Inside fal’s $1.65B Series E & F Led by Sequoia & Bessemer
Molly O'SheaBurkay GürGorkem Yurtseven
- Generative media has the spending intensity to sit alongside coding agents as a major AI market. The founders describe individuals spending $10,000–$20,000 a month generating content. Burkay recalls Veo 2 charging $5 per five-second clip against earlier models at roughly $0.30–$0.40: better quality unlocked professional buyers and what he calls “essentially infinite demand.”
- fal’s changing revenue mix challenges the assumption that video simply replaces images. Video climbed from zero to around 70% of revenue before image editing revived image demand, bringing the mix back to roughly 50/50. The founders’ premature conclusion—“image models were done”—missed editing’s usefulness both independently and in preparing frames for video generation.
- Advertising offers a compelling feedback loop, although the complete automated workflow remains unfinished. Molly raises the mismatch between effective 60-second ads and short generated clips; Burkay responds that quick cuts and imperfect, homemade-looking content can suit UGC advertising. Generating variants, measuring engagement and automatically favoring winners is the opportunity, but “no one has done it perfectly.”
- The founders’ margin thesis favors businesses that innovate in the compute path. They argue application margins are “close to zero” because competition and marketing consume the economics, reversing the conventional expectation that infrastructure commoditizes while applications retain value. fal’s role includes anticipating model launches and securing scarce inference capacity before demand arrives.
- fal is testing whether a roughly 150-person company can scale through two reporting levels and hands-on technical leaders. Its “Super ICs” combine engineering, leadership and commercial judgment; Gorkem estimates AI makes the engineering team three to four times faster. Burkay says a flat structure may hit communication bottlenecks and require more management; Gorkem thinks coding agents may let Super ICs lead five to 10 engineers with two levels of hierarchy as long as the models keep improving.
- Open models retain practical value even as fal expands its closed-model partnerships. Studios want fine-tuning, LoRAs and control over unreleased characters and styles; open releases also give trailing labs an “incredible marketing tactic.” Across both model types, customers’ confidence that inference runs on fal’s GPUs matters because confidential creative material is often the input.
- Growing Hollywood revenue has yet to translate into the labor disruption the founders hear predicted. Burkay claims “zero impact” on Hollywood labor despite the industry becoming a meaningful revenue contributor over the previous six months; legal approval remains an adoption constraint. His next-year milestone is “timeless AI content,” enjoyable after the demonstration of technical capability loses its novelty.
- World models could create another compute-intensive market, although product-market fit remains a possibility. Gorkem moved from skepticism three months earlier to enthusiasm about faster, more persistent interactive worlds. He says gaming might become a fourth category, with a GPU running continuously while someone plays.
1. Quality gains expand spending without making images obsolete
- Gorkem’s market map has three product-market fits: search and consumer assistants, coding agents, and generative media. The distinguishing behavior is individual spending: someone at an AI-native or conventional studio can spend $10,000–$20,000 a month generating videos, comparable to heavy coding-agent users. “This is the only other use case in AI that an individual can spend that much money”; search gets a pass because of its scale.
- His first reversal concerns revenue mix. Video climbed from zero to around 70%, prompting the assumption that “image models were done”: images seemed too simple and insufficiently compute-intensive. Image editing then revived demand and improved video workflows by creating frames for interpolation. He now puts revenue at roughly 50/50, overturning their expectation that video would completely take over.
- Burkay’s pricing example: clips started around $0.40 per five seconds and settled at maybe $0.30 before Veo 2 arrived at $5. “Who the heck is gonna pay” was their reaction. Higher quality brought in advertising, marketing and movie-production buyers whom cheaper models could not serve. Usage rose alongside prices, producing what he calls “essentially infinite demand.”
- Gorkem says the best video model currently produces maybe 10–12 seconds, with newer ones potentially reaching 30 seconds. Until generation reaches real-time, high-definition streaming, he expects continued expansion. In his explanation, duration increases inference attention and resolution increases pixel counts, making costs rise “exponentially.” Burkay says this market is “nowhere near” diminishing returns.
2. Professional demand runs ahead of finished video workflows
- Molly’s pushback comes from her AppLovin conversation: the best-performing ads are about 60 seconds, while generated clips can be around five seconds and do not fit every brand. Burkay responds that an ad need not be one continuous generation; quick cuts can work better in certain cases. UGC-style ads also tolerate imperfect lighting and disposable output: “If it doesn’t work, make a new one.”
- The larger opportunity is a feedback loop: publish variants, observe engagement, make small changes and automatically show more of what performs. Molly offers a concrete example of that testing mindset: raising her arms as a hook reportedly got her 10 times the views; she also tests thumbnails, titles and copy. “No one has done it perfectly,” Gorkem says. Burkay adds that sophisticated creators and some large companies, including fal customers, already do versions of this. The reinforcement-learning and technical work remains difficult; he thinks it becomes the norm “in a few years.”
- Sora illustrates demand alongside operating constraints. Burkay says fal saw huge growth during the last three months before it was taken down; Molly calls it too expensive, and Burkay agrees. Gorkem says they saw substantial volume, while the explanation that there wasn’t enough compute for other products was “the rumor”; he also says Sora was expensive to run. He thinks it had a future: “Maybe it’ll come back sometime.”
3. fal kept its compute thesis and changed its market
- Burkay dates fal’s start to 2021 and a thesis drawn from Snowflake and Databricks: valuable data infrastructure businesses owned the compute. Their initial bet was that Python would follow SQL into elastic cloud workloads because AI was next. They served data-science and machine-learning teams before ChatGPT, Stable Diffusion and DALL-E forced a strategic choice.
- With five or six people and limited money, competing in open language models meant joining a price race they could not finance: “We can’t really go and sell tokens at a negative margin.” Instead, they accelerated image models on existing infrastructure. Burkay acknowledges they had no fully formed thesis about where those models would go.
- The commercial signal settled it: the earlier business ran for a year without making a dollar, while image-model revenue started growing. Roughly three to six months later, they recognized different use cases and buyers, especially ads and marketing, and put “the generative media company” on the website. Within about 18 months, people were discussing feature films, an application they had not initially imagined.
- The name FAL comes from “Features and Labels,” old machine-learning terms; “fal” also refers to having a fortune told from Turkish coffee grounds. The founders describe fragmentation across image generation, editing, video, 3D, text-to-speech and speech-to-text, with perhaps five to 10 companies per subcategory and frequent leapfrogging. Their explanation for continued progress includes abundant video data: nobody claims to have trained on all of it, and selecting interesting footage is the hard part. Modeling physics and the wider world also leaves multiple directions for improvement.
4. A flat company needs engineers with commercial agency
- Burkay’s recruiting model comes from joining Coinbase in 2016, when it had almost 100 people; fal now has about 150. Coinbase’s Reddit crypto quiz screened for knowledge before a phone call. fal similarly wants people consumed by the subject: “If you didn’t hire them, they would be doing this anyway.” Early engineering hires also came heavily from trusted former colleagues.
- The organization has at most two reporting levels and builds teams around “Super ICs”: leaders who remain excellent hands-on contributors. Burkay says this approach has a limit: communication bottlenecks may force more management layers, and they are watching for when the structure breaks. Gorkem thinks coding agents may let them maintain two levels, with Super ICs leading five to 10 engineers while agents do much of the work, as long as coding models keep improving at this pace.
- Gorkem says agency includes commercial responsibility. Because customers switch models so quickly, fal wants engineers to work with customers and make business decisions—a broader remit than the software-only roles he describes at large companies. He looks for independent action, responsibility and ownership of decisions, but says these are hard to assess without having worked with someone.
- Burkay pairs hunger with humility and uses “reverse selling” to explain the workload: crunch periods can mean 60 or 80 hours. He rejects toxic work hours: “We go home during weekends,” and not doing so is “nothing to brag about.” He also wants technically exceptional people who are low-ego, friendly and easy for customers to work with.
5. Compute planning and customer trust anchor the platform
- Burkay describes fal as infrastructure partner to model providers and API partner to applications and studios. Customers use multiple models and often synchronize product launches with model releases. A successful lab may not procure enough inference capacity itself, so fal forecasts demand and invests ahead. Otherwise, scarce compute can go to coding agents or whoever commits more capital; coordinating both sides has become an ecosystem responsibility.
- On open versus closed models, Gorkem expects a swinging pendulum: trailing labs gain attention by open-sourcing, while studios have technical reasons to want it. He calls open source “the only way” to fine-tune, apply LoRAs, change lighting and use unreleased characters or styles. fal initially feared closed models because it mainly ran open ones; partnerships serving closed-model inference now give it exposure to both.
- Molly frames the broader issue as protecting IP across media, source code and customer lists, and notes that local execution can help. Burkay says customers trust inference more when models run on fal’s GPUs, open or closed, because unreleased content is often an input. Open models provide a self-hosting option; customers choose fal when they believe it can run them better.
6. Hollywood adoption is early, with lasting content and games ahead
- Burkay’s Hollywood claim is sharply bounded: business from the industry reached a meaningful share of revenue over the last six months, yet AI has had “zero impact” on its labor and barely produced worthwhile Hollywood content. He calls four years of displacement discussion fearmongering while expecting change ahead. Adoption requires company lawyers to approve the workflow as well as models becoming good enough.
- His next-12-month milestone is the first “timeless AI content”: work people enjoy equally when revisiting it later. Existing clips mostly demonstrate capabilities. His benchmark is Toy Story, which he dates to “1996, something like that”: computer-generated work that remains enjoyable. He thinks the next year will bring a widely accepted work, maybe a short movie or animation, whose appeal survives its technical novelty.
- Gorkem’s change of mind concerns world models: “I was very skeptical even three months ago,” but recent progress made him bullish. These are interactive worlds navigated with keyboard actions. Better speed, quality and architectures allow more persistence: previously, you could see a flower, look away and return to find it gone. Keeping it there is his concrete example of improvement.
- He thinks prompting a playable game could become another product-market fit: “There’s a big chance that may happen,” and “it might be this fourth category, maybe.” The constraint is continuous compute: playing requires a GPU running all the time. Molly responds, “Very expensive.” He identifies compute as the big challenge even as he expects game quality to improve.
7. The margin thesis follows the compute path
- Gorkem answers AI-capex skepticism with internal ROI: he estimates fal’s engineers move three to four times faster and would have required roughly four times the people before AI to match today’s pace. The company pays less than that extra headcount would cost. He urges measuring actual users rather than relying on reports or nontechnical commentary, describing himself as “long compute.”
- Burkay’s framing is that “every big company is gonna be a compute company,” with different storefronts: coding agents, data platforms or fal’s media offering. Behind those interfaces, he says, revenue follows who runs the most compute. He also says headline figures such as $200 billion in CapEx include capacity not yet built; construction is slow and current supply remains insufficient.
- A founder’s margin hot take reverses the old infrastructure-versus-applications consensus. The argument is that application margins are “close to zero” because competition and marketing absorb the economics: “You can only create margin if you’re in the compute path and can innovate in that path.” Molly cites Tony Kim’s observation that chips have better margins, and both founders agree. The speaker’s instruction to investors: “Wake up, VCs.”
8. The closing bets span consumer consolidation and programmable media
- Burkay revisits a year-old OpenAI–Meta merger idea: Anthropic has established an enterprise position, Google has made its consumer move, and he no longer sees OpenAI clearly winning either segment. Meta also needs something; combining forces could strengthen its consumer position. He stresses that he has no information and is “completely speculating.”
- The other speaker says a merger would be bad for the whole industry, but does not elaborate on the mechanism.
- A newer closing bet is that coding is also part of generative media. Developers on fal increasingly make richer applications and programmable video, including websites that feel like videos, so the company may have to invest because customers already want it. The shared closing line is “Everything’s a computer.”
Full transcript
You guys have the coolest brand on Earth. You’re growing really fast, and you keep raising more and more money.
AI really has 3 product-market fits. One of them, we believe, is generative media.
3 years ago, no one really understood what that meant. 3 years down the road, and it’s a whole new industry.
You sit down in front of a computer, generate videos all day, and spend $10,000 or $20,000 a month.
And we were like, “Okay, who the heck is gonna pay all this money?” Turns out there’s essentially infinite demand.
Every big company is gonna be a compute company; just the storefront is gonna be different. Our storefront is unique and interesting. No one else has a storefront similar to ours.
Were you surprised by how much money there is behind this?
100%.
Okay, cool.
All right.
Gorkem.
Wow.
I already—
You don’t need to. You really don’t need to.
Gorkem.
Yeah.
Gorkem.
Yeah.
Wow.
Try not—
That last one was very—
That was the best one?
Very good.
Yeah.
Gorkem.
Yeah.
Gorkem. Now I’m just gonna keep—
Yeah.
…on saying it.
Yeah, that’s perfect.
Teaching AI.
Or my camera, internet. You’ve been pronouncing Gorkem completely wrong. It’s Gorkem. Burkay.
AI, learn from—
Burkay.
…learn from the YouTube videos.
Burkay.
Yeah.
Yeah.
Perfect.
You guys, this is great.
Easy.
You guys have the easiest names on Earth.
Easy.
Uh, welcome to Sorcery.
Thank you so much.
And we’re in a very special place, so do you wanna maybe tell us where we are?
Sure, yeah. We’re at Dalida. Shout-out to Sayat and Laura. Dalida is an Eastern Mediterranean restaurant with a lot of Turkish influence. Fun fact: I got married here, and Gorkem officiated it.
Great wedding.
Congratulations.
Thank you.
Amazing. Well, I’m excited. I can already smell the food. Everybody’s eating it behind the scenes, so as much as this conversation is gonna be rich and delicious, I’m sure the food is as well.
It’s amazing.
I think I’ve reached out to Michelle like a billion times because your name kept popping up. You guys have the coolest brand on Earth. You’re growing really fast. Every single time there’s a round—and we cover funding rounds—I’m like, “I gotta get them the next time.” No, you’ve already raised it. It’s gone, it’s over. The next one: gone, over. I keep trying to catch you guys, but you keep raising more and more money. So congratulations on the recent round.
Thank you.
Thank you. Thank you so much.
You did a couple of back-to-back rounds, like any fast-growing AI company these days, but this last one is the biggest one yet, so thank you so much.
I think you’ve also hit the point where you don’t wanna do any media on it anymore. You’ve passed the inflection point of the hype cycle, and now you just have a humming business.
Exactly. We’d rather focus on the business. The announcements give us some push in the beginning, but as you do more rounds, I think they’re irrelevant. It’s the business that matters more.
The adrenaline has worn off.
Exactly.
It’s probably worked for hiring and that kind of stuff, and for customers, but you gotta focus. Okay, so let’s talk about the evolution of the business today. Generative media has taken off in AI. It’s probably one of the coolest parts about AI, seeing all these videos and all the creative stuff getting made. I don’t know who wants to take it off, but feel free to talk about where we are right now and why this has just kind of exploded.
The way we talk about it is that AI really has 3 product-market fits. One of them is search, like the initial ChatGPT consumer product, and now Google and Gemini. The other really clear one that’s growing really fast now is coding agents. The third one, we believe, is generative media.
To put things in perspective, with coding agents, an individual can spend maybe $10,000 or $20,000 a month on their Claude or OpenAI subscription. Generative media is the only other thing where this exists. If this is your job, if you work at an AI-native studio or a real studio these days, you sit down in front of a computer, generate videos all day, and spend $10,000 or $20,000 a month. This is the only other use case in AI where an individual can spend that much money. Search is so big that it gets a pass, and there’s no other use case or part of the industry that has this dynamic. It’s just coding and generative media, and that’s why you’re seeing all this growth: more people are adopting it, and individuals are spending so much money on it.
I know this is an elementary question, but were you surprised by how much money there is behind this?
100%.
Okay.
We were, too. We thought—we were surprised all along the way. We were surprised by how quickly we switched from image models to video models. Now, I would say, we make 50% of our revenue from video models. After that transition, we thought video was gonna completely take over. It was at zero, and then it reached around 70% of our revenue. We thought image models were done, that there wasn’t enough compute, that it was a simple problem, and that from then on it was all gonna be video.
Then it turns out image-editing models brought life back to image models. They’re also useful for video models because you can create different frames, and video models do a much better job interpolating between those frames. Image editing all of a sudden made image models come back, and now our revenue is 50/50. Even that was surprising. We thought a portion of our business was dying, but it made a huge comeback.
Video models keep getting better. Currently, the best video model maybe creates 10 to 12 seconds of video. The new ones are gonna reach maybe 30 seconds. But until you can generate real-time, streaming, high-definition video, this is gonna keep increasing. As you increase the resolution and the duration, the cost increases exponentially because you have to have more attention during inference, and the resolution increases the number of pixels exponentially. Things are getting much more expensive, so now we have a framework for thinking there’s a lot more money to be made going forward, because the video output we get is still very rudimentary.
There was also a very interesting point sometime last year when Veo 2 came out. I think at the time Kling was probably one of the best video models, and there was also MiniMax. The prices were somewhat going lower. I would say they started at, let’s say, 40 cents per 5-second video, and then they settled around there, maybe 30 cents for a 5-second clip.
Then Veo 2 was, very surprisingly, $5 per 5-second video. So it jumped more than 10 times, and we were like, “Okay, who the heck is gonna pay all this money?” This is just another 5-second video. Turns out there’s essentially infinite demand, so people started using that model, and then even more expensive models came around, and people started using those models.
What was really happening under the hood was that we saw more and more professional use cases start happening for ads. Real people in marketing and advertising, and in movie production, started using these models because the quality went so much higher, and there was more demand that other models weren’t able to capture. The usage is going up, the capabilities are going up, and everything’s getting more expensive over time. So, yes, people ended up spending so much more money.
In terms of progress, it’s still happening. We’re nowhere near— with LLMs, I think people talk a lot about hitting a wall or diminishing returns. On this side of the market, we’re nowhere near actually hitting diminishing returns. There’s just so much new stuff happening with Seedance 2 and 2.5. The quality’s going up, prices are going up, and people still wanna use all these models.
For sure. Do you guys remember Sora?
Yeah.
Yeah. Feels like it was a lifetime ago.
I know. No, I mean, that was a really cool moment, but clearly too expensive.
Yeah, it was too expensive. In fact, we also hosted Sora on our platform, and we saw huge growth in the last 3 months before they decided to take it down.
We were like, “What’s happening? There’s essentially infinite volume here.” But it turned out the rumor was that there wasn’t enough compute for everything else. It was also expensive to run. But I think it had a great future. Maybe it’ll come back sometime.
It’s really interesting because earlier today we were down at AppLovin. This will probably get released later, but I was talking to Adam, the CEO, and Gio, their CTO. We were talking about how they’ve created their algorithms for ads and what works best. The best ads are actually 60 seconds long. I asked, “Are you thinking about using a company like Higgsfield, or any of these other companies, and offering that—or acquiring one of these generative ad platforms?”
Mm-hmm.
Some of those are really—most of these cases are ads. It’s just really not there yet. The best they can do is 5 seconds, and not every brand can do that.
Yeah.
In terms of where you guys came from and where you’re going, do you think advertising is going to be the biggest use case?
I like ads as a business. The internet runs on them, and the content these models can make is perfect for UGC-style ads because the content is usually throwaway. If it doesn’t work, make a new one. It works great. The lighting doesn’t have to be perfect. It needs to look homemade, so it’s easier to do.
The best ad may be 60 seconds, but you don’t have to do a continuous 60-second generation. You can do fast, small cuts, which may work better in certain cases. Then you can get a lot of data back, right?
Mm.
I don’t think anyone is doing a particularly good job at it. But because you can post these ads and see how people interact with them, you can keep improving them, or make small changes to the—
Totally.
—to the—
Yeah.
—to what you’re posting, and show more of the things that are performing well. This can all happen automatically, right? I think it works great with video models. No one has done it perfectly, but we’d love to be part of this automated ad story. I think it’s going to be a great business.
It was really funny. I was going to mention this in the interview, but then I realized it was probably kind of inappropriate. The example you just mentioned of reinforcement learning models optimizing themselves over and over again—there’s something I don’t know if you guys see in your Instagram feed, but I do this because I make clips and stuff for the internet. People say, “One of the best hacks I had was showing my armpits. I got so many more views from showing my armpits.”
As the hook, right?
Literally. I got 10 times the views just from having my arms up. I was like, “What the hell?”
Yeah. Imagine you could have yourself in 100 different positions, and then let’s go.
Yeah, but that’s a thing.
Yeah.
You’re always testing the thumbnail. For YouTube, we have to test the thumbnail, the title, the copy, all of that kind of stuff, because it works the algorithms differently.
Exactly.
Yeah. I think there are really sophisticated users doing it on a smaller scale—maybe they’re creators—but large companies are doing it right now too. I don’t think it’s widespread yet, but it’s out there for sure. We work with a few companies that actually do it, but I think it’s going to become even more available. It’s not very straightforward to do the whole reinforcement learning process and all the technical stuff behind it. It’s not fully there yet, but I think in a few years this is going to become the norm.
Let’s talk about the engine behind fal, and then we can get into how your culture is keeping up with the growth, because it’s crazy. You were saying you started off with compute. You started off as a compute company and found your way into the AI space. What was that evolution like for you, and how have you completely embraced this new category?
Right, yeah. We started in compute, but in a pre-AI era, let’s call it. We started in 2021, and the thesis was: the previous data infrastructure wave led to amazing companies like Snowflake and Databricks. What do they have in common? They own compute. So, in order to build the next generation of data infrastructure company, we also had to focus on compute. That was the initial thesis, and we thought the next big thing was going to be Python.
Basically, what Snowflake and Databricks did, to a certain extent, was enable SQL to run in the cloud. They own the compute and let you scale it elastically. We thought Python was going to be the next thing because AI was the next big thing. We ended up working with some really awesome teams, trying to bring their data science and machine learning workloads into the cloud.
Mm.
That was our initial starting point. Soon after we started, some insane things happened in the market. From a technology point of view, we had a massive breakthrough with ChatGPT, and in our space, Stable Diffusion and DALL·E. We had a choice at that point: we could stick to our initial data infrastructure idea, or go into this new world of AI.
There were language models, and at the time people were very excited about a bunch of open-source models. If you remember, LLaMA came out shortly after, and there was a huge price race to the bottom. We thought, “We’re five or six people right now. We don’t have that much money in the bank. We’re a pretty early company, and we can’t really go and sell tokens at a negative margin. Let’s do something more interesting.”
The trend still held, so we focused on image models because there was something interesting happening there as well. There was no sign of video models. No one even talked about them back then, but we thought we could take image models and make them run a lot faster in the cloud, on the same infrastructure we’d built. So we did that. We had a very technical team, and we thought we could do a really good job running these models faster. I don’t think we had a very formed thesis about where those models were going to go.
We kept doing it, and that got a lot of attention from very early customers. We thought, “With this other business, we ran it for a year and didn’t make a single dollar. With this one, our revenue is starting to go like this. There’s clearly something here.” So we decided to focus on the image model space.
Three to six months later, roughly, we realized that this trend was probably much bigger. We thought these models were useful for very different things than what LLMs help with. We thought they’d help with ads and marketing. The buyers were going to be completely different. So we decided to call it something else, just to be able to tell the story to people. We came up with the term “generative media,” put it on our website, and said, “fal, the generative media company.” We always had to explain what it meant because three years ago, no one really understood. That’s how we got into it and found our niche, which, three years down the road, is more well-established. It’s a whole new industry.
And people want to make feature films out of it.
Exactly. Yes, yes. I don’t think that when you first play with DALL·E, you’d ever think, “Okay, we could make movies with this,” but it took only—
It came here very, very fast.
Yeah.
Yeah.
We came here really fast. It only took about 18 months for people to start thinking, “Oh, this could actually make feature films.”
We’re in San Francisco. People here are super nerdy. Maybe you guys are too, but as you’re growing super fast during this inflection period, how are you keeping ahead of research? Are you paying attention to research? How are you learning as you go?
In the beginning—
Because it’s new.
We were extremely involved, obviously. We had a very small team, and we had to be part of every single little edge we could make. Nowadays, we have a really talented leadership team. Most of them we worked with before at different companies. I would say that's the best part of our job: we can rely on these leaders.
That's a superpower—to ask their opinion about what's coming next and what they're paying attention to. A lot of the information I get these days is from that team. I'm also very curious and read everything, but I learn more from the leadership team than from the internet these days.
Oh, that's pretty good. What about you?
Yeah, I think, just to echo what Gorkem said, it's very important how you build this team. I got a lot of inspiration from my days at Coinbase in terms of recruiting the early team.
Mm.
One of the things they did there—I joined Coinbase in 2016. The team was almost 100 people, maybe. It was very early days, similar to our size now. We're about 150 people today.
One of the things they did was post a crypto quiz on Reddit. They assessed your knowledge about crypto: how much you knew about it, a bunch of technical terms. This was pre-AI, so it was hard to find the answers to these questions. There was literally a quiz, and you had to pass it to get on a phone call with someone on the team.
The question was, “How crypto-pilled are you?”
Oh my God.
I think this was a very key factor in Coinbase's success in the long term, because you really want fanatics about the space so they can keep things going. Most of the time, the whole job becomes about staying up to date.
Yeah.
How do you stay at the forefront? You could collect really great people and get them on your team, and hopefully they'll feel motivated. Or you can get people who are already motivated, who can't think about anything else, and who are most excited about this today. If you didn't hire them, they'd be doing this anyway.
That's still the criteria we have for bringing in early hires, because it helps us naturally stay up to date all the time. I think it also helps people stay constantly motivated about what they're working on. You don't have to try to do that; people are already AI-pilled or Gen Media-pilled.
This episode is brought to you by Brex, my favorite. You become what you spend on, and I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes—the companies building what's next, from Vercel, OpenAI, Anthropic, Granola, and Deepgram, all made the same call. They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically, enforce policy before spend happens, and close your books in minutes. That's why Sorcery runs on Brex, so I can spend time on building and not busy work. It's time to get Brex AF. Learn more at brex.com/sorcery. That's B-R-E-X.com/S-O-U-R-C-E-R-Y. Bye. From handshake to signature faster, Harvey, the AI platform built for legal work, is trusted by 80% of the Am Law 100 and over 20% of the Fortune 500. Learn more today at harvey.ai. That's harvey.ai. AI needs more than chips. It needs power, land, and infrastructure. Zone develops next-generation data center campuses, partnering with AI companies, site developers, and technology leaders to bring compute online faster and at scale. Zone is building the foundation of the AI frontier. Visit zonefrontier.com to learn more. That's zonefrontier.com to learn more.
Founders scale faster on Deel. Set up payroll for any country in minutes, hire anyone anywhere, get visas handled fast, and get back to building. Visit deel.com/sorcery. That's D-E-E-L.com/sorcery.
So where are you hiring people from? What were their roles? This is all very new. This is a common question I ask in these conversations, because I have to remind people that none of this really existed 3 or 4 years ago. Not many people are trained.
And even in the beginning, it was an overlooked market because it happened at the same time as the growth of LLMs.
Yeah.
Maybe if it had happened by itself, it would have been the hottest thing, but it was the second-hottest thing.
It's also hard because it's the creative industry. It's kind of at odds with technology to some degree.
There is a lot of technology involved in creative industries, right? Even an IMAX camera is the size of that table over there. There is technology involved in different parts of animation and filmmaking, in bigger ways, and then it plateaus. Maybe we're going through one of those times where, in the next 5 to 10 years, it's going to have another outsized impact on the creative industries.
AWS and AMD have massive media businesses where they sell to studios and media companies. This is just the next wave of it.
Yeah. I think I was listening to something about Christopher Nolan and his infrastructure, where he's keeping all that IMAX film.
Yeah, 900 pounds.
So heavy.
Yeah, it was.
Okay, so let's talk a little bit about the team and how you guys all gathered together. You came from Coinbase.
Mm-hmm.
You came from Amazon?
Amazon, yeah.
Yeah. So how did you guys meet, and what were the early steps in gathering a group together?
You go.
We're both Turkish, but we didn't know each other back home. We met in San Francisco when I first moved here. I met him immediately in 2014. We were both working at different companies, minding our own business, and it was during COVID.
We've told this story a couple more times before. We rented a house in Palm Springs, and that's when we first thought, “Okay, maybe there's a working chemistry between us.” It took us maybe another year to start the company, but that's how we met. We were longtime friends for over 10 years and never worked together until that COVID Palm Springs house. Here we are. We've been building the company for 5 or 6 years now.
And you've been pulling people from Coinbase, from Amazon.
Coinbase and Amazon. It's funny.
Really? Oh, damn.
It's from both places. Exactly.
What were you looking for in the early days? What were the types of hires?
I would say just really good people in many ways. One thing is that they obviously have to be excellent technically. Early on, we definitely had to get the technical part right because we were doing a lot of deep technical work. We still are, but at that time, that was our differentiator.
I would also say people we get along with. Early team chemistry is really important. Figuring out product-market fit is probably the most challenging thing for a startup, right? In our mind, you have to keep a pretty small team that can work really well together and stick together, with a great way of experimenting with ideas, throwing out what doesn't work, and just keeping moving and looking for—
I think this is one of the biggest advantages of starting a company in your mid-30s: you have your whole working life of people you worked with, trusted, and didn't trust, whom you can bring together. As soon as you reach out to people, if you have something worthwhile, they come and join you and want to be part of the story.
If you're younger, you unfortunately don't have that. You have to go on your own, try to hire people, and build trust. But here, we've built trust with so many people. That early engineering team was almost all people we knew, had worked with, and trusted. That helped us navigate problems so much more easily.
Yeah.
What are your views on scaling up or keeping flat?
Culture-wise, our company is flatter than your run-of-the-mill startup. The deepest level we have is 2 levels of reporting. That's by design. We really believe in these “super ICs.” I think it's becoming more popular now that it has a name. I don't think we had a name for it. I don't think we had—
It’s just Super IC?
It’s Super IC, yeah.
It’s Super IC.
So we have forward-deployed—
Yeah.
…deployed agent—no, not agents. Is it agents?
Engineers.
We’ll have maybe agents soon.
No.
Forward-deployed—
I mean, I should know—
Forward-deployed engineers.
…because I know Palantir quite well. So we have forward-deployed engineers—
Yes.
…and now we have Super ICs.
Super ICs.
Super ICs.
Super ICs, I think, is becoming a more common term. Basically, you want your management layer to be Super ICs: people who are so good at their jobs that they can do hands-on work and also lead teams. I think we have a similar structure. Without calling it this, we’ve always had a structure where we don’t have pure managers, and we don’t plan on having peer managers.
Yeah.
We can gather the team around really magnetic people who are really good, and that’s a natural way of building leadership within the team. That’s how we’ve been building it out: we add really good people, see leaders emerge, and gather the teams around them. But how long can you go this way? I think there’s a limit. You start hitting communication bottlenecks, and I think that’s part of the scaling challenge.
We think a lot about how to communicate with the team broadly and keep everyone aligned. As you grow, those start becoming the challenges. So far, we just try to have lots of check-ins with the team. We’re still keeping everything flat and seeing when things break; then we might add more management layers. But essentially, we’re really focused on this Super IC strategy.
In a way, our growth was timed perfectly with coding agents and LLMs becoming a huge part of software engineering. I think we can keep this up as long as agents and coding models improve at this pace. We just need really talented leaders—the Super ICs—to lead maybe 5 to 10 engineers, relying on coding agents to do most of the work. I think we can keep up this way, with maybe just 2 levels of hierarchy, and that’s it.
AppLovin does multiple billions of dollars in revenue. They have around 400 employees and 100 people on their engineering team.
Yeah.
Yeah.
Big fan of Adam.
Big fan.
Big fan.
I was there today, so it’s crazy. Their market cap peaked at around $250 billion.
Yeah.
Now they’re at around $130 billion to $150 billion. He’s such a proponent of what you’re saying. His whole thing is built on—
Yeah.
…individual contributors, and—
Yeah.
…people—not just because we now have access to coding agents. What about the other side of the brain? What about the human, personal things—
Yeah, yeah.
…and also holding the business weight of it and making business decisions—
Yeah.
…but having the team effort around it.
Yeah.
I think that’s a structural component that’s fundamentally changing. You can’t just be a 10X engineer. You have to be—
Yeah.
…a lot more.
I think other CEOs are talking about it too, right? Brian Armstrong from Coinbase—
Mm-hmm.
…recently talked about it and literally described a blueprint: “This is how we’re building the company going forward.” I think Elon is a big proponent of this, for better or for worse. He talks about it a lot. I think it’s really picking up, especially among tech companies. It’s harder to do at non-tech companies because there’s so much hierarchy.
Tech companies are better suited to pull this off, though they still need to go out of their comfort zone. But it’s something a lot of people are talking about.
So what are the key traits you guys look for?
—
You mentioned good people, but—
Agency is one thing that—
Yes.
…we talk about in the office. Everyone would want people with agency, but you just mentioned software engineers even making business decisions. I think that’s unheard of in a big company like Amazon or Coinbase. Software engineers do software engineering.
But because the market is moving so fast, and customers are emerging and changing between different models so fast, a lot of the time we want our software engineers to make business decisions or work with customers, and therefore make business decisions around them. That’s the hardest type of engineer to find. I think you look for agency: if this person can act independently, take responsibility, and own the decisions, then the other things we want come naturally with it.
It’s hard to test for. I think the best way to know is if you’ve worked with him or her before. That’s the best way to know. But that’s something we care about a lot.
Yeah. I think we also look for hunger, which may correlate a little. You need to really push in this market to actually win. You have to really want to win and build a big company.
We try to test for that, which is very hard to do, but I think you can get a sense when you sit down and talk to a candidate. We do a little bit of reverse selling, which may help identify hunger. We’ll say, “Hey, it’s a lot of work.” We don’t believe in toxic work hours or things like that. We go home during weekends, for example.
Really?
Yeah.
What?
Imagine that.
This is the first I’ve heard of that.
Yeah. Nothing to brag about—not going home.
You go home during weekends and you’re still successful.
Yes. Imagine.
Possible.
Crazy.
But I think there’s a certain baseline we try to be realistic about in terms of what to expect. That’s part of the reverse sell: “Hey, there can be a lot of time crunch, where we really push and have to spend a lot of time working—60 hours or 80 hours.” I think that filters out certain people.
Looking for hunger is very important. Culturally, anyone who has interacted with our team on the customer side would probably attest to this: our team is pretty low-ego and easy to work with. We look for humility in general. They’ve got to be really talented, but not be a jerk about it. I think that’s pretty unique to find.
Anyone who’s interacted with our team probably knows that, and that’s why they really like us: people are easy to work with and friendly. I think you can have both at the same time, so we try to filter for humility as well.
What have been the biggest surprise challenges you guys have come across? Not just the standard stuff, but something you had no idea you were going to expect.
I’d say being at the forefront as a company is pretty hard. First, you have to really want it.
Mm-hmm.
You have to want to build a big, big company. Second, you have to keep up with it, which requires a lot of effort. In order to build a big startup, you really have to be very committed and very hungry. That takes a lot of work and takes a toll on you. I think Adam would probably say this. But it just requires a lot of hard work.
I don’t know. I asked him what he’s most excited about. He says he doesn’t feel excitement.
So he’s beyond that point.
You’re going to get there one day. No emotions.
Our industry is incredibly dynamic. When there’s no good model for a month, everyone starts panicking.
Oh.
“Oh my God, it’s been one month; nothing new has happened.” But the answer is just to wait another month, and something incredible is going to happen. It’s been true since we started, and that keeps the excitement going. It keeps people encouraged that the next model will unlock other use cases, which will unlock other types of customers. That has brought so much growth to our business, and so far it just keeps happening for us.
Where have you been seeing some of the best models come from? Do they just come out sporadically? Is there any consistent theme between them?
Right now, we play a bigger role in the ecosystem because we’re an infrastructure partner for model providers, but we’re also the API partner for the application layer: studios and people who are consuming the models. So we get a good sense of supply and demand on both sides. We’re in such a compute-constrained moment that, if you’re coming up with a new model, you really have to plan ahead. You have to make sure you have enough compute to run inference.
If you’re successful, you’re likely not going to be able to procure it on your own, so it’s usually a good idea to partner with an inference provider like us to do your launch. We work very closely with model providers. Customers usually use multiple models at the same time, but they really care about model releases because they often tie their product releases to them as well. So we keep in very close communication with them.
We also sometimes have to anticipate how much growth is going to come on the model side, so we can procure our own compute and make investments to enable the whole space. Someone really has to do that, because otherwise all the compute is going to go to coding agents or whoever is making the bigger investment.
Mm.
By planning ahead and investing in compute, we can enable both the model providers and the applications being built on it. That’s become our responsibility in the ecosystem over the last couple of months because compute is so hard to come by.
I’ll add a few more points. I think generative media is very unique in two ways. One, there are just more modalities. When people think of generative media, maybe they only think of video.
Yeah.
But you have to think about text-to-image, image editing, video, 3D, text-to-speech, and speech-to-text. There are many modalities within generative media. There are five to 10 companies in each of these subcategories, so there’s a lot more happening and more fragmentation.
The other point goes back to what I said about diminishing returns. It’s still so early in the space that there’s a lot more leapfrogging happening between players. We’re not at the 98-point-whatever-percent level; we’re only at maybe 50% of where we’re ultimately going to go. There’s more leapfrogging and more diversity in the number of labs. It’s not just OpenAI, Anthropic, OpenAI, Anthropic. There are more like eight players, and they’re all leapfrogging each other, which is very—
One big reason is that we’re not running out of data. There’s so much video data out there, and so much new video data is being produced every day, that no one claims to train on all of it. The hard problem with training a video model is actually filtering out the interesting video data so you make a good model.
If you need to put more data into the model, there’s an abundance, and it seems free. There’s enough free video out there that it works very well. But the problem space is bigger, right? You’re trying to reason around the whole world, physics, everything, so it’s not as specific as coding or as certain LLMs are trying to optimize around. There are a lot more angles people can take to create a more interesting and appealing model, and it’s been happening almost every other month.
I think that’s where the naming comes in, too.
There you go. You have to differentiate with your naming.
They’ve got some funny names.
Yeah. Happy Horse, Nano Banana. What do you like?
There’s Happy Oyster.
Happy Oyster.
It’s like a game model.
Aw.
There’s MiniMax—
Seedance, Seedance, and Seedream.
Seedance, Seedream.
I think those are—
When Nano Banana had its moment and everyone was talking about it for three days, even my dad was talking about Nano Banana. I was like, “What the hell is going on here?”
I think the backstory is that it was an anonymous model. People do these cloaked releases where they release a model but don’t disclose who it is.
Yeah.
They use a code name to get people excited about it because it ranks highly. I think they somehow used Nano Banana as a placeholder, a code name.
Yeah.
And then it stuck, so it became the original name.
If you guys were a model, what would your model name be?
Sorbet.
Sorbet.
Both of you are saying it.
This is Berke’s favorite word.
I wanted to rename the company Sorbet.
I don’t know why.
I’m just supporting it.
Sorbet?
He likes sorbet a lot.
Yeah. I also think about names. I like these exotic colors, like burgundy and turquoise. I think that would be a good series for model names.
That is a good name, I guess. What about you?
I don’t know.
Sorbet could be a color, apparently, too. It’s like the peach color—
Yeah.
It’s associated with sorbet as a color.
I like birds. There are so many birds.
Birds.
Yeah, you can do birds.
Do you have a favorite bird? I’m asking for a name.
I don’t know.
Yeah, yeah.
You can even do owls. There are so many owls.
Owls, yeah.
You can literally just do different owls.
I like owls.
You can do insects. Nature, I think, would be my theme.
Well, how did you come up with the name FAL?
All right. So FAL—
Let’s talk about this.
Yeah, yeah.
We have a longer name, Features and Labels, and it’s the—
It’s a double entendre.
Yeah.
Yeah.
Wow.
It’s the initials of Features and Labels, which—
And it’s not “fal.” It’s FAL.
It’s FAL.
Mm-hmm.
FAL. It’s FAL. It’s Features and Labels, which, if you’re a real OG, you’d know are two terms from the machine-learning world. They’re very old, pre-AI. So it’s the initials for that, but also, have you had Turkish coffee?
Yeah, I have.
Have you ever been told your fortune from the coffee grounds?
No.
All right, so that’s “fal.” It’s like having your fortune told by the coffee grounds. You should get it done. We’ll do it sometime.
Okay. Okay, let’s do it.
It’s like a prediction of your future.
Someone looks at what’s left when you finish the coffee.
Are you guys trained at that? Can you tell fortunes?
No, everyone in Turkey is trained. Everyone has done it before.
Really?
Yeah, yeah.
Yeah, yeah. You, you…
Can we do it right now?
We should—
We could try.
Yeah, we’ll get some coffee. I think they have Turkish coffee here.
Oh my gosh. Well, that’s fun.
Yeah.
Do you think the open-source/closed-source debate is still going to be going on in two to four months, six months? How long do you think that’s going to keep going?
I think so, because anyone who’s falling behind a little bit can use open-sourcing their model as an incredible marketing tactic. The whole world starts talking about you if you open-source your model. And there are real advantages. If you talk to anyone at these studios in Hollywood, they want to be able to fine-tune the model.
They want to apply different LoRAs and filters to change the lighting, and to use their unreleased characters and styles on the models. The only way to do that is open source, so there are real technical advantages to it. It’s also an incredible marketing tactic, so people are just going to keep doing it. Someone’s going to create the frontier model, and people who are falling behind are going to open-source theirs. The pendulum is going to keep swinging.
We thought this was a bigger threat to our business in the beginning because we were running mostly open-source models. Then we started partnering with closed-source models and running inference for them as well. It’s a win-win situation for us, but open source makes for a more interesting ecosystem. I think it’s good for the industry in general, so we always welcome more open-source models and support the community as much as we can.
I think the fact that people want to protect their IP is going to continue, right?
Mm-hmm.
Not just for media, but you want to protect your source code, your customer lists, whatever. All of that is always going to be a thing. If you have something local, you don’t have to worry about it. I think this debate is always going to be there.
When we talk to our customers, if they know the model is running on our GPUs, they trust the whole process so much more. Open or closed, it works the same, but that’s been a big differentiator for our business. When it’s open, they know for sure that they can even self-host, but they trust us because we can do it better than they can.
Trust is really important with video models, especially because people don’t want their unreleased content to leak. That content is usually an input to the video model when creatives work with it. We’ve been really pushing on that with our enterprise customers, and it resonates.
How good is AI at real professional work? Mercure's AI Productivity Index, or APEX, tracks it with every major frontier model release. Built with partners like Harvey and Cognition, see where the frontier actually stands at mercure.com/apex. Today's episode is sponsored by VC-X by Fundrise, the public ticker for private tech, allowing investors of all sizes to invest in venture capital. Learn more at getvcx.com. Public is a multi-asset investing platform where you can automate your investing with AI agents. Create your own custom agent with a simple prompt. Agents can monitor markets, execute trading strategies, and manage your cash across stocks, bonds, options, crypto and more. Wanna buy at market close and sell the next day's open? Just tell your agent. Automate your investing at public.com.
Talking macro, what do you think the biggest question is that people aren’t asking right now?
Not exactly that, but if you listen to any Hollywood podcaster or look at the news about Hollywood, they always have this angle of, “Oh, AI’s going to kill Hollywood.” We’re in this business, and we talk to these people all the time. So far, I’d say AI has had zero impact on anything in Hollywood. People are projecting a little bit into the future.
The business we get from Hollywood has been increasing in the last 6 months. I think it’s reached a meaningful percentage of our revenue, but I’d say it’s had zero impact on labor. It’s barely made any content that’s worthwhile in Hollywood. It’s going to happen; there are signs of it. But we’ve been talking about it for 4 years, and it’s always been fearmongering and nothing real. I think that’s what people don’t realize from a distance.
Yeah. Most of LA, that’s where—
People, yeah.
…this is happening.
It’s really regulatory in what they’re doing there, with different laws.
Exactly.
It really has nothing—
And that’s been the reason adoption has been slow. It’s not just the quality of the models. The lawyers at the companies need to be okay with it as well. But finally, slowly, we’re seeing that everyone is going full force at it.
What do you think people should be asking? What’s going on out there?
I think people are asking everything.
They’re asking too much.
They’re asking too much. There’s a lot of skepticism about use cases, speaking more broadly about AI. There’s skepticism about all this CapEx investment: where is that going to go? Is AI actually useful? But I’m more on the optimistic side. A lot of the people making these claims don’t really understand what’s going on or how AI is actually being used at companies.
Generative media is still very early. If anything, it’s just starting. We’re starting to see it in ads, Hollywood, social, and elsewhere. I think there’s real value. We were just estimating how much faster our software engineering team is moving. I estimated somewhere between 3 and 4X. Pre-AI, we would literally need 4 times the people to go at the same speed, and we’re definitely paying less than that.
Mm.
There’s clear ROI for us. If you really go deep into the ROI for companies using AI and actually try to measure it—not read it from a bunch of reports or nontechnical people—I think you’ll see a different picture. I’m very much in favor, very long compute, as they say.
Yeah. We talk about it: every big company is going to be a compute company. Just the storefront is going to be different. Our storefront is unique and interesting; no one else has one similar to ours. Some have coding agents. If you go to a neocloud, there’s no storefront. If you go to Databricks, it’s about data. Everyone has a different story and a different way to talk about it.
But behind the scenes, it’s about who’s running the most compute. That’s what brings in the revenue. We have a unique, interesting storefront, and it keeps growing.
Yeah, and there’s not enough of it. All the numbers you hear, like $200 billion in CapEx and whatever, a lot of that stuff isn’t built yet. It’s very slow, right? Building these things takes a lot of time, and there’s not enough compute right now.
Yeah.
I have 2 final questions, and then I’m going to ask you what your hot take is. I’m letting you know that now so you can get it—
Think about it.
…prepped. I want to—
This was a hot take.
Don’t want to take—
Yeah, you wasted his hot take on that.
You guys have been crushing it. You have great answers, and they’re quite opinionated, so it’s awesome. But before we get there, what are you most looking forward to in the next 12 months?
I’m looking forward to the first timeless AI content. The way I define this is: you and I are going to watch this, and then a year later we’re going to watch it again, and we’re still going to like it the same way we like it today. So far, AI content has been about capabilities: “Oh, look, this is possible to do with a video model,” or “That’s possible to do with an image model.” But we need to reach the point where things look fresh all the time, and I’m looking forward to that.
I think in the next year there’s going to be something widely accepted—maybe a short movie, maybe an animation—that people will look back at in a couple of years and still enjoy the same way they enjoy it today.
That’s pretty good.
Evergreen content.
Yeah.
It’s very important.
I try to do that, but—
There you go.
…it’s hard.
People say Toy Story is that for computer graphics. That was the first time. It came out in 1996, something like that.
Isn’t it crazy how long ago that was created?
And, you know, we can just go watch it, and it’s going to be good.
Yeah.
I’m going to enjoy it.
Did you like The Odyssey in that way?
I haven’t seen it yet.
You haven’t seen it?
No.
All you’ve been doing is talking about Christopher Nolan. You haven’t seen it?
I’m waiting for the IMAX, the big IMAX. I couldn’t get a ticket yet.
Go to the Sphere.
Yeah, sure. Did they show it there?
I don't know. But they should.
It's too big.
Too big.
Too big. Not as Nolan intended.
Yeah, no.
For me, I would say world models are something we haven't spent too much time talking about.
Yeah.
I think they're super exciting. I was very skeptical even 3 months ago. World models are these models you can play around in and move around in. It kind of looks like a game, and you can put in actions with your keyboard and navigate the world. But seeing what I've seen recently, it's pretty incredible. I actually started feeling very bullish about it.
I think it's a matter of the quality going up and the speed getting so much better. I think there are new architectures now that let people have more persistent worlds. Before, you could look this way, see the flower, look away, come back, and it would be gone. Now they're able to make things persist. You'll be able to build a game with a prompt.
I think that space is just going to blow up. The big challenge there is compute, because you need a GPU running at all times to play that game. But I think the quality of the games themselves is going to get a lot better, and I think it's probably going to be yet another product-market fit. Gaming with AI isn't a thing yet. I actually think there's a big chance that may happen, and it might be this fourth category, maybe.
Very expensive.
Yeah.
Damn. Okay, so we have one more question until the hot take.
Okay.
I'm just letting you know again. I'm warning you. Okay, so you guys have worked for legends, whether it's Brian Armstrong or Jeff Bezos. You like chips, so you might like Jensen Huang. And you also like Christopher Nolan. I'm curious: who are some figures that really inspire you and keep you motivated? Who keeps you running at it?
Great question. Of the previous generation of founders, I think Jack Dorsey is very underrated. I think he has more range than anyone else, and now he's maybe coming back. Watching his journey with Twitter and then Square is inspiring. I never got to work closely with him, but when I was younger in Silicon Valley, I always looked up to what he was doing.
I wouldn't say he keeps me going, but I really like Ali Ghodsi.
Mm-hmm.
I think he's transformed his company many times over time. I like to think—
Keeps reinventing, yeah.
—he keeps reinventing. I think that's very important. I think that's a secret to building a long-lasting company and navigating through all these changes that are happening. So, yeah, I think he's a really great leader.
The time has come. What is your hottest take right now?
Do you have it?
No.
Nah.
There's so much pressure. We can take our time.
Yeah, let's take our time.
Right? And then just—
We can cut it.
And then it's dark outside.
Should we prompt you with some?
No, no. Let's do it.
I have a good one.
It could be on IPOs. It could be on companies disappearing. It could be on fake GPUs that people sell for— I don't know where they get them. It could be—
Yeah.
I had a hot take a year ago: the OpenAI–Meta merger.
Yes.
Ooh. Could still happen.
Again, it could still happen.
They're not doing too great.
I think it makes sense.
My hot take is that it would be bad for the whole industry, but—
Sure, sure. Okay, so I guess it's not too spicy or anything, but I think it makes a lot of sense. OpenAI was known for its consumer edge, right? That was their whole thing: OpenAI is winning in consumer, therefore they're going to win, because consumer is so much harder to win. Then Anthropic came with its enterprise thing, and Google started making its own consumer move.
I see OpenAI not necessarily winning very strongly in either enterprise or consumer. Meta, on the other hand, is in a similar situation where they need something. I thought, “What if they combine their forces? That would be really cool.” They could really start winning in consumer. I thought it made sense. But again, I have no information about any of this, so I'm completely speculating. I thought it would be a good move.
Mm-hmm. That is very interesting.
Another one—
Well, this is a common hot take. I'll do it for both of us.
Sure.
I have more now. It's coming.
Okay, good, good.
Yeah.
Yeah, yeah.
Good. It's coming.
The consensus view was that there'd be more margin in the application layer.
Yeah.
I think it's completely reversed. Margins are close to zero in the application layer. There's so much competition. People need to spend so much money on marketing. You can only create margin if you're in the compute path and can innovate in that path. That's the only way you can create margin, which wasn't the case when we started the company 4 or 5 years ago. Everyone thought the infrastructure layer was going to end up with prices going to the bottom, and applications were going to have margins. Looking around, the application layer doesn't have any margin today, either. So wake up, VCs.
It's a good one.
Yeah.
Yeah, very good.
We talked to Tony Kim about this at RAISE, and he was like, “The chips have much better margins than all of these other companies.”
The chips have much better margins.
I mean, they're the only ones with margins right now.
They're the only ones. Yeah, yeah, yeah.
Right?
Yeah. I have another one. This is a very recent one. I've been kind of toying with this.
Yeah.
I think coding is part of Gen Media.
Oh.
Really?
Yes. Well, it's also part of Gen Media, right? It's its own thing, but at the same time, we're seeing so many developers building richer applications that aren't just websites. They can be programmable video, right? Content. Maybe you've seen these websites that look more like a video. I think coding is becoming a very key part of what developers on fal are using. So I think we're going to have to invest in it because there's so much demand from our own customers.
Everything's a computer.
Everything's a computer.
Everything's a computer.
There you go. Everything is a computer.
Everything's a computer.
That's the hot take.
That is a hot take. Well, that's a great place to end. Thank you guys so much. I'm very excited for dinner, and—
And the coffee.
And the coffee.
Yeah.
I can't wait to know what my fortune is.
There you go.
Cool.
Thank you.
Thank you, Molly.