Roblox 的 David Baszucki 打造了地球上最大的游乐场

You just said something interesting right before we started recording.

Yeah.

I asked you how old you were when you started Roblox.

That’s right.

You told me the age, and then you said, “Oh, there’s actually an interesting story and a metaphor there.” What did you mean by that?

First off, it’s a super interesting story because, arguably, one could say I started Roblox when I was 2 years old, because a lot of this stuff really fascinates me.

Before Roblox, I had started a company called Knowledge Revolution, which made Interactive Physics, educational software—a simulator for learning physics. We were in this very unique position with Knowledge Revolution, way back in the early days of the Mac, where we had all of these kids who, instead of using it just to do their physics experiments, were starting to build stuff.

It was 2D at that time. It was just in the pre-Internet era, so we could see them all trying to build and share their stuff. We thought, “Oh my gosh, there’s going to be a whole new market here: immersive 3D multiplayer—playing, working, learning, listening to music, all of that stuff.”

What happened along the way to the founding of Roblox is that, in the period after Knowledge Revolution, I took a 2-year sabbatical. In a way, I went a little astray. I started going almost more logical: “Okay, I want to start a company.” I wasn’t thinking about all of the stuff we learned at Knowledge Revolution.

About a year in, I had a bit of a vision. I was saying, “Whoa, you can’t be logical on this. You have to be intuitive and go back to some of the roots of Knowledge Revolution,” which was all about fun, play, and building something very innovative.

So instead of following this logical track, some people from Knowledge Revolution and I said, “We’re going to do this very unorthodox thing and build this wacky new product: immersive human co-experience, multiplayer, cloud-based, creator-led UGC.” It was very illogical and very risky.

It was logical and risky because, at the time you were doing this, this was around 2002?

It was very—

2003?

No one got it back then.

Yeah.

No one quite thought of it. When we finally raised some angel money, we had a business-plan slide that was actually very accurate to today. It was a little bit about the history of storytelling and communication.

The history of communication was the mail system, voice, texting, maybe video. But in science fiction, everyone was talking about the holodeck and that immersive stuff we would see on Star Trek. We actually believed it, and that was part of the idea behind starting Roblox.

We thought 3D immersive digital stuff would combine communication, being in the same place, storytelling, and the rest. That’s how we got our launch.

The other cool thing about the launch is that we initially thought it was so fun and cool to work on. Even a 4-person lifestyle company at that time was very appealing.

Okay, so a 4-person lifestyle company. The first company you started, you didn’t raise any outside capital for, correct?

Knowledge Revolution got very far without raising any money.

And you sold it for $20 million, something like that?

That’s exactly right.

That’s when you took the sabbatical?

That’s correct.

Explain to me how you knew that you were being logical instead of following your intuition during this sabbatical. What do you mean by logical?

I think “logical” goes to what someone is trying to optimize for. Literally, at that time, I had been a CEO, so in a way I was optimizing for being another CEO and dropping into CEO-ness when, in fact, a lot of the magic of Knowledge Revolution had been about inventing new stuff.

That was actually the superpower of some of the early Knowledge Revolution people and myself. It wasn’t just about being a CEO. So when I came out of Knowledge Revolution, I went and looked at a bunch of CEO jobs that seemed like the logical thing.

Wait. You were applying to be the CEO of a company you didn’t start?

That’s right. What a mistake, right?

What’s your distinction between a founder and a CEO?

I kind of learned at that time that my founder mode from Knowledge Revolution was something I couldn’t find a position for. Obviously, it was a mistake to be looking for that type of thing. That was much more of a founder—

You’re a world-builder.

That’s right.

There’s no way you’re going to jump into somebody else’s world.

What you’re describing is literally a vision I had one night. I was on this path, and there was a big barrier there. But this other path, which was more about being a world-builder and a creator, was like, “Boom, I got it.” That’s when Roblox started.

Okay, we’ve got to pause here because there’s almost a paradox between those two things. I had dinner with Ho Nam, who’s been one of the earliest investors in Roblox. This guy’s been telling me about Roblox for as long as I’ve known him. I think he’s been involved for a decade and a half.

Yeah.

He’s like, “The thing about Dave is, if you ask Dave the time, he’ll build you a clock.” It’s one of the most unique descriptions of another human being I’ve ever heard.

You’re known as this eccentric genius systems-builder. But then I watch your talk at Stanford, and all you’re talking about is following your intuition. Can you reconcile those two things for me?

That’s right. They’re both very valuable. I would say that early CEO lesson was definitely a sign of not following my intuition. It was, logically, “Be another CEO,” rather than following my intuition and building Roblox.

I would say combining intuition with tenacity and taking the long view—if those things can coexist—is super, super powerful. Along the way, with Ho, for example, that metaphor of building a clock is funny because when we started Roblox, we used to joke, “We want to start a perpetual-motion machine.”

What is a perpetual-motion machine? It’s something that can keep going and get better and better. That was the notion behind building a cloud-based 3D UGC system. If we kept building that system, creators would make more and more amazing content. We could keep tuning the system, and we’d get that perpetual-motion machine.

The metaphor of the clock is really interesting because, if we dive into that, it’s harder to build a clock. But if you ask me the time every day for the next 20 years, it’s probably easier to build the clock than to tell you the time every day for the next 20 years. That was part of the thought behind Roblox.

I was just on stage at an event with my friend Eric, who's the co-founder and CEO of Ramp. When I looked over to my right, I noticed that on the sleeve of Eric's jacket it said, "We win when our customers win." Ramp is the presenting sponsor of this podcast, and the way that Ramp helps their customers win is by helping you save time, save money, and grow revenue. The median company running on Ramp cuts their expenses by 5%. The median company running on Ramp also grows their revenue by 16%. So, when you're running your business on Ramp and your competitors are not, you have a massive competitive advantage that compounds over time. Ramp is the only all-in-one platform designed to make your finance team faster and happier. Many of the top founders and CEOs I know run their business on Ramp. I run my business on Ramp, and you should too. Go to ramp.com today to learn how they can help your business save time, save money, and grow revenue. That is ramp.com.

Okay, let’s go back to what you were saying. The idea behind Roblox: You had the sabbatical, you started Roblox, and there were 4 people. You described it as a lifestyle business going into that.

It’s interesting. I think the lifestyle business could be a metaphor for having no expectations and being so excited about the area we were going into. It was like a validation of our intuition: “We’re going to go do something enormously fun, and something that, according to our business plan, could be really big.”

But at the time, it was so fun. I think we were thinking we’d be very satisfied if we could work on this for many years. What happened is that it got a life of its own, and the responsibility just got bigger. The second we were live, we could see this thing was going to keep growing.

What was the original product?

There was actually an original product that was an incredible failure. The original product that was the incredible failure was arguably us knowing what the viral product would be.

We intuitively knew the viral product would be online in the cloud, multiplayer, digital, physically simulated, accessible anywhere, user-created, and built around creating cool stuff. We kind of knew that. We wanted to get something out sooner.

What we tried to get out—which luckily was along that trajectory—was more of a single-player, puzzle-builder-type game. We thought, “Oh my gosh, if this puzzle game gets viral, that will give us a little bit more runway to build the full Roblox.”

But as you can imagine, that single-player puzzle-type game is not quite as fun as multiplayer immersive 3D with your avatar, going with your friends, and doing all of these things. Sure enough, within a couple of weeks, we were just like, “Yeah, we knew it. It’s not going viral. What are we going to do?”

So how did you get from the single-player puzzle game to building the platform?

Just take a deep breath. Six more months of engineering. Let’s go.

What did the team look like during those 6 months? Was it still 4 people?

Initially, it was just Erik Cassel and me. Then John and Matt Dusek were just coming on board. So that decision happened when it was the 2 of us.

And had you raised any money?

Nope.

Just having a good time building this thing.

And funding it yourself?

Yeah.

You said something very interesting. You started this with no expectations. Say more about that.

After selling Knowledge Revolution and taking my 2-year sabbatical, I tried that CEO thing when I wasn't using my intuition. Then I came back, and the revelation was almost so big: “Oh, I could work on invention, fun, and inventing Roblox. What a luxury. That is such a fun luxury. I could do that for my whole life.”

I think that was more coming off of that other linear track, just being so pleased with it. Obviously, the second we started going, a lot of other instincts kicked in: the responsibility, how big can we make this? The second we saw the perpetual motion machine starting to work, that was really interesting.

When it was just you, your co-founder, and these 2 other people, were you using the term “perpetual motion machine”?

I was using it, yeah.

Where did it come from?

Even now, if I go on a short-form video platform, every few months I can see some crazy mechanical gadget. If you look at it, they’ll say, “You know what? The water falls out of here, and that falls out of here, and it’s a breakthrough in physics,” and the machine will just go forever. That’s what a perpetual motion machine is. They’ve been around for hundreds of years.

Obviously, it’s physically impossible. Thermodynamics—more energy in than out—and friction. No perpetual motion machine ever works. But it’s an interesting moniker for the notion: “Are you building a system that will have a life on its own? Will it grow on its own?”

I think, in the context of Roblox, it’s something that will organically gather traffic rather than buying traffic. In that case, the content on Roblox was perpetual. It was made by creators, and the acquisition of users was perpetual, and it was word of mouth.

You were how old when you started Roblox?

20 years ago, roughly. So, early 40s.

It’s very fascinating to me that you’re saying, “I want to build a perpetual motion machine”—something that has no end, something that carries on forever. Did you know the next company you wanted to build was going to be the last company you worked on?

I did.

One of the things that Ho Nam told me about you is that you have unbelievable patience and endurance.

Yeah.

He said you want to work on something for the rest of your life.

Yeah.

But you knew that in your early 40s.

I think I actually knew it in my 20s, and even very early on, when I was programming Apple II stuff. For some reason, I intuitively wanted to build world simulations.

Were you playing these games like “Civilization”?

I was more building them. When I was hacking up Apple II code, I always wanted to build “How do we simulate reality?”–type games. Then, when we built Knowledge Revolution, that was arguably a 2-dimensional world simulator.

I would say video games have historically been more about how they look. I was pulled to not just how they look, but how they function as well. How much fidelity can we get out of it? I was really interested in that.

When we formed Knowledge Revolution, it was very early in the days of simulating physics on a computer. There were papers in SIGGRAPH at the time—very early algorithms for how to do collisions and all of that kind of stuff—that we had to pull that research out to build the Knowledge Revolution physics simulator.

That instinct carried into Roblox as well. When we sold Knowledge Revolution and started Roblox, I was thinking that immersive, 3D, physically simulated stuff should have a 20-, 30-, 40-, or 50-year trajectory.

You know what’s confusing to me? You had that understanding so early, and yet you fell off your path for 2 years.

Yeah.

The chance that you were going to build something you could work on for 30, 40, or 50 years—a perpetual motion machine—wasn’t going to happen if it was created by somebody else.

Yeah.

You had to build it from scratch, the way you wanted to do it.

No, that’s right.

I mean, you definitely made the right decision. So wait, let’s go back. Now you’re building this thing, and you’re saying, “We’re not going to do this single-player puzzle game.”

No, no.

“We started Roblox.” You knew it was going to be a platform from that point?

We did.

Okay, but who’s making the games?

At that point, we had 4 people. Matt and John joined Erik and myself. John was a really good game maker, so we had a little bit of a warm-up phase where John made a couple of very famous first Roblox games, like “Classic: Crossroads” and “Chaos Canyon.” They were really pretty fun, and they tested the system.

So we had in-house creation. The 3 of us were coding, and then we had some users starting to play on “Classic: Crossroads.”

How are you getting them?

It’s really interesting. I don’t know what entrepreneur I heard this from. They said, “Look, just go buy 50 users a day from Google. That’s it. You can buy them for a buck a user.”

So we said, “Oh, that’s a really good idea. Let’s go buy 50 users a day.” That was the germination of all of Roblox.

If we were able to reverse-engineer the whole social graph tree, all the way back to the start of Roblox, we’d probably see a 1-month period times 50 users times 30 days. We’d probably see 1,500 users who saw some ad like “building game, come try it out” and who were the initial social graph of everyone on Roblox.

Now, in the afternoon, the 4 of us go online. There are maybe 20 or 30 people just hanging out in these games, and we’re watching them. We had been making those games with Roblox Studio, which was our creation environment.

We were in a rush: When can we publish the full closed-loop system? The closed-loop system is that anyone can download Roblox Studio, make an experience, and push it live. There’s a page where you can see all of the experiences. You can jump in and start playing, and that’s a closed-loop perpetual motion machine.

So we were just saying, “Okay, we have to get this Studio thing out. What’s going to happen? We don’t know. I hope someone likes it.” We had about 1,500 users.

So you went from trying to get them from being users to creators?

That’s right.

Then, a few years later—which I want to talk to you about, because I don’t think many people know the size of the businesses people are building on top of your platform—

They’re big.

Yeah, they’re massive. At this point in the story, you’re turning your users into creators.

And it happened very virally. There was a day when we were in our little office on Chestnut Street in Menlo Park, the 4 of us. It was 1 o’clock in the afternoon: “Okay, let’s push Roblox Studio live and see what happens.”

Then, probably around 3 or 4 o’clock, we started seeing activity. “Whoa, someone just published something. Look at that.” “Oh, someone published something that other people are playing. Oh, look at that.” “Oh my gosh, now there are 20 things published, and that has a bunch of people in it. Oh my gosh!”

We went home that night saying, “Okay, closed-loop, viral system.” These users now have way more word of mouth than anything we’re buying. This system is growing on its own.

Why do they have way more word of mouth than anything you’re buying? Because they’re telling their friends to play the game that they made?

They made a game. All of a sudden, there are 100 different things. It’s not the boring stuff we made. For the existing users, I’m seeing something new every minute rather than this boring stuff we made 3 weeks ago.

New content, breadth of imagination, creators bringing friends—all of those are just viral.

What do you think when people compare Roblox to YouTube?

I think it’s interesting. It’s a little different.

Well, Roblox is inherently social in a way that YouTube will never be.

I think both consumers and creators are part of it, but you hit it exactly right. The comparison I would make to YouTube is the difference between the phone system and reading a magazine.

Reading a magazine on a cloudy day 50 years ago, you could do by yourself. On the phone system, on a cloudy day, you could call your friends in the Leave It to Beaver ’50s and say, “Hey, what’s up?”

I feel the difference is that they’re both content platforms. The content in a video platform is typically solo.

I’m by myself when I watch YouTube. I don’t watch it with other people.

That’s right. The content in Roblox is really a scaffold for communication and being together.

So, is this a game platform, or is this a social network?

We’ve actually said we had slides in our early business plan showing 2 viral loops rather than 1 viral loop.

What are the 2 viral loops?

In YouTube, there’s a bit of a content viral loop. The better the content, the more retentive it is.

Mm-hmm.

YouTube has very thoughtful ways of everyone finding good content. In Roblox, the viral loop is both the quality of the content and the users being with each other. There is both a content viral loop and a communication-connection viral loop.

I’ve heard that other platforms, other messaging platforms, actually see you guys as a big threat. Snapchat is one I’ve heard, because the user behavior on Roblox is, “Yeah, I can play games with my friends.” But the use case I’ve seen is that your younger users will come home from school, and when we were kids, we’d get on our bikes and go find our friends.

That’s right.

They get on Roblox, and all their friends are there. But sometimes they’re not even playing the games. They’re just sending messages to each other.

I would say it’s interesting to think about the utility of messaging, even very visual messaging, or ephemeral or non-ephemeral messaging, versus immersive 3D. In a way, I don’t think we think of it in terms of how much time doing this versus how much time doing that. I think there’s a natural evolution of a wide range of types of platforms.

We’ve seen the natural evolution of text, the natural evolution of photo-type messaging, and the natural evolution of short-form video. I think we actually think less about comparing with the time spent on another activity, and we’re typically saying, “How can we increase the quality of this experience?” Our niche is immersive 3D co-experience. How can we make it better, better, better?

We think it’s a naturally emerging niche, and I would say we feel the spec of our product has a long way to go. Roblox is a very primitive product relative to what will be possible someday.

What other co-experiences besides games take place on the platform now?

We have all kinds of organic things popping up. There are, arguably, of course, concerts and music. We’re starting to see more and more of that being done organically rather than by us. I think Bruno Mars showed up on one of Jandel’s games without us organizing it.

I’d say that’s a good sign. I actually think Bruno was the peak concurrent of music anywhere on any immersive 3D platform, and it was actually within a game. I think we’re seeing early signs of shopping. Who knows? Someday, older people are going to go to church on the platform. Older people are going to work together on the platform. Older people are going to do other things.

I think the natural evolution of a communication platform is figuring out the ways people use a communication platform.

One of my all-time favorite quotes is from the book "Zero to One." It says, "The single most powerful pattern I have noticed is that successful people find value in unexpected places, and they do this by thinking about business from first principles instead of formulas." This is exactly what AppLovin has done with their new advertising platform, Axon. Axon is the most powerful advertising platform in a generation. Axon allows you to capture undivided attention. Axon ads are full-screen videos that are watched for an average of 35 seconds, retention that blows other ad platforms out of the water. And you can launch in minutes. You set the goal, and Axon achieves it. No complex setup, no expertise needed. And Axon scales quickly. They can put your ads in front of over a billion potential customers. Other businesses have seen immediate results, scaled to hundreds of thousands of dollars of spend per day, and increased their revenue by millions. And most advertisers aren't even thinking about this channel yet. Less than 1% of advertisers have access to Axon, so you want to get started quickly. And you can do that by going to axon.ai/senra. That is axon.ai/senra.

I still don’t think I understand the total depth and breadth of your vision. So, if I come to you, and I don’t know anything about Roblox, and you’re the systems designer, and I say, “Okay, explain to me what you have built”—take as long as you want with this question.

If we had to talk super expansively, I would say, look, there’s a continuation in the human development of how we communicate. We said earlier it’s a bit like the holodeck. Right now, the holodeck that we’re building is very primitive. It’s not photorealistic yet. You don’t go inside of it yet. But you do go into a 3D environment, mostly that you could imagine, or think about, or build.

You can go there with your friends and play hide-and-seek. You can play a traditional game. You can be in a fashion show like “Dress to Impress.” All of the places you go in the holodeck are made by a creator community.

We’ve incorporated an economy into this system, so the vast majority of people who go do that do so for free. Our creators, though, are very thoughtful and savvy, and they come up with interesting ways to spend money. So, when we go into “Dress to Impress” and go into a fashion show, there may be something I can buy, if I so choose.

In a way, they are creators because they have this platform and are able to innovate new types of things that we think of sometimes as games or experiences. It’s funny that the top hits on Roblox recently have either been “Fashion Famous,” “Dress to Impress,” or “Grow a Garden.” It’s like, that’s pretty cool.

Okay, but explain how you built the world, though.

Behind the scenes, there’s a massive amount of technology in Roblox. To make it seem as transparent and easy as it does, there’s an underlayer we call the “Roblox Operating System,” which is the company that’s then building this Roblox thing.

Kind of in line with systems thinking, we think of our company as the system, and the company Roblox is really running almost as if it’s 9 separate companies. They are all very well connected. We all get together once a week and connect all these companies together.

There is a 3D cloud simulation and toolset company running within Roblox. There is a mini cloud infrastructure company running with 40 data centers and hundreds of thousands of computers.

Why did you decide to do that?

Initially, we were actually pretty naive. Rather than going down this cloud path, I would say right when we were starting Roblox, AWS was just starting to be a thing, and S3 was just starting to be a thing. We liked the idea initially of building it ourselves.

Five years later, I think there was a lot of discussion about, “This company’s all-cloud. This company’s all build-it-themselves kind of thing.” At that point, we started to see building infrastructure ourselves as being a cost benefit and a scale benefit.

I would say we’re great partners with AWS, GCP, and others. We do burst into those platforms. But at the scale we’re working and the cost we’re working, we do this very efficiently. All of Roblox runs for less than a penny an hour for every person around the world. That’s super critical.

Let’s go deeper. I want to go back to the 9 companies because this is stuff—this is what I meant before we started recording. I was like, “You’ve got all this crazy shit in your head that I haven’t heard in any of the podcasts. We’ve got to find a way to get it out.”

Yeah. So, underneath Roblox—Roblox Operating System—it’s how we run the company. We have a group of engineers working not on Roblox, but on the Roblox Operating System, looking at who the various teams are, how we’re organized, and how we do all of our things in the company.

Hold on, back up, back up. You have a team of engineers working on the Roblox Operating System, which is just the operating system for how to run the company—the company building the world?

That’s correct. Yeah. Because running the company efficiently is the most powerful way to then build this vision.

Why won’t you go into it?

I don’t want to say anything nonpublic, but I would say that is right in line with the Roblox Operating System. How does every engineer, product manager, data scientist, and designer work in optimal ways? How does the leader of an individual group, like that game engine and cloud group, actually operate almost like a mini CEO? How do they run that vertical stack? How do they control the types of engineers and product managers and compensate them?

That’s a really big part of how we run our company. I’d say that the common thread of what we try to do is have as much autonomy in these groups or companies within, and then simultaneously, horizontally, just continuously try to connect it together.

So, we have lots of groups who can go and do things very quickly on their own, and then we glue them together constantly.

How did you come to the design of the way you have the company operating?

There was a pivotal moment a long time ago at Roblox when we were running it as 3 horizontal stacks. We were running it as the web stack and the infrastructure stack, and what we would see is someone working on a specific feature, like the social graph. They would have to go and try to get enough bandwidth out of the web stack, the infrastructure stack, and the front-end stack.

So, there was always a negotiation among those, and it was hard to say who was building the social graph. There was a rotation of that where we said, “No, someone is going to be in charge of the whole social graph: user-facing components, web components, infrastructure.” All of a sudden, we got a lot of acceleration because they could load-balance between all of those pieces.

That lesson has been with us today, and it actually runs recursively. So, within, for example, this one group or company I’ve been talking about—the game engine group—that is subdivided into smaller pieces that also run kind of in that same way.

Essentially, you created a series of primitives?

Yeah, in a way. We have a primitive system for how we run the company recursively, from groups to teams to pods, and we’ve tried to make that somewhat of an accelerant in how we build the business.

But the way you think about these 9, you think of them as almost individual companies inside the company? Is that the term you use?

We call them groups, but with the leaders of these groups, we want them to feel that autonomy as much as possible—running a pseudo-autonomous thing that intersects with each other. It’s almost like we would have a game engine simulation company talking to an infrastructure, cloud computing, and storage company. They work with each other.

This sits on top of that. But we want as much speed and autonomy in each of these groups.

Say more about how this is organized. This is very fascinating.

No, we have an executive staff meeting every Tuesday. We get together for 4 or 5 hours. The leaders of all these 9 companies come together. We go through the things they're working on by themselves, but we also go through a lot of the horizontal things that are bringing them together.

What are the horizontal things bringing them together?

It's funny, but we track things at a ridiculously high level of fidelity, even in our executive staff meetings. We have probably 50 to 60 things that are core to hitting the objectives we have this quarter, next quarter, and the rest of the year. It's almost like a hand-curated list at the company level and group level of things that we think are emblematic of our success. Every week, we have all of the key leaders of all those groups come together, and we just go through that list of 50.

So, who's in this meeting? Is it these 9 people? Are they reporting directly to you?

Some of them report to me, and some report to someone else. We'll bring in a few other engineers as well, so there's probably 15 or 18 people in that meeting. We spend an hour just going through them rapidly: choo, choo, choo, choo, choo, choo.

Well, no, what's the “choo, choo, choo”? You have to explain that.

Key deliverable 1 through 50. It's like, “This key—”

In an hour?

It's 2 minutes. It's like 1 to 2 minutes per issue.

I just recorded with Evan Spiegel, the founder of Snapchat.

Yeah.

He was explaining his design meeting. He says, “We'll go over hundreds of ideas.” “And we just go through it really, really, really rapidly.”

We have our core list that goes between 50 and 70. The items cycle every month or 2 or 3 or 4. We're constantly putting new things in.

How do you keep track of all this?

We have a pretty good Office of the CEO team. We have the Roblox operating system, and there's some software in there. My office has StratOps people—senior people who work with me to do this on the side—so it's very real-time.

Okay, can you say more about these 9 separate companies? I think you just described 2 of them.

Yeah.

Or 3 of them.

We publicly talk about all of them. I would say there's a people systems company that actually builds the Roblox operating system and has all of our recruiting, people, and performance management stuff. There's an infrastructure company, and that's running our 40 data centers and all of our compute and bandwidth. There's a persistence group; they're building out cool persistence systems as well. The game engine, simulation tooling, and Roblox Studio sit in a group or a company.

There's a group called the economy group. It's literally all of the things around how we have Robux, how we monetize, how we collect money, the ledger—all of that. There's a very large group, arguably the largest: the safety and civility group, with engineering, product, policy, and live ops under a single-threaded leader, Chief Safety Officer Matt, who can run that whole thing and make decisions. There is a user-facing group that's all involved with the social graph, virality, and the various apps we have for having Roblox run in different places. So, there's a wide range of these user-facing groups.

Do all 9 of these companies have the single-threaded leader?

Some of them are still matrixed. Many of them are single-threaded. We do have some leaders that have engineering under 2 of them, so it's not absolutely perfect. Some of these are better run by maybe a product person than an engineering person. But generally, there's 1 or 2 leaders for each of these groups, and we know exactly who they are.

Do you think that'll be the case a year from now?

I think so, yeah.

Okay, so you don't see the need to change that?

No.

All right. So, let's go back. This is very fascinating. Essentially, you've designed a system—

That's right.

—which is your company, to design another system which is the—

That's right.

—platform that you're building.

That's exactly right.

I still don't understand the jump. So, within a couple of hours of letting people start building their own game, now you're turning users back into creators.

That's right.

But there's no economy there. There's no Robux.

Pure fun. Pure fun.

Okay. And there's no revenue for the company at the time?

No revenue. Pure fun.

Okay, so what's—taking it back to your decisions, you're seeing this like, “Oh, we're onto something. This is going to work.” What's your next thought?

What starts happening is that Roblox started in this persistently viral mode. We could track the growth day by day, week by week. It's like, “This is organic growth. This is not paid acquisition.” We're just sitting back and watching that. The fun started being there. It's just like 10 concurrent, 100 concurrent, 1,000 concurrent, 10,000 concurrent. Creations are getting better.

We know that everything we do with Roblox Studio—the better Roblox Studio gets, the more viral it is. The better the game engine is, the more viral it is. Very early on, to Erik's credit, I think literally in week 3 or week 4, in the afternoon, we're sitting there and we know the names of a lot of these people. There are hundreds of people. Erik said, “We need to build the first safety moderation system.”

John, Matt, and I all look at it and say, “You're right. He's right. Oh my gosh. We've got to get on this thing.” So, along the way, we build a moderation system. We have user reporting, all of that. For a long time, Matt, Erik, John, and I were the moderators. We would switch off so we could get a vibe of that.

This keeps marching on, and then finally we say, “Okay, how can we start making some money on this thing?” It's growing. We have costs, all of that. Interestingly enough, this is another example of— We had a big intuition around virtual economy. We said, “We can get this out a little faster if we build this club membership thing.”

At that time, Club Penguin had a club membership. Can we just charge people 5 bucks a month for something? We worked on that for a while. We had this thing called Builders Club. It was kind of fun. We started monetizing.

We had this really fun moment, just like when we started, where I was on a camping trip and I could check how many people were buying Builders Club every day. The first day it was 5, then it was 4, then it was 6, then it was 3, then it was 8, then it was 15. So, we're like, “Okay, we're monetizing. This is great.”

Builders Club kept going, and then for about a year or 2, because it was working so well—wow, it's all viral growth, and viral Builders Club is starting to work—we just kept working on the system, making it better and better. It was interesting, though. Just like in the early days when we started the puzzle game and then said, “Finally, the market has shown us that's not the ultimate system. Let's build what we really believe.” Something very similar happened with Builders Club.

We entered this interesting period where user growth was going like this, but revenue and bookings were kind of going like this. What that meant is that Builders Club was getting a little old and stale, and for the wider range of users, it was really not the right thing.

In retrospect, it's funny because when we started the company, we used taglines like, “You make the game.” That's a really good early tagline. It's aspirational. Everyone's making games on Roblox. It's definitely not the reality for 1 billion users. For 1 billion users, communication would be very important for many people without having to make a game. There are a lot of people who just want to show up and play something.

Yeah. I think right now you said the last number I saw was, I don't know, 150 million daily active users or something.

That's right. Roughly, yeah.

About 140. Whatever the last public number was. Out of those 150 million, how many are building games?

The 150 million?

I don't know if we give out our latest Roblox Studio numbers, but it's definitely a lot of people.

But it's not 10 million.

A small percentage.

It's 1 or 2 million. And I would say the good news is that vision may actually return because I do think, over time, with AI acceleration and all of the things we're working on—just like maybe in the very early days of video or YouTube, which was mostly consumptive, there was a little bit of early creation. Then with short-form video, everyone's starting to make videos. I do think there's that potential for something like Roblox.

Yeah, especially if they can do it in natural language now.

Exactly. Along the way, we had this revelation that maybe the reason Builders Club was getting tired is that not everyone wants to be a builder. A lot of people just want to have a lot of fun.

We also had a big intuitive vision. What would really be cool—another perpetual motion machine—is a virtual economy. People spend money; they buy Robux. If we can allow the creators to figure out all the different ways someone might buy Robux without hurting their experience or making it any less fun, then a small portion of users will buy some Robux. They'll spend it in the game. The creator will accumulate those Robux in their Roblox Robux bank account. They'll cash them out for real money, and that should create another perpetual motion loop.

Interestingly enough, we actually had a lot of board discussions: “Is that a good idea or not?” The board came around. I was probably the biggest pusher of it because I believe some people who were hobbyists would then be able to go and earn a living. You could go from hobbyist to earning a living on Roblox if you could do this.

We did that, very similar to the Roblox Studio situation. We had a full closed-loop system. The closed-loop system means I’m a creator, and I can make a game or an experience. You can buy something extra for Robux—it could be a flashlight or something.

As a user, I can go buy some Robux with a credit card. I can buy that. From a discovery standpoint, we started showing the top games where people were spending Robux. Within about an hour or two of going live, it was, “Oh my gosh, that’s the top game where people are spending Robux.” More people went into that.

And that’s going to scale with user growth in a way that the Builders Club never would.

Yeah. What we hit was an elegant economy that scales more than linearly with hours, basically. What we have found, while still optimizing generally for engagement and retention, not for money, is that developers have an incentive to balance engagement and money. If everyone needs to spend money to have fun, your game isn’t viral. Creators have figured that out, and so that’s been another perpetual motion machine.

Deel will help your business hire, pay, and manage any worker anywhere in the world. Deel is the best company in the world at building infrastructure for global hiring. Deel is one platform for payroll, HR, benefits, and device management across 150 countries. Deel gets you everything you need to run a high-performing global workforce on a single AI-native platform. From first offer to final off-boarding, Deel handles the complexity so you can stay focused on your business. The best founders and operators in the world have one thing in common. They control as much of their business as possible, and the founders of Deel do exactly this. When you use Deel, you aren't using a third-party payroll processor or a messy network of in-country providers. Deel built and owns the rails. That means faster speed, better service, and total accountability. The founder of ElevenLabs, who I use to make transcripts for this podcast, has a great description of the value that Deel can give your company. He said, "We built Eleven Labs to break down language and communication barriers. With Deel enabling us to hire and support exceptional talent anywhere, we can accelerate our innovation and bring more voices, stories, and ideas to every corner of the world." Deel is trusted by over 40,000 customers and growing fast. Learn how they can help your business by going to deel.com/senra. That is deel.com/senra.

You launched this second perpetual motion machine, this virtual economy and currency, how many years into Roblox?

Very early on. It’s got to be 15 years ago.

Okay. So, you’ve been—

10 to 15 years.

—building this economy for 15 years.

That’s right.

Is there another flywheel where it’s, okay, I was making games for fun. I’m a user. You turn me into a creator. That was all fun. Now you’re turning me from a creator into an entrepreneur. Therefore, I now make more money. It has to increase the quality of the games.

Quality’s gone up, up, up. I would say quality is ultimately going to be limited not just by the creativity of our developers, but also by something interesting about Roblox. In the video space, the camera and the display are getting pretty refined: a 4K camera, a 4K, 60 Hz display. That’s a pretty mature technology relative to, maybe, black-and-white movies with no sound 100 years ago.

In the 3D co-immersion space, we’re not there yet. We’re really excited because there are a lot of technical people at Roblox, and I think the ultimate spec is to see what human co-experience can be: huge crowds of people, 10,000 people—

But realistic.

Photoreal.

Yeah, because right now the games look shitty.

We believe we’re going to get to photoreal.

I’m not obviously insulting you.

We believe we’re going to get to photoreal.

Yeah. But isn’t it crazy how far you’ve gotten? You hit on something deep in our nature.

That’s right.

This fact that you don’t want to do things alone.

That’s right.

You don’t want to shop alone. The core bones of the platform—

You don’t want to play games alone.

—are very interesting. I think it’s almost like Roblox is very sophisticated and very technically complicated, but in a way, that’s the same traction one would’ve gotten from that early black-and-white movie in a movie hall without any sound and with some text between the scenes.

I just thought of something. The fact that you got this far with—

We got pretty far with black-and-white, no-sound movies.

If I was in your position, I’d feel very confident investing and continuing to reinvest into this platform, given how far you got. Could you imagine how big the platform could be if it’s photorealistic?

Can you imagine 10,000 simultaneous users, photorealistic, on a 2 GB Android device?

What’s the most concurrent users? You said that Bruno Mars was the most you’ve ever had on the platform, or no?

I think there are 2 types of concurrency. On Bruno Mars, we’re talking about 20 million people. But those are sharded amongst copies of experiences.

When we say the technical goal is, say, 10,000, that’s in a basketball arena. Running 10,000 concurrent people is really technically complicated. They may be all around the world trying to get together, and so when we say concurrency, it’s in the exact same environment where you can see and hear every single person.

This is why you keep saying that it’s a little secret that Roblox is actually an infrastructure company.

That’s right.

Okay. You have not yet figured out how to create the technology to do what you’re describing, what you want to do, correct? And that’s what you’re doing every day?

I would say part of what makes the job interesting and fun is that I think we have a reasonable idea of how we’re going to get there. There’s no timeframe—I’m not saying when Roblox will get 10,000 users, photorealism, all of that—but I think technically we have a reasonable idea of how we’re going to get there. That’s kind of exciting.

And very hard to compete with.

We run the company in a way that’s interesting. We’ve been through many crazy times, right? We’ve been in the PC revolution, the web revolution, the mobile revolution, the crypto revolution, and the AI acceleration revolution.

We have to run the company almost as if we’re imagining a virtual other company that really loves this space. We have an imaginary competitor, so we literally think about running the company that way.

Michael Dell said this exact same thing on this podcast. I think 5 years ago, he stood up and said, “There’s a company that’s going to compete in every single area we’re in. They’re going to be doing it faster and better than us. They’re going to be using the latest technology.”

That’s right. And he goes, “And we’re going to be that company.”

That’s right.

He believes that even if you don’t have a crisis, you make a crisis.

We have to look over here at that imaginary company and then be that company.

Very similar to how Dell thinks. I just did this episode on Elon, and he said the same thing. He’s like, “Even when it looks like we’re going to win, I always assume that we’re losing.”

That’s right.

To keep that mentality, I want to go back to this. You stumbled in, and now we’re a couple of years into Roblox’s history. Now you have this virtual economy. You can see that it’s going to scale way better than Builders Club.

Ho Nam, again, I want to read this tweet because I’ve got so much information about you from him. He says, “In the early days of the journey with Roblox, a next-generation Roblox received a $500 million seed round prior to launch. The company was called Improbable.”

Yeah.

He says, “Roblox used less than $10 million of equity to build their business to cash-flow breakeven.” Is that true?

Yes, it is.

Okay. And then, over the coming decades, you reinvested billions of dollars of free cash flow into this vision, unlike the competitor you’re talking about.

That’s right.

And then his point was that $500 million—that seed round of this competitor that died—is nothing over the long run.

I think they’re still around doing military simulation. I think the reason it was $500 million—I think that was at the peak of venture capital. I think it was Masayoshi Son or whatever.

Oh, God.

And that was just the check you get.

Yeah.

So, sorry, there’s no $100 million; there’s $500 million.

Talk about how you were so efficient: $10 million to get to breakeven. How did you do that?

I think we’ve always been very good at product management of the surface and at doing the right minimum viable walk through the space: do less, take the long view, know where we’re going, and get a lot of stuff on the way. I do think we’re pretty good at eliminating distractions, so we’re going down the right path. I think we do that pretty well.

What are some of the distractions that you avoided or eliminated?

Oh, man. It takes a lot of hubris to build a platform because I think it’s so fun to build content. We all want to build games and big, complicated features, and a lot of times, the features our creators need are boring and purely performance-based.

We have a saying in the company: “Performance is a growth feature.” We put an enormous amount of work into raw performance features, scale features, those kinds of things. That takes a lot of hubris.

What’s a raw performance feature?

We watch how long it takes on a wide range of devices from when someone clicks, “I want to play that experience,” to the time they’re interacting. The vision would be video. We’ve come to assume in short-form video that I can just scroll through—choo, choo, choo, choo. We’re used to that. Gaming has traditionally never had that.

Gaming has traditionally meant that you download, or you do this.

Loading screen.

You could have this background thing. I text you or something, “Let’s go try this.” Choong. “Okay, we’re in.” People actually like that without maybe knowing it. So, we just made the decision: We want to get the time to jump into any Roblox game down to zero, basically. That’s very technically complicated, but we do believe it has long-term growth aspects for us.

Okay, so you have this engine that is now making money.

Yeah.

Right? You are staying focused on just building this perpetual motion machine.

That’s right.

So, you’re not going on these other side quests. If I’m understanding what you’re saying correctly, then if you just limit your investment to things that make the user experience better, that will lead to more user growth, more money. Now you have this other kind of flywheel going on very early.

That’s right.

I’m trying to figure out—basically, the line of my questioning is just, how the hell did you only burn $10 million?

Yeah. We got viral very quickly, and then we started monetizing very quickly.

And then you were just very careful about what you used the company money for?

That’s right. And we did great. We had a secondary. We went public. There’s a certain amount of financial prudence where I think we didn’t spend any money, but we always made sure we had enough cash along the way. So, if we raised a secondary round, we might put some padding in there, but we actually never dipped below that initial $10 million.

I heard from Ho, and I think other people, that you were very careful when raising money. It was almost like they had to slowly commit. You kept saying, “No, no, no, no.”

In that sense, I think we did a really good job intuitively connecting with go-long VCs. We got very fortunate, whether it was Altos or Greylock or Index or others, that, in a way, we got VCs who could risk it all, who wouldn’t be like, “Oh my gosh, this is one of my first deals. What am I going to do?” They could really go along with us, and I think that was very helpful.

So, why was it important? You were just worried about dilution? Are you a control freak?

In some areas, probably yes. Hopefully in other areas, not.

No, but you’re pretty vertically integrated. You like to control everything. You build your own data centers, your rendering systems, your AI creation tools. You’re funded by operating cash flow.

I think there are areas of my life where hopefully I’ve decided not to be.

We’re talking about work.

But for Roblox, knowing what we’re building, I would say it’s not just control. I think, in a way, it’s owning our destiny. So, I would say building data centers is—

Yeah, but the only way you own your destiny is if you maintain control, right?

Exactly. So, data center cost, performance, control, yes.

What other areas of the business are you like that with?

Our game engine, for example. We always imagined we needed a multiplayer 3D immersive engine that does 10% of everything really well. Early on, it was like, “You should use that game engine. You should use that game engine.” We said, “No, that’s a critical core part of our whole platform. We have to build that ourselves.”

I think you just like building things.

We do like that, too.

It’s like, to go back to your description, “I built an operating system.”

That’s right.

Like, “I built the company, and then I built the operating system—

We do. Yeah, we do.

—to operate the company.”

Yeah, it’s interesting because we’re going to be building some other interesting things that we just decided on.

Oh, of course, but do you see where I’m going with this? It’s like, I want to tie that natural inclination or propensity you have for—

Yeah.

—we’re going to call it control—

Yeah.

—for the time being—to also being very careful not only about how much money you raised, but about who you raised it from. Is there a connection there?

There might be a thought around efficiency, growth, and control.

Because you can control your destiny.

That’s right.

Okay.

There are many components of our stack where we went exactly the opposite way, and we said, “We don’t need to control that.” In the game engine business 15 or 20 years ago, what audio system are you going to use? We’ll use FMOD. Everyone uses it. We’ll plug it in. So, we don’t build everything. When we feel we can plug that thing in and use it, we’ll go that way.

Brian Halligan founded HubSpot 20 years ago, and he has this line about AI that I keep thinking about. He said, "Most companies are using AI to make their teams more productive, but the companies that will thrive make the company itself the intelligence." And that is exactly what HubSpot does. HubSpot gives you AI that works, AI that actually knows your customers and your business. Your AI needs to know what you know, your actual customer conversations, your sales history, what worked last quarter and what didn't. HubSpot connects AI to your real customer data, so when it writes an email, it knows this customer asked about pricing three weeks ago, it knows what campaign brought them in, and it knows that they already contacted support twice this month, and that's when you start seeing actual results. Visit hubspot.com to learn more. That's hubspot.com.

Let’s go back to this idea that I’ve heard you say a few times: the dirty little secret of Roblox is that we’re actually an infrastructure company.

That’s right.

Okay, explain this.

Running Roblox behind the scenes is very complicated. So, when “Grow a Garden” hits 20 million people at the same time, there’s a lot going on there. That’s 20 million people all around the world. They’re playing on a phone, a tablet, a computer, iOS, Android, PC, Mac, Xbox, PlayStation, or Meta Quest. They are running translated versions of that experience.

They are hitting Roblox bandwidth very quickly. They are connecting to data centers in Poland, Singapore, or Brazil. They are connecting to our core databases as well. Running that is interestingly complex. Running a 3D simulation cheaply is very complicated. I said a penny an hour.

That’s part of why our infrastructure costs aren’t crazy: We can do a good job even in some countries that don’t monetize very well. So, we have core data centers, edge data centers, and we’re running a lot of our own CPUs and GPUs all around that. That’s what allows us to do that.

And you started this from the very beginning of the company.

We did, yeah.

Okay.

The very beginning of Roblox was running on 1 giant server that was right over there in my office.

Can you describe what you see as the flywheel of the world that you built?

I think the flywheel would be akin to the early phone system. Way back in that day, when AT&T comes out with the phone system, there’s probably a time when only 5% of people in America had a phone. But imagine the virality of going over to your neighbors and seeing them make a phone call across town. “Oh my gosh, I want to do that too. Look at that.”

And so, when we look at the phone system, it most likely grew pretty virally. There were some early adopters; they were probably really expensive, but over time, everyone saw that. I think that’s how we see Roblox going, a bit. There’s a lot of word of mouth. As the quality of the experiences grows on the platform, I think we can see correlation in growth there.

So, if we can technically support what we ultimately think should be on the platform, I think there’s potentially a lot of growth there.

Blake Robbins is a friend of mine. He hangs out on the edge of the internet. He always finds these weird things for me.

Mm-hmm.

He’s the biggest Roblox bull that I’ve ever come across. This is the way he describes it. I’m curious if you agree with this: “More creators bring more games. More games bring more players. More players spend more Robux. More spending attracts more creators.”

We’re ruthlessly pushing forward the quality of our technology stack, our economy stack, and our safety stack. So, I would say, I think what he’s saying is that if we walked away from Roblox right now and just kept the lights on, it’s possible we would keep growing for a while. But that, at the same time, would be very, very dangerous and something we would never contemplate.

Yeah, the way he described it to me is that Roblox is a compounding machine with network effects and a fully functioning economy.

Yeah, yeah. We don’t know what that full effect is. We’d never claim that we could walk away and have all of that continue. But I do think, simultaneously, we know what we believe is technically possible, and we’re literally racing to build that.

You mentioned safety a few times. Ho Nam, when we had dinner to prep for this, said something interesting. Essentially, it’s the biggest playground in the world, right? That’s one way to think about this.

That’s right.

The biggest playground in the world. Obviously, it’s the safest playground in the world, but it’s also huge, so there are going to be some kinds of issues that come up. But he was actually saying that, because of the investments you guys have been making for 7, 8, 10 years around safety and AI, way before anybody else was, you’re getting so good at it that it’s actually a moat.

Oh.

Because over time, it’s going to get safer and safer.

That’s really beautiful that he said that. We’ve seen this in self-driving, and we’re not going to claim we’re self-driving, but with self-driving, say it ends up being 10 times safer than the average human. When there is a self-driving incident in the news, you don’t hear that. It’s just like, “Ah, self-driving, ah,” kind of thing.

I do believe the pressure we get, in a good way, from the media, from this, from that, is an incredible motivator for that moat. The vast majority of what we’ve done on safety and civility, we’ve done on our own, in a visionary way.

I'd say age checks—we made the call on our own. It's not because of laws or anything like that. But the ultimate moat and the ultimate belief in what is going to be possible is that we're going to know the age of everyone. We have AI systems watching content and watching communication. We're banning things. We're not sharing images, all of that.

It's an enormous opportunity, and I think more and more we're starting to say, "Look, we know what the gold standard for safety is. We're building it. We're pretty far along." We're actually starting now to see other companies say, "Oh, maybe we should do what Roblox has done," or seeing more and more governments say, "We like where you're going with this. This is really cool." So I'm really optimistic about having this force of innovation in what we're doing there.

Okay, but is this something that you've been building and compounding since your co-founder—

All the way since we built the first moderation system.

Like the first year or two into that?

That's right.

Okay. Explain what—

That's right.

I don't know if you can talk about this publicly. What have you built to make it so safe?

I think we have seen AI become more consistent over time and scale more than humans, and so the ability to run every asset—everything on the platform—through AI, all content, images, everything everyone's building, everything everyone's saying, and just get better and better and better at that, it's been an enormous thing.

Part of AI is very much the human-facing thing, like, "Oh, I've got an AI prompt," or, "I can do all of that." Behind the scenes, though, at Roblox for many, many years, we were pushing very early AI—BERT models, primitive-type models—to drive safety. So I do believe it's a moat, and I do believe we've done something very unique in that.

What's going on out there right now is that we're very upfront: We do have people under 13 on our platform. It's an enormous responsibility. We focus, once again, on all of the aspects of how they play and learn and do things, but that's very different from being on some other platform that says, "Oh, yeah, it's okay. No one's under 13. We're good." So I do think leaning in long-term and being part of that is an enormous capability.

Something came to mind when you said that now other companies are looking at what we've done in this, and they're like, "Hey, I want some of that, too." Would you ever sell this kind of technology to other companies?

I think it's worth thinking about. I believe we will end up with the best AI tech for text, for voice, maybe for video, doing a wide range of safety things—either monitoring for critical harms, flagging, adjusting, all of that. It's not unreasonable to think we might do that.

We've already started open-sourcing some of our stuff. There's actually a consortium with various companies involved in safety. We've open-sourced one of our voice models and one of our Sentinel models, so we are starting to go somewhat down that path.

So the majority of the revenue that Roblox makes now just comes from this platform, this virtual economy, this world outside, correct?

That's right.

But you'd have to have a million of these giant model companies coming and wanting to train on your data, because you have 12–14 billion—what's the number of hours a month?

It makes it so simple, right? Like, we're—

But how many hours of—

13 billion-plus a month.

So, 13 billion hours of actual people interacting. What does this data set look like?

There are a lot of data sets out there that are video coupled with WASD. It's just like a human interaction model. What's cool is that because we're running this 3D simulator, and we're running it on our own cloud, and all of the experiences and games are running somewhat on the same simulator, we have really interesting data.

We have the 3D location of everything. We have how people are trying to move around with their avatar. We have, obviously, what they're texting and typing in a privacy-safe way. But what's fun about it is that we would, of course, never sell that data.

That's what I was going to ask.

Never, ever, ever, ever. For several reasons. One is just from a values standpoint of respecting the community. Can you imagine what kind of whacked-out decision that would be? "Roblox is selling their 3D user data." We would never do that.

Also, I believe over time that this is going to be very interesting data to, for example, imagine a future where a creator wants to make a really beautiful Roblox game and have a bunch of agents working all night long in iterative Ralph Wiggum loops—recoding, testing, recoding, testing. Imagine, just like we do with code now, a creator wanted to do their whole game that way.

So, wait, go back to you guys are buying 50 users a day from Google. Now, if you're building a game on your platform, you can actually use real data from actual users on Roblox to test what you're building?

There's a level of indirection there. I would say the—

What's the indirection?

The level of indirection is making amazing NPCs that look and behave and act like humans by training on that data from our users. So what we'd like to be able to do is, you're building your cool experience and you want 10,000 virtual testers. You want to describe how they act and what they do, put those into your experience, and use that to test overnight.

I like that idea a lot. That's very interesting. Your stock seems to be all over the place sometimes. Why do you think—like, you have these huge Roblox bulls, or it's just like this is a compounding machine—and then what do you think is most misunderstood about what you're building? Why wouldn't this keep getting bigger and more valuable in the future?

I feel like I can't predict the stock market, obviously. And I think, though, ultimately, it's showing that the future for many companies is—we're in a time right now where raw user growth and engagement growth is mixing with a lot of factors.

I think technical excellence, continued innovation, and having people understand where we're going with this—there's just going to be more and more of that, so I'm optimistic.

And you see AI as a giant tailwind for you guys.

It's an interesting time, right? It's a really interesting time. People are questioning all types of different companies: What companies are going to grow twice as fast? What companies are going to—this happened.

For us, it's so many areas. It's not just making games better and faster. It's increasing quality and reducing the time to market of experiences. And I do think, ultimately, AI will power some of what we've said around our ultimate spec.

Are there completely AI-generated games already on the platform?

I would say we're getting close. Right now, Roblox Studio is starting to embed a pile of AI-capable functionality. The beauty, though, is that we have all of Roblox Cloud open to MCP servers and all of that stuff. Now we're just seeing developers push at this with AI coding tools and creator tools and start to create that flywheel.

And these can be relatively small teams, even individuals, that make wildly popular games, correct?

Absolutely. Absolutely.

I don't even know why it's taken me so long to get into this part, because this is the part I'm actually most fascinated with—

Yeah, yeah.

—the size of the businesses that people are building on the platform.

I mean, DevEx is well over $1 billion a year. That's the amount—

Explain to people what DevEx is.

That's the amount of raw money flowing to creators on the platform. I'd say it's a really interesting time because, in the midst of AI, throughout the game industry as a whole, there's been a lot of perhaps legitimate concern among creative people that AI will displace what they do.

The thing we've said is that we believe the money flowing to creators is going to increase, not decrease. And so that's pretty good news. If that money flowing to the Roblox creative community is increasing, we should see even more opportunity for creators.

We've also seen something really interesting: Top creators on Roblox are making 10s, 20s, and 50s of millions of dollars. These are pretty serious—

I heard some of them are making that a month.

I can't say that, but I would say they're making—

They say that. Okay. I don't know if it's true, but I heard three kids in the middle of nowhere making $25 million a month.

I won't confirm or deny, but I would say one thing: When you hear that as a measure of platform health, if we look at the year-on-year growth rate of Creator 1, Creator 10, Creator 100, and Creator 1,000, Creator 1,000 year-on-year is consistently growing faster than Creator 1.

Wait, say that again.

The curve is flattening even more. The growth rate in bookings per year for Creator 1,000, ranked by how much they're making, is growing faster than Creator 1.

How?

The curve is flattening: wider use around the world, more opportunity for vertical content, and more opportunity for some content for older people. So that is a flattening of the curve, which bodes well for Creator 1,000.

What does the long tail look like at the very end? They're making a couple of hundred bucks a month? What is that, $10 a month?

I think the top 1,000 are, on average, making $1 million.

I heard it's higher than that. I think it's $1.3 million.

Okay. And then I think as we go beyond that, the curve goes way out to Creator 10,000 and beyond, where there's significant money being made out on that part of the curve.

Do you see a world—think about how, in the last decade and a half, when you ask people what they wanted to be when they grow up, "YouTuber" wasn't even on there 15 years ago.

Yeah.

Now it's always in the top 5 or 3, or whatever the case is. If you do have this perpetual-motion machine, you keep going, and I think once people understand the size of the business that they could build on this, and they have a love for what they're doing.

You're usually only really great if you actually love it.

I think being a Roblox creator could displace being a YouTube creator.

Who knows?

Why'd you make that noise? Before you said, “Who knows?” What was that noise?

It's funny because I get these short-form videos on various platforms of really cool Roblox creators living in this giant studio, driving these cool cars, and making short-form videos about how cool it is to be a Roblox creator. I think what they're saying is, if you do this, there is the potential to make a great business.

So, what do you guys do to foster that message or foster the size of the businesses that can be built?

It's pretty viral. People know. I think what we're trying to do is create, literally, like you said, the technical infrastructure so more and more of them can bring their ideas to reality. We've said publicly we want to get to 10% of global gaming on the platform. I think that means taking that black-and-white film projector and turning it into a 4K projector in the gaming world.

So, we think a lot about the responsibility of building the platform for them. The more photorealistic it gets, the wider the market gets. I think, in many dimensions—concurrency, performance, cost, photorealism, ease of creation, AI acceleration and creation—all of those things help bring ideas to reality.

But out of all of those, what is the most important, would you think?

I'd say it's well-balanced, actually. I think one of the good things we've done is, if I looked at that mix and looked at where our engineers are working, it's not like everyone is working on that one thing. We've got a balance of those things that are pushing forward.

Do you have funds coming in and trying to buy up these games and roll them up?

There is a market out there. I think, if anything, we're trying to get the message out to our creators. Our DevRel team can help you assess what your game might actually be worth, because a lot of them are worth a lot of money.

The platform itself has never tried to buy any of them?

We never have. I think what we want, though, is some transparency in the market, so early creators can actually know what kind of gem they're sitting on.

Have you seen people being taken advantage of?

Not really. I've seen amazing profit sharing and revenue sharing, things that have really worked out for buyer and seller. But I do think, over time, we want there to be a lot of transparency in that.

It's almost like you've built a game to play entrepreneur, and you turn them into real entrepreneurs.

In a sense, it's an entrepreneur game.

Yeah, if I wanted to teach—

That's right.

a 9th grader about entrepreneurship, Roblox might be the best place to start. You could literally teach them how to use AI coding tools, come up with a crazy idea, and get it live. On Roblox now, they could buy traffic. They could spend $50 to buy users on Roblox—the whole cycle.

Yeah.

I'd be very interested in that because—

Yeah.

when I talked to Tobi Lütke, the founder of Shopify, he said something interesting. He thought his job was to create more entrepreneurs. That's the way he looks at what he's doing.

I think that's right. I think that's right.

Yeah, I think it's more important than ever, especially with young people.

That's right.

If you look at their empathy, they're way more attracted to socialistic ideas than I think your generation was, and definitely my generation was, especially in America.

It would be fun to imagine an entrepreneurship class that kind of connected all the dots: I build this, do this, and can see this happening. That could be super educational.

What's your biggest threat that would inhibit you from building what you want to build, given this long-term view?

I actually feel the biggest threat would be not imagining that competitive company and not building what we think that competitive company is. So, the biggest threat could be complacency, rather than saying, “We can see what's technically possible. Let's build that.”

How often does this imaginary company come up inside the discussions you're having in your real company?

Now and then, I would say, given the speed at which we think some stuff is happening. Actually, it's interesting: the whole history of the company has arguably been a motion from quarterly things to monthly, weekly, and daily. The whole pace of our company, I think, has just been a historical acceleration of the pacing of the company.

The pacing at which you move?

The pacing at which we make decisions, the pacing at which we gently adjust without swinging the tiller widely, the speed at which we check in, and the speed at which we track things. It's gotten faster over time.

So, the way I would think about what you've been telling me is, you have this excessively long view. The fact that almost 25 years ago, you're like—

That's actually one of our values: “Take the long view.”

Okay. Well, 25 years ago, 20 years ago, you're like, “This is going to be the last company that I'm going to work on.” So, therefore, you even said, I think, early in the conversation, it's a 30-, 40-, 50-year kind of view.

Yeah.

So, an excessively long view in the room, basically. I think about this with an infinite number of daily iterations toward what you're trying to build.

Yeah.

A combination of those two?

Totally. The infinite iterations would be a complementary value, which is “Get stuff done.” And so we say “Get stuff done” means continuous iteration, but pointing in the right long direction.

Something we haven't talked about, but I think I heard you speak—I think that you said this in your Stanford talk, which I thought was interesting—that there were multiple near-death experiences or almost-death experiences at Roblox.

Mm-hmm.

There were several. I think one was the economy thing. If we had not figured that out, that would've been very dicey.

Looking back now, I mean, you were obsessed with science fiction. There were all these other virtual economies, right, at the time.

There were. Second Life.

Second Life.

There.com.

Yeah.

All of that stuff.

So, were you taking ideas from that as well?

There.com and Second Life really were early and in parallel to us. In a way, they were so early and so parallel, they made these very big visionary things. Second Life's architecture was a contiguous Earth kind of thing, which had some issues around scaling, because in a contiguous Earth-type situation, which is just the way the real world works, there's only one copy of that roller coaster, and when people want to go on that roller coaster, no more can go.

So, I think we were a lot more practical around scale. We're just like, “No, man, if there are 100,000 copies of that roller coaster, we're going to figure out how to route people to that, so we could have 10 million people riding the roller coaster.” But there were a lot of early visions out there, I would say, at the same time, for sure.

What was the argument from opposing board members to doing the economy?

We had a lot of fun going on the platform. We're just like, “Whoa, everyone's having fun building this viral stuff. Would we distort it if people could make money?” But the idea was that, look, in the traditional game market, you have studios with hundreds of people. We could probably get better quality if we nudged in that direction.

Another near-death or rough thing was that our economy got hacked once, so we had to shut down the whole economy for 2 days, literally. That was somewhat scary, and the reason is we had just been so early, moving so fast, that the way our economy was working at the time wasn't like double-entry bookkeeping and journaling and all of that, and we had no fraud control anywhere in the economy.

So, this was a famous economy hack very early in the days of Roblox, where the second the economy was hacked, money could bounce from place to place very quickly. We caught it pretty quickly, and we just said, “Shut everything down.” We had to shut Roblox down for a while, then bring it back up without the economy on.

Luckily, we caught it pretty quickly, and only a small portion of the money had moved around. But we had to run for several days with no economic activity.

Were you already a public company when this was happening?

No.

Okay.

Whoa. No, no, no. The SEC does some really good stuff, and I would say some of the good stuff—if we go through a lot of the controls in the SEC—those were things we did 15 years ago, as far as this economy, backups, and all of that.

When you were designing this, were there any books you were reading, or were there any examples? Were you thinking, “Okay, I'm going to take an economy that I see functioning well in the real world and just make the virtual version”? How did you even come up with the set of rules that you had?

I would say all 4 of us founders were very into high-integrity systems: what's the amount of float in the economy, how much currency we have, and do you trade it or not trade it?

We saw very early on in our economy that we had something really interesting. In the early days of video games, people would have multiple currencies, and we had something that, in retrospect, the Roblox community really loves, but isn't really a good idea, I don't think. That is a participation-based currency called “Tickets,” as well as a money-backed currency called “Robux.”

The thing that came out was that when you have a participation-based currency, you get 1 ticket for every day you log in. That's really not a very good way to have people log into your system. It's much better if they just love chatting with their friends or playing games.

The other thing is, any currency that a user can get from working or logging in will immediately be botted now. And so, we could see people trying to bot that currency all day long, and we had to ultimately get rid of it. So, we had some good learnings with that.

But ultimately, our current economy, as I've said, is scaled better than linear with user engagement.

You mentioned there's an advertising business inside of Roblox.

That's right.

How long has that been going on?

It just started, pretty recently. But if you go onto the Roblox homepage, you can see that some of the experiences say “sponsored” right now. We have a long-term vision of what percent that is. But it's actually really helpful because most of our discovery is primarily organic. We've been very transparent with discovery, and one of the things that Roblox creators like is that we share all of the signals of what's boosting something.

It's almost as if we were YouTube, TikTok, or Reels saying, “Here's all the factors that bring you up here.” But for some creators who want to have that entrepreneurial experience—buying 50 users very early, when their game isn't viral yet—they can use that sponsored thing, just like when we bought traffic from Google.

I got one more question for you.

Okay.

We're almost out of time.

Yeah.

Are you jumping up? Do you want to leave already?

No, I'm stretching because—

Okay.

I want to give an expansive answer.

Oh. Why do things like this? Why are you doing podcasts?

I do feel there's a lot of depth to the way we run the company, and a podcast is the format to get it out. I love podcasts. In today's media, if I know what the podcast is, that's all I need to find great content.

I like podcasts because they're typically not edited. They're not showing up in a discovery mechanism, so I feel they're one of the truest forms of media.

Okay, I'm glad you used the word “depth,” because what I want to do, since we're out of time now, is run this back as much as you want—maybe every 6 months or something like that.

Yeah, yeah, yeah.

Because I do these selfishly.

Yeah.

I'm obsessed with making podcasts, but I'm obsessed with entrepreneurs. Something Tobi Lütke said to me that I thought was fascinating—he said, “There's not one right way to do something in a company.”

Yeah. That's right.

There's probably like 100 right ways. And you have to figure out what makes sense for the context you're in, and then who you are as the founder, and what you're trying to accomplish.

I have a better understanding of you—way better understanding than when we started the conversation—

Yeah.

Even though I've listened to every single interview you've done. Your company is very misunderstood, and there's not a lot about you as a person. And yet I find these people whose opinions I respect very much, and I'm like, “This guy is special, and the way he's building his company is interesting.” I feel like I just—

I'll take that as a compliment.

Scratched the surface. You should take it as a compliment.

Thank you.

You built something amazing.

Thank you.

So, thank you very much for the time. I hope you accept the future invitation.

Great.

Every time we have this conversation, we can just peel more and more layers of the onion of what you're building and why you're doing it. I think it'd be very fascinating. It'd be fun to peel another layer. All right, perfect. Thanks, man.

Thank you.