compute?

The last one introducing— so really what happened is the academic community started to really adopt you know, training neural networks, whether it's ConvNets or early language models. Sometime around 2017, that started becoming very widespread within the academic community, and thus a market opened up to provide essentially scientific compute for researchers and engineers. And so 2017 is when we started to like actually have a business that made money. So, you know, we did $3 million of workstation sales that first year in 2017, and that's kind of when things started to How did you survive long enough, 5 years, you know, to make it to that? So, so interestingly enough, going back to kind of the old school way of thinking, you know, uh, of startups, uh, Joel Spolsky taught— spoke at, um, like a YC startup school or YC Startup Day. And I went up to him afterwards and said, hey, Like, tell me about like consulting versus building a product company because, you know, what was it, you know, FogBugs or FrogBugs or FogBugs was one of their first products. But, you know, they had a software consultancy. And he said to me, he said, you know, look, you look around this room and it was an auditorium that was packed with people. He said, maybe one person or no people will ever IPO a company. in this room. And so it's important to just like stay alive and, you know, build some sort of revenue stream to keep the company going. And so I took that to heart. And I always just did a lot of consulting on the side. So I would do iOS app development, computer vision and AI consulting work.

Like you as an individual, or like you had like—

As an individual.

Okay.

Wow.

As an individual. And I would sometimes get like a friend of mine to kind of work with me in the early days, and I would kind of do a consulting project for 2 months out of the year, save up enough money, and then just like, you know, really like kind of live on ramen, if you will, you know, and just have very low cost of living. And we even did that in between DreamScope and the workstation business. We ended up doing a project with Airbus where we were doing some computer vision consulting for them for their— for drone detection for one of their flying vehicles, this Vahana project. And so it's always been a really important part of the company. And we did that type of consulting even through 2020. I know it's crazy as it sounds, but we actually did some consulting with the creators of South Park who were doing some early AI face swapping, um, uh, videos and stuff. Um, and so that's always been a part of the DNA of the company, is just like survive.

But you also like learning about the market during these— like, this wasn't like unrelated consulting stuff. Like, this was like in the market, and you were kind of learning about like how companies were using this kind of like software and things like that.

Yeah, yeah. I mean, it, it, it wasn't just— it wasn't always random. It was, it was oftentimes like very much sort of I developed a particular skill set that allowed me to charge a nice rate because it was a specialized computer vision and image recognition skill set. And it was always sort of like seeing, hey, well, what is the state of the art? What are people doing? Where's their money in the R&D side of things? And sort of eating your own dog food.

So, I mean, you could have got a job at Google or Facebook or I guess any of these companies around you and got paid way more money. What kind of kept you going for 5 years? or even longer, it seems like, before you like really had product-market fit.

I remember really distinctly starting Lambda and thinking, okay, I've always wanted to build a really big company. I've always wanted to sort of build a company that was well-known and was financially successful. And I've— I even like wrote down, Before this, I actually went through some of my old notebooks from 2012, and I had like long-term goals. And written down in one of these notebooks I've got here from 2012 said, raise $1 million for a company you found, create disruptive technology that changes the world, and IPO, right? So that was like part of my goal. But the way that I kind of thought about it is you can't IPO a company, uh, if you don't try. I know it sounds like a little bit just like kind of like naively optimistic, but, you know, you certainly don't wake up one day, you know, 2 decades down the road after working at a big company for a long time and then just suddenly IPO a company, right? And so you have to have a, you know, uh, you have to have a line in the water to catch fish is sort of how I When was your first round?

When did you raise your first, uh, because I know you've raised tons of money since then, but it sounds like you were financing through services and consulting before then, which is really interesting model that a lot of people don't talk about actually. Most people, especially founders these days, they're like, you know, it's raise money or go home.

Briefly back on that, sort of like what kept me going. I mean, I honestly, it was more just like, I was so passionate about this. I really wanted to build a company, and I just just don't run out of money. Don't fold the business. Right? And it's just like keep you'll you'll eventually find something that'll that'll work. I just sort of one of my first one of my early mentors is a guy named Robert Haddock, and he always used to tell me keep the faith. And it sounds like kind of just a weird phrase, but I would just think about that like keep the faith. You know, just know that it will work out if you just keep. keep going on it. Um, but in terms of like when we ended up like raising our first real money, um, it was with when Dreamscape came out and it was having this sort of amazing exponential growth curve in the users and the images processed. And we raised about $600,000 in, um, in, in, uh, 2015-ish. Um, but we had— I had this guy Gary Bradski put in some money in, um, early 2015 before DreamScope. Um, he was the creator of OpenCV, which is like a computer vision library. And, and then a friend of mine, Austin Russell, who, who was our first investor, and he put in $20,000 at a $400,000 post-money valuation, which is, I think, a great, great investment.

Um, $400,000 valuation. Wow. People don't raise that.

Post-money, $400,000 cap.

Yeah, that's crazy. People don't even raise for like less than $400,000 these days. So there's a cap with that. Wow.

Oh yeah, I, I forgot we also had the Lambda Hat product too. I forgot about that pivot, but that was like part of the face recognition API, was sort of like to, to, to, um, gather datasets as well for image recognition. Uh, it was basically a camera embedded in the tip of a baseball cap.

You had a hardware product?

Oh yeah, yeah. That was our first hardware product, was this baseball cap, um, with a camera embedded at the tip of the brim. I flew out to Shenzhen. I found like a, a smartwatch manufacturer and we like embedded the board like inside of the, the hat here, a little, uh, lithium battery. And then you had like a flex PCB that ran through here and embedded into the tip of the, the brim. And, and then it just took a picture. every 10 seconds to gather datasets for image recognition.

Of your face or what else?

Uh, it was just like basically a wide-angle lens of just whatever you're seeing. It didn't have audio because that was actually like— I realized that that was like kind of— it felt like too much of a privacy. I didn't want to have all my conversations recorded, but it was, it was, it was mostly for gathering pictures of the world for training neural networks.

Oh, very cool. I mean, this whole area, everything you were working on 10 years ago has like become, you know, very big today. Like Snap just released their specs, right? And this augmented reality thing has become quite big. What are your thoughts on that space? And also, what are your thoughts on the image, AI image space? Because that also went through its own growth phase.

So it's funny because, you know, you've seen a bunch of different iterations of both sort of like these types of wearable computers that can gather data. So Snap's product actually came from a company called Vergeance Labs. This guy Eric and then John Rodriguez started Divergence Labs, and I actually ended up meeting them out in the Bay Area. And I— they had at some point offered to have me join the company or whatever. And I was like, you know, not interested in joining another company. But the— that is like just a really difficult space because I just think that there wasn't— you know, there's not a lot of product market fit in terms of— I guess I see some people wearing the Facebook ones these days now, um, the, the Ray-Bans. But I still think it's a little bit awkward and people just prefer to take their phones out. Um, and then, uh, the, you know, the image space. So we went from like sort of Instagram filters, DeepDream, and then we have kind of like a renaissance of the image generation with when Midjourney came out. And, um, you know, there's a lot of people who are just kind of like, I again, old hands in Silicon Valley. So like Dave Holtz, who originally started Leap Motion, which I think it was called Occipital before, which was basically a hand tracker for virtual reality. You know, he— I remember him texting me in like 2022 or 2021 and saying, hey, like, check this out. We're, you know, there's—

there's—

we don't really know what this is, but it's like going to be really huge. And then I remember One of our biggest competitors in the DreamScope era, a company called Prisma. I don't know if you remember this. It was like some of these Russian guys or something. It was called Prisma Labs and they ended up kind of taking away. And there was a few different acquisitions. The Snap Lens— Lens— Lenses was a company called Lensa. And then Prisma kept on going and then they had a resurgence in 2022. Um, and I feel like this is, you know, you, you kind of wonder what the next sort of— what's the next rocket booster on image generation or video generation? You know, there's so much, so much activity now, but like, I— we haven't really kind of broken through that next barrier. I think that it's going to unlock everything, you know, which is to say the one prompt that makes a full Hollywood-quality feature-length film, you know what I mean? I think that that's like, it's a high bar, but I think that that's like the next unlock. And it seems like we probably will get there.

Well, there's several YC companies now and also early-stage companies who are doing these sort of like movie labs, um, where they're like, uh, funding creators to like make these AI movies. Um, I think there's an interesting model there where like you can make an AI short for pretty For relatively low cost, and you can actually test the concept out. And and there's this whole trend now of AI-generated short-form video serials, like almost like soap operas, and that are very popular actually in China.

In China, they seem to be able to monetize it. I was talking to somebody about this. What is the platform that they're monetizing on in China? Is it is it TikTok?

Is it like you know Douyin or yeah Douyin okay Douyin is like the TikTok. Yeah, yeah, yeah. Uh, the TikTok equivalent.

And they're able to monetize enough because I've, I've, I, a friend, a couple people I've talked to said that they've been able to monetize. I haven't seen any American companies be able to monetize it yet.

Yeah, I've seen a few on TikTok and others that like TikTok now is trying to encourage you to like watch these things as well.

Is the Fruit AI slop, you know, the Fruit AI slop videos?

Yeah, those are fun.

Fruit AI slop, was that one?

There's been like a few kind of meme videos with like fruit.

Yeah, they're like meme videos. Is that like, are these people just printing X, X, X bucks?

I thought X stopped its creator program or like revenue share. Yeah.

I, I, I, I just wonder, I don't, I haven't seen the sort of AI generated, um, you know, um, in the American and Western market yet.

Yeah, it hasn't fully taken off here, um, for sure, but I am seeing a lot more interesting content now. I would say a lot better. uh, created content. Uh, YouTube actually has a few good ones now where like people are transporting themselves into different times, you know, in, in the past.

Um, I went back on Facebook recently, uh, I was like on, on Facebook Marketplace and I started to see these sort of like narrative AI storytelling ads. And that seemed to be now is like a thing now.

Yeah.

Yeah.

Well, actually advertising is actually has product market fit with AI. So like in the World Cup, there was a lot of, uh, um, AI-generated. Like, they don't want to— the soccer players don't want to come and shoot a video, so they just get their face and they shoot an advertisement with AI. It's all AI-generated now. So I think that actually has product-market fit.

If the AI is good enough, you can't tell it's AI, right? Like, that's the end kind of result with this. Fatma, my wife, has been producing a lot of these kind of music videos using AI. I posted a couple on X, and they're like trying to promote her fashion design firm. So it's kind of like an ad. Uh, but I feel like music videos kind of fit the paradigm, uh, cuz you don't need like a very long kind of, and it doesn't matter if like the scenes are like a little disjointed.

You know, going back to just like thinking about the early days of Lambdic, I, I, I remember when Prisma came out in 2015, 2016, it went super viral. And, you know, like it was really hard for me to see that because we had started earlier. And we, you know, were being kind of surpassed by these, these, you know, Russian guys in their basement in the App Store rankings. And I remember being with my, my brother, who's my co-founder, and I think I just like broke down. You know, I was just like really upset by this and was just like saying like, look, it's You know, we were so behind because they were able to make it run locally on the phone, which allowed them to scale. And I just remember being so devastated by this. And it's just like really funny to me that like no matter what stage in the business that you're at, when like stuff like that happens, it's actually like maybe to some extent much harder to run a company when you don't have a ton of revenue than when you do have a ton of revenue and you've got a very clear gradient.

I mean, they're both hard.

It's hard to see people leapfrog you, right? And the leapfrogging, there's almost a sense of life should be fair. If you start earlier, you should get more of an advantage, but that's not the way it works.

There's no bonus points for being early in capitalism.

That's right.

It doesn't pay to be a hipster.

Steven, I feel like we kind of missed the end of this Lambda story. Can you talk a little bit about kind of I guess like the— I don't know what success metrics you have, like how much you've raised, things like that, just to like give the audience a bit of context.

Yeah, yeah, no problem. So we started selling these workstations and servers in 2017. We hit about $3 million of revenue run rate that year. In 2018, we did about $10 million of revenue, again expanding the product line from workstations to servers, clusters, and then even a laptop called the Lambda TensorBook, which we did in collaboration with Razer. And then we did in 2019 about $30 million of revenue. In 2020, it was— it got up to about $40 million. The pandemic was really hard because in the depth of the pandemic in April, the loading docks were closed. You literally could not recognize revenue as a physical hardware company because the loading docks for your customers were sometimes closed. And that was really, really difficult. Then by 2021, we had really started to see a lot of progress on the cluster side of our business. And I think we got to $70 million in 2021. 2022, we just passed a little bit over $100 million of revenue in that year. 23, 24, you know, we can continue on. The cloud starts to take off at the end of 2022. And that business has just like, it grew stupendously fast. And, you know, today we're, we're, I think, I believe, you know, we'll probably print out, you know, we're at around a billion dollar revenue run rate for just our cloud business. And in fact, last year we ended up shutting down the hardware business, and we don't even do workstations and servers anymore. And so from a revenue perspective, we're at that billion-dollar revenue run rate and are, um, yeah, have totally pivoted the company to pure cloud infrastructure for AI.

That's amazing.

That's incredible.

When you were talking to investors, you know, when you were like $3 million, $10 million in revenue, like Did you have to answer a lot about like, oh, this seems like a niche market and it's only academics using this thing? Like, how did you kind of answer those questions?

Oh yeah, I'm not sure I had great answers for any of them because the vast majority of investors we talked to didn't invest. You know, their critique of it, which, you know, was valid if you're like very SaaS-brained VC and, you know, is going to be Oh well, it's a 30% gross margin hardware business, non-recurring. We don't like hardware. It's— they would say it's capital intensive when it's like really actually just working capital intensive. Like they don't even have— so it's like there's many degrees to capital intensity. There's actual capital intensity, which is what the business we're in today, which is an infrastructure business. There is a working capital intensive business and then there's no capital you know, a capital-light or SaaS model, and they didn't even know how to parse it. For many of those years, we were profitable like we were. As one of the prominent VCs in Silicon Valley said, we were punk rock profitable. And, you know, we'd go back and pitch every single one of the mainline VCs on Sand Hill Road, and we would get nos, and I would just collect the nos and You know, it was, it was very hard to convince them. They didn't believe in AI. They thought that AI was a bubble in 2021. And they, you know, they thought that it wasn't a big enough market. They weren't sure if, you know, crypto seemed more interesting. So yeah, it was always really hard to raise money as a company.

And did it flip after ChatGPT? I guess probably your whole world flipped.

To be honest, if you actually look at the way that the infrastructure market has rolled out, It— within Silicon Valley venture investors, it didn't flip. Because if you look at all of the great infrastructure companies that are out there that were, you know, that blew up around that time, many of the traditional Silicon Valley venture funds did not invest in those companies. So I don't even think that ChatGPT caused it to flip for them because they just didn't know how to parse a capital-intensive infrastructure business. And I think the same was true for Anthropic and OpenAI themselves, right? I mean, it— many of the traditional investors have come on in later, later on for sure. But like, you know, they, they weren't seeding it.

Yeah, it's interesting to see the investors who are in those companies, right? Like they're often non-traditional investors or some of the funds who are not top tier have actually leapfrogged the other ones because of— so the same leapfrogging thing we talked about, right? Like, the, you know, by, by seeding Anthropic or being early there.

Yeah. And, and it's like the same for like other, I guess you'd say, whatever people would call it, hard tech or deep tech, you know, hardware companies, right? You know, there's now many great examples like Anduril, you know, um, uh, there's just like, there's a whole suite of defense tech companies. I'm seeing like Saronic, and you look at like Valor Atomics. And, you know, there's like a whole new set of things. I think that, you know, Lux Capital did a good job, I think, of pioneering this sort of, um, deep tech or hard tech investment model. And but, but yeah, I mean, a lot of these were just like very non-orthodox bets, right? They were, they were, they were, they were very sort of non-orthodox investments. That ended up panning out.

One thing that's been fascinating to me to see is like how many of these companies have started as something completely different and like, um, and then, you know, kind of natural curiosity led them to this place. Um, and, uh, you know, Hugging Face is another one. Um, you know, kind of similar story to you in some ways, right? Like started off as a consumer play really. And, um, and, you know, their curiosity led them to this other space. Um, and, um, you were with Scale as well, right? Scale was doing something also very different back in the, in the early days.

Yeah.

Yeah.

And also at like pretty humble margins, right? I remember analyzing that business and thinking, hey, you know, maybe we could get into that business or something. And, and, and, you know, I, my, my, my pencil, I penciled out like a 30% gross margin, you know, services business in a sense. And, um, Again, yeah, I think there was a lot of it, just like many other businesses met with a lot of skepticism. But the interesting thing with the AI infrastructure side is coming back to like how investors' mentality has evolved over time with it. So if you actually zoom out, the capital markets are a lot bigger than just Silicon Valley, right? You know, we have institutional Wall Street capital, which if you like study the history of it, a lot of the American financial system was really created in response to the railroad boom in the late 1800s. And that is why the United States has such a vibrant financial infrastructure and ecosystem. And those investors, you know, traditional private equity, traditional or alternative asset managers, traditional credit, like big— the biggest investors in the world, they're starting to really actually understand the GPU asset class. They're seeing that like it's not all, you know, that basically the usable life of the GPUs is much longer than that was originally underwritten by the early pessimistic creditors in the space who did a phenomenal job because they made so much money on their initial investments. And they're starting to actually like understand the asset class and it's becoming like building a power plant. Building an AI factory is like becoming building a power plant, building a Uh, you know, a major infrastructure project. And there's— they just, they love it because it's, you know, it's a very fungible asset. It's something that is, um, you know, highly in demand. It's like, it's like building an oil well.

One thing I learned is also building Presto, I had very similar problems as you, you know, selling hardware to the restaurant industry, right? Um, and, uh, we pretty— we definitely could not raise in Silicon Valley. We had a very difficult time actually. And a lot of our investors came from very unorthodox places. And actually, the advice I got was at a startup school event or something, or like a talk as well with Ben Silberman, who gave a talk at Pinterest. And I told him, I'm like, I can't raise money for this thing, what do I do? And he's like, he's like, there's money everywhere, they don't need to make money from here. Like, you know, money is green, just go find money, you know? Like, everyone has money, just go find it, right? And that, that opened my eyes as well. I was like, oh, So I don't need to just like depend on the Sand Hill guys. And, uh, and so that got me thinking and I started doing this whole circuit, like global circuit, you know, we raised money from like so many different countries, like Recruit in Japan and, you know, random like Asian countries, European players, um, and, you know, family offices directly. So we got really creative with our funding strategy because we couldn't raise from the traditional guys. Um, and I, I think entrepreneurs, that's good advice for all entrepreneurs is like, you just gotta find if you need capital for your business, you get— you gotta get creative. Don't just depend on the normal players. And it's not like universities, you know, people compare— think of it as universities. It's like, oh, if you don't get to the top university, like, what's the point of going to like a second tier or whatever? It's like the same— it's the same money, you know, no matter where you get it from.

Yeah, yeah. I mean, the def— you know, isn't the definition of a currency that it's like maximally fungible and it's, uh, you know a perfect medium of exchange. And so it doesn't really matter where it comes from. We saw the same thing. I mean, you know, so our Series D was some huge percentage of it came from Taiwan and, you know, Taiwanese OEMs, ODMs. I flew out to Taipei, met with a lot of the folks that we do a lot of business with these days, whether it's WeWin Wistron, Pegatron, companies like Chenbro, which, you know, you wouldn't have even thought this is like— these are just like sort of much more nontraditional investors, but they're all businesspeople who understand the industry and who better to raise from, you know, Charles Liang, who's a good friend of mine now from— who's the founder of Supermicro. So we've always kind of gone in and really tried to find, you know, find out who you can be strategically aligned with to get capital that way.

Have you been thinking, Steven, about this kind of goal you had in 2012 about going public and IPOing? Is that kind of closer now? I mean, I have that goal as well mentally. Like, that's one of my entrepreneurial success metrics. Like, I feel like going public is not cool anymore. So like, how are you thinking about that as like a goal for you?

Well, it's amazing because just of how much the markets changed from, let's say, the late '90s, which is like I, I wasn't in business at the time, but, you know, you would kind of see the dot-com era. And I— there's, there's a, there's a bunch of different reasons for why that likely is, whether it's, you know, Sarbanes-Oxley or, you know, the regulatory overhead and burden. But, you know, these private rounds would have been blockbuster IPOs, considered to be blockbuster IPOs years ago. And so I think that it always makes sense if you're a capital-intensive business to be out in the public markets. The public market's about 10 times the size of private markets in the United States, about $100 trillion versus $13 trillion in the U.S. In terms of just like the gross size of it. And, you know, now look, in terms of— I don't— you know, we never discuss what our plans are for fundraising. But, you know, eventually with any mature company, you want to be public. I think that I definitely always kept that in mind. And, you know, we'll— when the time is right, we'll do that.

Immad, I know you've talked about this before, something you want to do. I actually want to talk about your new product that you just launched yesterday. This is very exciting, this, uh, the books product. Stephen, did you see this one?

No, I didn't. Tell me more about it.

Uh, yeah, we just, we launched it yesterday. It's, it's basically like a Mokri, uh, it's called Mokri Books. It's kind of a QuickBooks competitor focused on kind of small business accounting. Uh, obviously in the AI era, we're doing a bunch of things kind of to automate categorization and reconciliation and things like that. But it's still actually targeted to bookkeepers. You know, one of the things I learned while I was like researching this product is like a lot of people think you launch a product like this by targeting it to kind of business owners and founders first, which I think is like how you end up failing. Because, uh, at the end of the day, bookkeepers— Dentists.

You need dentists.

Yeah, but even they have bookkeepers and accountants, right? So like, I think, I think the way they think about accounting is very different. So We've actually been working on it for a long time. We made an acquisition, I guess, a year and a half ago that was in this space. And then we had to— it's actually kind of a complicated product. It's got a lot of edge cases and it's one of those things where—

What was the company that you guys acquired? Was that— it was like a quiet one or?

No, we announced it. It was called Teal, but they weren't like— it was a relatively early company kind of doing like API for accounting. So it took a lot of steps to go from that to like a full-fledged kind of books product.

I always tell people that I, you know, that are founders that I talk to, I always tell them that like having a really solid accounting process early on in the company and getting a really strong grip of your balance sheet is like one of the most important things, you know, especially when you start making money. And having a real business, working your cash conversion cycle and actually really understanding what it means to manage cash flow in a company and like is like super important in having good foundations.

Steven, you, you ran like a struggling, somewhat profitable startup for a long time. So you build up that muscle. I think most people— well, a lot of people nowadays go raise like $5 million day zero and they just don't have that kind of muscle. Like we— in my last startup, we struggled for many years and it was very much like You know, yeah, it was even worse for us because we had a freaking marketplace and we had people who'd have to pay us before we could pay out our publishers. And it was literally like, we were like, oh, we just literally can't pay out the publishers until we get this invoice paid. And it was every month was this like super stressful, like timing to make sure the bank account didn't go like negative because we had to like have these like receivables and—

Yeah, working capital.

Yeah, it was such a pain. But you learn a lot when you go through that kind of pain in terms of like, okay, what are you managing against and how to manage it? I don't think it's actually that complicated. You just have to spend at least a day a month thinking about finances, which isn't that high a burden.

Well, I mean, I can imagine with the state of the art with these models, I mean, when I use something that's like a frontier model to edit a spreadsheet, even very complicated, like 5-tab financial models, it finds like the perfect little bug that was missing or, you know, some even some very subtle errors in the logic. Not, not, not necessarily. It's not like an error, like, you know, it's not throwing ERR inside of the sheet, but like, you know, a little logic error in terms of how you're maybe amortizing a fixed cost. It does like a— it does a really good job and better, better than even a diligent human sometimes. And so I can see how that, that's like getting to the point where it can be a really valuable thing for people.

Yeah. Ironically, I was actually Googling AI bookkeeping services the day before you announced it. I was like looking for like, you know, 'cause, uh, my bookkeeper was making a lot of mistakes. I was like, this is, uh, and you know, slow response times. And, you know, I was like, well, what are these guys actually doing that can't be done by AI? Um, so like I was actually looking for it. So I immediately turned on the product and started playing with it. Because, um, I was looking for something like this. So it's cool. So you'll be taking a bite out of Intuit.

That's the plan. I mean, I think like I really used to hate this kind of non-real-time nature of bookkeeping. Like it was just like, I'm like, what, you know, like you want to know your costs went up in something as soon as it happens, not like at the end of the month, 15 days later, because like someone categorized it kind of thing. So I think you can achieve all of that.

Yeah. And then, and then you're trying to do, you know, really at the end of the month and then realistically it's like, quarter, quarter views is a little bit better because, you know, you'll have sort of out of, out of normal scope expenses that you want to average out over the quarter. And it's just like, you know, it's like sort of that old economics joke, which is like sort of like, you know, economics is sort of like trying to drive a car by looking out the, your, your, your rearview window. And it's like really difficult to do that. It doesn't really. And it's same thing with accounting if it's not, if it's not real time. I could see that being super useful for people.

So what did you think, Raj? Did you— we had a bit of a delay with the categorization of everything, so I don't know if you got hit by that because suddenly when thousands of people sign up online, you're supposed to be like— we were like, oh, we think it works. But then it was like, oh shit, actually we weren't ready to categorize everything. But I think that's fixed now.

It looks very cool, very powerful. You know, I think I need to play around with it more. I had a bunch of questions. I was thinking you probably should have a community to like, you know, for people who are playing with it to ask questions. Questions. Um, but, you know, I think, uh, it looked very powerful from the beginning. Like, the dashboards that came out of there were way more involved and detailed than anything I've seen before, um, you know, especially from like an agentic product. So, um, I, I need to do more playing. I was like trying to figure out how to, you know, get better at categorization. That's one of the things that annoys me is like going through everything and categorizing. I'm like, it's pretty obvious what this category is. Um, and so, uh, I think that that's one of the biggest time sucks, right, of bookkeeping is categorizing.

Did it not auto-categorize correctly?

It did, but then it had a bunch of stuff it couldn't categorize, and so I was trying to figure out why, why it couldn't figure it out. Yeah.

Yeah, well, we just launched it. We'll keep improving it, but give us feedback. Uh, we're definitely working on it.

Yeah, yeah. Uh, Stephen can run his billion-dollar business, you know.

Yeah, but we're not quite there yet.

But once you get the NetSuite and then eventually, you know, SAP integrations, then let me know.

Mercury runs everything on Mercury, but we, we're not quite there with Mercury Books yet since we have a fairly complex business. But I do really want to get there. I think that would be cool.

Stephen, I had a couple more questions for you. Um, actually, one big question is, uh, you know, you stepped down from CEO to CTO, or you stepped to the side, I would say. Um, and, uh, tell us about that decision. Now, I know, um, Both of us are long-term CEOs and I've sort of grappled with this in my career as well. But yeah, we'd love to hear your thoughts.

Well, first of all, like, I've never been happier than I am now. Like, I get to actually work on the stuff that got me into the business in the first place. I'm focused on some really critical sort of innovative new ways to build AI factories. And, you know, you'll hear about that in a little bit. You'll see what I'm— what I've been working on, what I've been cooking up. But You know, I think that it's kind of the way I describe it to people in terms of making that call is you find yourself— okay, if you're the CEO and you started the company because you love technology, because you love software or you love AI, you find yourself spending most of your time doing fundraising or HR stuff and recruiting people. And talking with customers. And but fundraising, especially in a capital-intensive business like Lambda, was like just all my time. And I did that not because I enjoy fundraising. I don't think I'm a particularly good fundraiser. I, I've never really been able to do anything other than just show the revenue growth and say like, all right, well, we're raising money. And I didn't like it. And I would openly talk to people about this. I didn't like that part of the job. And yet I had to do it because that's what the company needed of me. And it takes a while for a company to get to the point where you can afford to hire the quality of executive that's going to come in. And do it. You know, it's not, it's not inexpensive. You know, you can look up any public company that has a, you know, certain revenue run rate and has an outside, you know, has a CEO that's come from not within the company as the founder. And you can look up how much the average is per year. And it's not inexpensive. It's something that most companies simply cannot afford. And so like getting to the point in the company where one that was like affordable is like an important part of it. And then the other one is just you know making you're choosing the the the the right person. I mean that's that that was always like you know who knows how things work out. You know you do your best to predict the future, but it's been really awesome. Like honestly having Michelle Michelle Combs, who is our new CEO, he was previously the CEO of SoftBank International as well as the Previously the CEO of Sprint, the, the major telecommunication company. And he's— he knows the capital markets like the back of his hand. He's raised tons and tons of money. He's used to super large deals, very infrastructure heavy as an executive. And it's worked out so amazing. Like, I love working with him. He's come in. He understands our culture. He understands what we need to do to win. And so I'd say for the funny thing is that most founders, and I won't name any names because, but a lot of people who know me came out there like, oh my gosh, you're so lucky to be able to do this. Like, you know, there's a lot of people who are CEOs, I think, who are a little bit unhappy in their jobs secretly and wish that they could do that. So I think that You know, uh, it, it can, it can oftentimes work out.

I mean, being a CEO is a pain in the ass. There's no doubt about that.

Uh, yeah, it is not a fun job in many ways. Like, you know, there is a— and actually gets worse as it scales, actually, I think, um, because you end up being like a symbolic figure and like you have to deal with so many different stakeholders. Uh, I think even Sam Altman was like tweeting about how much he's not enjoying it, you know, like it's, it's, uh, Even the best.

It's difficult.

Yeah.

Even the best.

It's a very difficult job. And I think most— there's a reason why they're compensated fairly high that most people don't understand, is like you're doing a lot of really, uh, difficult things. But yeah, I totally understand.

I think you really have to believe though that someone could do a better job than you, which, uh, you know, I would— I don't know if anyone else could do a better job at Mercury than me. But I like, you know, obviously for you, you found someone that you feel like confident in, and it was maybe turning into a job you weren't enjoying.

Yeah, I totally agree with you. Like, for the longest time, I kind of, I mean, you know, look, privately and even with some of my investors, which, which, you know, I don't always recommend, don't, don't, don't, don't take these things as all advice, but like, you know, would have these conversations And the reality is that, like, for the vast majority of companies, you're doing it because there's just no other option, right? You know, like, it's just, it's just, that's, that's, that's what you have to do. And it takes a certain degree of scale to be able to, to find somebody who will do it, you know, much better than you with, you know, given the uncertainty of that transition, if that makes sense. Like sort of like the expected value of how good of a job they're going to do vis-à-vis the sort of uncertainty that's inherent in any hire.

But there's also like, I think, a skillset match, which is what you're talking about, where like the skillset for the CEO for Lambda became like amazing at capital markets, which wasn't necessarily your skillset.

No, right, exactly. I mean, I had to learn it up. Yeah. Yeah.

If his skillset continued to be like, hey, you need to be like an innovative startup-y, like technical person, then like it might've been harder to like replace that? I mean, I'm not— oh yeah, you might still have been the best person for it. Because there is this whole thing which I do believe in that like founder CEOs just have this like kind of ability to be more revolutionary because they have like that remit that they can like reinvent things in a way that like non-founder CEOs can't. Uh, hopefully you could still do that as the CTO founder and in support of the CEO, but that is a difficulty.

Yeah, I, I, I think, you know, there's, there's the founder remit. There's oftentimes they They are heavily involved in the board of directors. You know, they have the sort of, sort of whole industry in their head. These are, these are some of the things I think that the interesting thing is that when I look to a lot of the people that I looked up to at the time and, and I still do, whether it's like Steve Jobs or Larry Ellison or Larry Page, go down that list, Bill Gates too. Every single one of these folks ended up stepping into basically a chief technology or chief architect officer's technology-focused role and brought in a CEO or promoted one from within the company. And so I think it's very common. And a lot of people don't remember that because there's still a lot of Obviously folks, whether it's Elon Musk or Mark Zuckerberg, who are sort of like this, are prototypical, have not taken their hands off the reins necessarily.

I think it's a big topic and, uh, you know, kudos to you to, um, you know, finding something you enjoy more. I think there is definitely burnout that can happen, you know, having done it for a while. So if you find someone that you really admire who can take the wheel, that, that's great. And it could take a lot of the BS out of your life.

Yeah, no, it's, it's been, it's been wonderful and I cannot wait to Uh, to, to show everybody what I, what, what, what's been the next, uh, big thing for us.

Ooh, I'm very excited to see that. And, um, I, I, one big topic we didn't get to talk about, but we'll get you back soon to talk about is the, the whole data center discussion. I know you've been very vocal about it, how communities, you know, shouldn't be afraid of data centers, right?

Yeah. Well, I, I, I think what I've gotta say on that is there's so much disinformation out there with regards to what a data center is, what it isn't. And I would just highly recommend everybody Go to a local big data center that's in your town, and it'll be a building that you've never even thought of or even heard or even, you know, there was a great video as I was tweeting about some of this stuff or posting about some of this stuff on X. Somebody said, I'm at a data center and it's a 250-megawatt data center. They've got their camera out and it's just like quiet. The road is far louder than whatever the building is at the, at the fence line and at the, at the, at the edge of the property line. And I was just like, see, exactly. There's so much misinformation because, you know, what will happen is you'll get a news story about a Bitcoin mine, right? That's a converted warehouse where they're throwing in 10,000 Antminers or something with the fans going, and then they put the microphone up and say, look, see how loud this is? It's like, that's not even a data center. That's a crypto mine retrofit.

And so there's just so much misinformation. Are you sold on this conspiracy theory that's like China and Russia kind of spreading this disinformation to try to hamper us?

I think there's there's tons of evidence that there's been nation-state actors that have have spread disinformation in social media. There's there's no question about it.

I think yeah, it does. It seems so like there's the water thing, and I'm like, you know, why are they guzzling water? It doesn't make any sense.

No, no. And the water thing doesn't make any. sense. It makes no sense because these systems are literally closed-loop dry coolers. It's, it's, it's no different than your AC. Your AC has refrigerant or a liquid that runs through and it doesn't, it doesn't leave. It just dissipates in through a heat—

I think the only, like, the only somewhat fair concern is it could drive up electric— electricity prices if we don't manage it correctly.

But it's a fair concern. But the Loudoun County, the county in Virginia which has the most data centers in the United States, has a lower average dollar per kilowatt-hour than— a slightly decreased one than most of the rest of the United States. These data centers have been bringing power, one, two, bringing tax revenue. I think Loudoun County is like $1 billion or more a year, maybe, I think is the statistic on it, in terms of how much data centers are bringing into the county. So again, there's just a lot of misinformation about it.

Yeah, that's crazy.

Seems to be a media bugbear too. Like everyone wants to talk about it. You know, it's disproportionate to its impact.

Yeah. You know, whenever there's something that's coming up in a local area and I think that here's the thing, the communities definitely have a right to kind of understand what's going on inside with the facts, not with hysteria or with a sort of misunderstanding of what it's like, right? I think they need to understand the benefits, they need to understand the downsides, and they need to then make a, a great decision for themselves. And, uh, so Lambda's actually recently— we've signed the, um, the The White House's Ratepayer Protection Pledge, where we've agreed that we're going to be bringing power, paying for power, and making it so that it's not possible for the ratepayers to have any experience increase in their price. That's been a pledge that's been signed by every single major AI data center and AI factory builder.

I think the AI leaders don't do a good job when they also scaremonger about AI because I think in people's heads it's all like this like big box of like, you know, data centers evil, like AI is going to kill us with like bioweapons, slash it's going to go AGI and rogue. Like all of this stuff becomes like this like kind of like thing that the media latches onto that like AI is evil and this is just one other part that's evil. Like I wish we could just be like a little more optimistic about AI and like talk about like, hey, actually like, yeah, everyone loves using it. Like it's like cool and it's the future. Like we should just be optimistic about it. And I think that would help.

But it's hard when like the tech leaders aren't optimistic I think that this is what my hypothesis is, because I'm a relentlessly optimistic person. I think that what we're going to see over the next 5 to 10 years with regards to new medicines and new therapies that are developed that are extending people's lives, addressing some of the worst diseases that are killing loved ones, and when AI is able to show that kind of a benefit and it's going to be so clear, people are going to say, well, we want— how many of these can we build? You know, we want more of these life-saving and life-extending therapies. I think that's going to be a really huge benefit, you know, in addition to the benefits that we all experience and enjoy now. But, you know, I think that these types of therapies resonate more with people, more than, let's say, something that's like better than Google Search or for automatically generating code, because most people don't have writing code as part of their day job.

Have you read Enlightenment Now? Um, this concept that like, you know, everything is so much better, but like humans aren't good with slow benefits over time. You know, like take a while to FDA to approve some cancer drug and all this stuff. So like we always look at like, what's the, you know, what's the mimetically bad thing that's like catastrophic rather than like the slow benefits that come later over 10 years. Uh, I agree with you though.

I should, I should read it. I'm, I'm, I, I'm definitely familiar with like sort of, you know, Steven Pinker talks a lot about, um, just how much more peaceful and healthy and brilliant our society is, you know, post-Industrial Revolution, and, um, and just how amazing things have gotten. And, you know, humans will always One of my favorite X posters is, of course, Pessimist Archive, right? You know, the world has always been getting better and better and better. And there's always been sort of naysaying around when we introduce air conditioning, right? Like, this is another example. That's like probably the biggest previous grid buildout in the United States. It's about— I think it's something like— I want to say it's like high tens, if not 80 gigawatts or something for air conditioning in the United States. Well, like, that's like a very clear advantage that you get from having electricity, right? You know, why should we build power plants? Well, we like not being sweltering hot in the summertime. We like having ice, right? These are— these were absolute luxuries, like the most luxurious thing you can imagine back then. And now it's a common everyday thing. And so I think that that's like Uh, an example of like, I think the healthcare thing as well as, um, um, other things like the way that self-driving cars are going, where it's going to reduce automotive fatalities, are going to be the things that people will hopefully be able to point to and say, wow, that was all worth it.

Yeah, I, I think, um, you know, the slow benefits are one thing. Also slow, uh, risks are also tough for humans, you know, like, uh, the debt accumulating over time, or, you know, um, you know, maybe our pandemic safeguards are eroding. Like, there's different things That could happen on the risk side as well. It's like the boiling frog, right? So, yeah, we're not good with slow things in general.

And we're probably getting worse at it.

Yeah, because everything seems to be happening right now, right? You open Twitter, it's like everything's happening right now. Even something happened yesterday seems like old news, right? I get a print subscription now to the Wall Street Journal, and I'm struggling sometimes to read it because it's like This is like a day-old news.

And so that's interesting, actually. And I talked to some founders about this. One of the things that I found is that the less news you consume, one is like you can be happier by consuming less news. And then also, I just feel like the benefit that I usually get is by reading about history and a little bit less about reading about current events. Um, as a way to help guide thinking. Um, and that to me has been, you know, you'll see sometimes Sam Altman talk about long-term thinking and like the powers of long-term thinking. And I think that it starts with like media consumption diets.

Hmm.

Yeah.

Good idea.

We better end this. Uh, we're over time, but this is great. Uh, Stephen, we should, we should get you back.

Why did you choose data centers at the last minute? I know.

It's a fascinating topic, guys.

Thanks so much for having me on.

No, this is great, man, and great to finally catch up. So we should, we should stay in touch more than once a decade.

Awesome. Let's do it.

Yeah.

Thanks everyone for listening. Uh, please leave us a review on whatever channel you're listening on. Uh, makes a big difference to the algorithm and to the success of the show. And see you all next week.