Inside fal’s $1.65B Series E & F Led by Sequoia & Bessemer

1. Key Themes

Generative media is AI's third product-market fit, defined by individual spend of $10–20K/month

Gorkem Yurtseven frames the market by who actually spends money: only coding agents and generative media see single users spending enormous sums. "AI really has like three product-market fits. One of them is search... The other really clear one that's like growing really fast now is coding agents. And the third one is, we believe, is generative media." [00:03:42] He adds: "This is the only other use case in AI that an individual can spend that much money... It's just coding and generative media." [00:03:42] Demand is professional, not hobbyist: "if this is your job, if you work at a AI-native studio... you sit down in front of a computer, and you generate videos all day, and you spend $10,000, $20,000 a month." [00:03:42]

Prices go up, not down, as quality rises, and demand follows

Burkay Gür describes a counterintuitive pricing curve in video. Kling and MiniMax-class models settled around $0.30–$0.40 for a 5-second clip, then Veo 2 arrived: "Veo 2 was very surprisingly it was like $5 per 5-second video. So it jumped like, you know, more than 10x. And we were like, okay, like who the heck is going to pay the pay all this money, right?... Turns out like there is essentially infinite demand." [00:06:51] Professional marketing and movie production users drove it: "it's like the usage going up, and then the capabilities are going up, and then everything is getting more expensive over time." [00:06:51] Gorkem adds the cost mechanics: "as you increase the resolution, as you increase the duration, the cost increases exponentially because you have to have more attention... during inference time, and resolution increases the amount of pixels exponentially." [00:04:58]

Product mix whipsawed: image → video → image editing revived images (50/50 today)

Gorkem walks through three "surprises" in revenue mix: video went from 0 to ~70% of revenue, then he thought "image models are like done." Then "image editing models brought life back to image models. And then it's also useful for video models as well, because... you can create different frames, and video models does such such such better job interpolating between those frames." [00:04:58] Revenue is now "50-50." The lesson: forecasting the winning modality was repeatedly wrong, so the platform bet beats the model bet.

Gen media has no diminishing-returns wall (yet), with many leapfrogging labs and unlimited data

Burkay: "with LLMs, I think people talk a lot about hitting a wall or like diminishing returns. I think, in this side of the market, we're like nowhere near uh actually hitting like diminishing returns." [00:06:51] Structural reasons for more competition: "there's actually more people leapfrogging. There's more diversity in in like the number of labs. So, it's not just like OpenAI, Anthropic, OpenAI, Anthropic... it's more like there's like eight... players and they're all like leapfrogging each other." [00:40:21] Gorkem on data: "we are not running out of data... The hard problem with training a video model is actually you have to filter out the interesting video data." [00:40:56] Gen media also spans many modalities (text-to-image, image editing, video, 3D, TTS, STT), each with "5 to 10 companies." [00:39:43]

fal's role: infrastructure partner to both model labs and application builders in a compute-constrained market

Gorkem explains fal sits between supply and demand: "we're an infrastructure partner for the model providers, but we are also the the API partner for the application layer or studios." [00:37:49] On compute scarcity: "if you are successful, you are likely not going to be able to procure it on your own. So, it's usually a good idea to partner up with an inference provider like us to do your launch." [00:38:13] And fal now pre-buys compute: "someone really has to do that because otherwise all the compute is going to go to coding agents... By planning ahead and investing on compute a little bit, we can enable both the model providers and the applications." [00:39:04]

Origin story: pivoted from "Python data infra" to image-model inference when revenue went from zero to real

fal started in 2021 on a Snowflake/Databricks-style thesis: "They own compute... we thought Python is going to be the next thing." [00:13:42] After ChatGPT and Stable Diffusion, they avoided the LLM token price race ("we can't really like go and, you know, sell tokens at a negative margin") and chose image models: "with this other business, we ran it for a year, we didn't make a single dollar. With this one, our revenue is starting to go like this." [00:13:42] They then coined the category: "we actually came up with this word generative media, and we slapped it on our website, and we said fal, the generative media company." [00:13:42]

Flat org with "super-ICs," and coding agents as the enabler

Burkay: "you want your like management layer to be super-ICs. So people who are just like so epic at their job, they can like actually do hands-on work and then they can also like lead teams... we don't have pure managers, we don't plan on having pure managers." [00:28:37] Max depth is "two levels of like reporting." [00:28:21] Gorkem links this to agents: "You just need really talented leaders, the super-ICs, to manage maybe be the leader of, I don't know, 5 to 10 engineers, and you rely on coding agents to do most of the work." [00:30:17] Burkay estimates ROI: "I estimated somewhere between like 3 to 4x. Like we would need pre-AI we would literally need like four times the people to go at the same speed." [00:51:28]

Hiring filter: obsession, agency, hunger, humility, and a "reverse sell"

Burkay borrowed Coinbase's crypto quiz idea: "you really want like these like fanatics about the space... If you didn't hire them, they would be doing this anyways." [00:20:14] Gorkem on agency: "we want our software engineers to make business decisions or work with customers... that's the hardest type of engineer to find." [00:32:42] Burkay on hunger and the reverse sell: "Hey, like it's a lot of work... there can be a lot of like time crunch... 60 hours and 80 hours. And I think that like filters out maybe like certain people." [00:33:56] Plus humility: "they got to be really talented, but at the same time like, not be a jerk about it." [00:35:31]

Margin has moved to compute, not applications

Gorkem's "hot take": "the consensus view was there's going to be more margin in the application layer, and I think it's completely reversed. Margins are close to zero in the application layer. There's so much competition... You can only create margin if you are in the compute path." [01:01:08] Reinforced by "Every big company is going to be a compute company. Just the storefront is going to be different." [00:00:25] and Burkay: "there's not enough compute right now." [00:53:12]

Next frontiers: automated ad optimization, world models, and "timeless" AI content

Gorkem on ads: "you can post these ads and see how people are interacting with it... show more of the things that are performing well, and this can all happen automatically... we would love to be part of this like automated ad story." [00:10:23] Burkay on world models: "I was very skeptical, like even 3 months ago... new architectures now that people are able to actually have like more persistent worlds... I actually think there's a big chance that may happen, and it might be like this fourth category maybe." [00:55:14] Gorkem's 12-month wish: "the first timeless AI content... So far, AI content has been about capabilities." [00:53:42]

2. Contrarian Perspectives

Application-layer margins are near zero; compute is where the profit is

The consensus VC thesis said infra gets commoditized and apps capture value. Gorkem says the opposite has happened: "Margins are close to zero in the application layer. There's so much competition. People need to spend so much money for marketing. You can only create margin if you are in the compute path, and you can innovate in that path... So, wake up, VCs." [01:01:13] Burkay adds it's the chip layer that has margin: "And there are only ones with margins right now." [01:02:01] This is notable coming from a company that is itself an inference layer, but they back it with their own revenue trajectory.

Better models mean more expensive models, and demand is effectively infinite

The standard infra assumption is price-to-the-floor deflation. Burkay's data point is the reverse: a 10x price jump with Veo 2 ($0.30 → $5 per 5-second clip) produced more usage, not less. "Turns out like there is essentially infinite demand... everything is getting more expensive over time." [00:06:51] Gorkem even frames output length and resolution as a multiplier: "there's a lot more money to be made going forward, because the video output we get is still like very rudimentary." [00:04:58]

"AI will kill Hollywood" is overblown, and the real constraint is lawyers

Gorkem works with studios daily: "So far, I would say I had zero impact on anything in Hollywood... it had zero impact on the labor. It barely made any content that's worthwhile in Hollywood... we've been talking about it for 4 years, and it's always been fearmongering." [00:49:25] The binding constraint isn't model quality: "the lawyers of the companies need to be okay with it as well." [00:50:14] Yet Hollywood revenue to fal "has been increasing in the last 6 months... reached to a meaningful percentage of our revenue." [00:49:25] Slow labor impact and rising spend coexist.

Open-source vs. closed-source is a permanent pendulum, and it's good for an inference platform

Rather than seeing open-source as a threat, Gorkem argues it's structural: "someone's going to create a frontier model, people who are falling behind is going to open-source it, and the pendulum is going to keep swinging." [00:45:44] Open-sourcing is "an incredible marketing tactic," and studios need open weights to "fine tune the model... apply different LoRAs... use their unreleased characters." [00:45:21] fal benefits either way because customers trust it to run unreleased content on its GPUs: "if they know the model is run on our GPUs, they trust the whole process so much more." [00:47:08]

Coding is part of generative media

Burkay's late hot take: "I think coding is part of gen media... we're seeing so many developers building like richer applications that are not just like a website, like they can be programmable video." [01:02:04] This blurs the "three product-market fits" taxonomy and signals fal moving into coding to serve its own developer base.

3. Companies Identified

fal

Generative media inference platform (founded 2021, ~150 people) offering API access to image, video, audio, 3D and other models; just raised Series E & F at a $1.65B valuation level led by Sequoia and Bessemer. Mentioned as the subject of the episode, with revenue "50-50" image/video. Quote: "we're an infrastructure partner for the model providers, but we are also the the API partner for the application layer or studios, people who are consuming the models." [00:37:49]

Coinbase

Crypto exchange where Burkay worked from 2016 and a source of fal hires. Mentioned for its hiring practice: "they had this like um crypto quiz that they posted on Reddit... you had to pass this quiz to like actually get on a phone call with someone on the team." [00:19:10] Also cited for Brian Armstrong's AI-era org blueprint. [00:31:45]

Amazon

Gorkem's former employer and a fal hiring source. Also noted for its media business: "AWS, AMD, they have massive like media businesses where they sell to to studios and media companies." [00:23:35]

AppLovin

Ad-tech company that Molly visited the same day; held up as an example of a high-revenue, small-team model. Quote: "AppLovin does multibillion dollars in revenue. They have like 400 employees and 100 people on their engineering team." [00:30:55] Gorkem and Burkay said they are fans of CEO Adam Foroughi.

Kling

Video model that Burkay cites as a top video model alongside MiniMax before Veo 2's price jump. "Kling was probably one of the best video models, and uh there was also like Minimax." [00:06:51]

MiniMax

Chinese lab with video models and a named leapfrogger in the space, listed among model names. [00:42:12]

Google (Veo, Nano Banana, Gemini)

Veo 2's $5/5-sec price proved demand elasticity. Google also appears in the consumer AI landscape: "Google started adding made their their own consumer move." [01:00:17] Nano Banana is described as a cloaked-release code name that stuck. [00:42:30]

Seedance (ByteDance)

Cited as an example of ongoing rapid improvement: "there's just so much new stuff happening with like Seedance V2 and V2.5. Uh, it's just like the quality is going up, prices are going up, and people still want to use all these models." [00:06:51]

OpenAI (Sora)

fal hosted Sora and saw growth right before shutdown: "we saw huge growth in the like last 3 months before um they decide to like take it down... the rumor is like there wasn't enough compute for everything else." [00:09:06] Also the subject of Burkay's OpenAI–Meta merger hot take.

Meta

Burkay's hot take subject: a merger with OpenAI to combine strengths in consumer. [00:59:51]

Anthropic

Cited as having moved into enterprise: "Anthropic sort of came with, you know, their enterprise thing." [00:59:51] Also a customer-style reference for coding-agent spend.

Higgsfield

Generative ad/creative platform that Molly mentions AppLovin might use or acquire. [00:09:36]

Snowflake and Databricks

The comps behind fal's original thesis: "this previous data infrastructure wave led to some amazing companies like Snowflake, Databricks. And what do they have in common? They own compute." [00:13:42] Gorkem notes Databricks tells a data story while still running compute. [00:52:24]

Happy Horse, Happy Oyster, SeaDream (named models)

Named in passing as examples of the diverse model ecosystem. [00:42:07] Happy Oyster is described as "a game model." [00:42:07]

Harvey, Brex, Mercor, Deel, Public, VCX by Fundrise, Zone (sponsors)

Sponsors named in ads. Brex is also described by Gorkem as a tool fal uses: "we are proud users." [00:57:02] Harvey is described as "trusted by 80% of the Am Law 100." [00:22:20]

Palantir

Mentioned in passing by Molly with the forward-deployed-engineer comparison to super-ICs. [00:28:27]

Twitter / Square

Cited as the context for Jack Dorsey's range. [00:58:09]

Pixar (Toy Story)

Used as the archetype of timeless content: "people say Toy Story is is that for computer graphics... It came out in 1996... we can just go watch it and it's going to be good." [00:54:37]

4. People Identified

Gorkem Yurtseven

Co-founder of fal; ex-Amazon, Turkish-born, based in SF since ~2014. Framed AI's three product-market fits and the margin-in-compute thesis. Quote: "Every big company is going to be a compute company. Just the storefront is going to be different." [00:00:25]

Burkay Gür

Co-founder of fal; ex-Coinbase (joined 2016). Source of the pricing-elasticity story, super-IC org model, and world-models conviction. Quote: "We just add really good people, and then we see leaders emerge, and we gather like the teams around them." [00:28:37]

Adam Foroughi

CEO of AppLovin, admired by the guests and Molly as a proponent of individual-contributor-driven companies. Molly: "he is such a proponent of what you're just saying. Like, his whole thing is built on individual contributors." [00:31:04] Referenced as a hardworking founder who "doesn't feel excitement." [00:36:58]

Brian Armstrong

CEO of Coinbase; cited as publicly publishing a flat, AI-native org blueprint. Burkay: "Brian Armstrong from Coinbase recently talked about it, like literally like describes a blueprint. 'This is how we're building the company going forward.'" [00:31:48]

Elon Musk

Mentioned as a "very big proponent" of lean, super-IC-style org building. [00:31:48]

Jack Dorsey

Inspiration for Gorkem as a founder with range. Quote: "Jack Dorsey is is very underrated. Um, I think he has uh, more range than than anyone else." [00:57:42]

Ali Ghodsi

CEO of Databricks; Burkay's inspiration for reinvention. Quote: "he's like transformed his company many times like over time... that's like a secret to building a long, you know, long-lasting company." [00:58:27]

Christopher Nolan

Filmmaker discussed for his IMAX film infrastructure ("900 lb.") and The Odyssey. [00:24:23]

Jensen Huang and Jeff Bezos

Named by Molly as legends the guests might draw inspiration from. [00:57:34]

Tony Kim

Previous Sourcery guest, referenced for the point that chips have better margins than applications. Molly: "We talked to Tony Kim about this... he was like, 'So the chips have much better margins.'" [01:01:53]

Sayat and Laura

Owners of Dalida, the restaurant where the episode was recorded, thanked by Burkay. [00:01:25]

Michelle

fal's comms contact whom Molly says she contacted "like a billion times." [00:02:05]

5. Operating Insights

Use a "how obsessed are you" screen as a hiring filter and a content-freshness engine

Burkay copied Coinbase's crypto quiz idea to screen for fanatics rather than motivating non-fanatics: "Or, you can get people who are like already motivated, this is all they can think about... If you didn't hire them, they would be doing this anyways." [00:20:14] The effect compounds: "it helps with just naturally like we're up to date all the time." [00:20:14] Gorkem's complement: he learns more from his leadership team than from the internet. [00:18:21]

Reverse-sell the job to screen for hunger

"We do a little bit of like reverse selling... it's a lot of work... there can be a lot of like time crunch... 60 hours and 80 hours. And I think that like filters out maybe like certain people." [00:33:56] Pair this with a humility filter to preserve a low-ego culture that customers explicitly notice.

Hire your mid-30s network and grow leaders from ICs rather than hiring managers

Gorkem: "one of the biggest advantages of starting a company in your mid-30s, because you have your whole working life, people you worked with, you trusted, you didn't trust, that you can bring in together." [00:27:03] Burkay on leadership: no pure managers, let magnetic ICs emerge, and add layers only when communication breaks. [00:28:21]

Let engineers make business decisions because the customer base shifts monthly

Gorkem: "because the market is moving so fast, the customers are... changing between different models so fast that a lot of the times we want our software engineers to make business decisions or work with customers." [00:32:42] Test for it via prior work history: "the best way to know is if you work with him before or her." [00:33:05]

Pre-buy compute ahead of model launches to be the ecosystem's gatekeeper

fal anticipates model growth, procures its own compute, and partners early with labs, because otherwise "all the compute is going to go to coding agents." [00:39:04] This turns capacity planning into a strategic moat, not a back-office function.

6. Overlooked Insights

Image editing's comeback implies the real product is multi-modal workflows, not any single model

Mentioned inside a revenue anecdote, Gorkem notes image editing is "also useful for video models as well, because... you can create different frames, and video models does such such such better job interpolating between those frames." [00:04:58] This signals that image and video are complementary parts of a pipeline (keyframes → interpolation), which makes a unified inference/API layer more valuable than point-solution video companies, and explains why a "dying" modality returned to growth.

Trust in where the GPUs run is a quiet enterprise moat for hosted models

In a throwaway line, Gorkem notes that customers trust inference more when it runs on fal's hardware: "if they know the model is run on our GPUs, they they trust the whole process so much more... people don't want their unreleased content to leak, and it's usually an input to the video model." [00:47:08] Combined with studios' need for LoRAs and unreleased characters, this positions an independent inference host as the secure intermediary for Hollywood and brand IP, and it works for both closed and open models.