Transcriptly

How to Build $1.5B AI Startup in Just 3 Years | fal's co-founders

EO12:41Added Aug 31, 2026

This episode features co-founders of fal, Burkay Gur and Gorkem Yurtseven.fal is a generative media platform. They just raised their Series C round of $125M,...

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. From Zero to $1.5B: fal's Milestones

  2. 00:00Language models at the time was like all
  3. 00:02the hype, crazy optimism about language
  4. 00:04models. We felt similarly about image
  5. 00:06models. Yes, this is a niche place.
  6. 00:09Where could this be going?
  7. 00:10Finding a niche market that is fast
  8. 00:12growing is the key to startup success.
  9. 00:16After these big models were released is
  10. 00:18that the number of users 10x, 100x,
  11. 00:22maybe over a millionx, we we saw this
  12. 00:24this change early on. We have to build
  13. 00:27systems that are ready for this change.
  14. 00:29We're fast at everything we do. We put
  15. 00:31the models, usually we have day zero
  16. 00:33releases. Space is moving so fast. We
  17. 00:35have to be like very ahead of getting
  18. 00:37these models in front of people, making
  19. 00:38it easy to uh use. There's a lot of
  20. 00:41startups out there, you know, they will
  21. 00:42be stuck on an idea for like months and
  22. 00:44years, right? With no traction. You have
  23. 00:46to really take that to the extreme. Like
  24. 00:49I don't think people stress that enough.
  25. 00:51Think of moving fast. take that and
  26. 00:53multiply with like 100 and move that
  27. 00:56fast.
  28. 00:56If we are wrong, we can always revisit
  29. 00:59our decision. Focusing on image and
  30. 01:02video is going to be an important
  31. 01:04differentiator. We just raised our
  32. 01:06series C round 125 million which values
  33. 01:09us at a 1.5 billion valuation. We're
  34. 01:11very prepared, you know, we're prepared
  35. 01:12to scale. Chat GPT moment for video is
  36. 01:15that I don't think we've hit it yet.
  37. 01:17Right now, we we employed, you know,
  38. 01:18language models at massive scale. We're
  39. 01:21going to have to do that for AI video,
  40. 01:23AI image, AI audio, even AI games. And
  41. 01:26we want to be the place where like all
  42. 01:28the builders that are building with this
  43. 01:30technology, we want them to do that
  44. 01:32through FA.
  45. 01:37So my name is Burkai and I'm co-founder
  46. 01:39and CEO at FAL. FAL is a generative
  47. 01:41media platform for developers. We host
  48. 01:43models that can generate images, videos,
  49. 01:453D audio. Typically these models are
  50. 01:47very hard to host. So the problem we
  51. 01:49solve is like hosting these models as
  52. 01:52APIs which makes it very easy to consume
  53. 01:54for developers. We also have a inference
  54. 01:57engine that we built inhouse that is
  55. 01:59specifically optimized to run diffusion
  56. 02:02models to run two three times better. I
  57. 02:05think latency kills creativity, latency
  58. 02:07kills productivity. We work with
  59. 02:09customers like Adobe, Canva, Shopify,
  60. 02:12Perplexity. We are at uh 90 million
  61. 02:14annualized run rate revenue.

  62. How Two Immigrants Without Connections Built a $1.5B Startup

  63. 02:20My co-founder and I have been long-term
  64. 02:22friends. We've actually we're both from
  65. 02:24Turkey. I grew up in Turkey. I moved to
  66. 02:26the States for college. There was
  67. 02:28definitely like culture shock. I think
  68. 02:30like even a decade makes a big
  69. 02:32difference here, right? Like I moved to
  70. 02:34the States 2007. I think Facebook had
  71. 02:36just come out. It definitely felt like I
  72. 02:38wasn't as tapped in to like the culture,
  73. 02:40right? So there's like a big gap between
  74. 02:42how I how I grew up in Turkey and like
  75. 02:44what people like to do there versus like
  76. 02:46how they're in the US.
  77. 02:47I would say like school work felt a
  78. 02:50little bit easier than than I thought
  79. 02:52because we have a pretty good education
  80. 02:54system in high school in Turkey
  81. 02:57especially with maths and sciences. the
  82. 03:00the the biggest challenge actually was
  83. 03:02like the understanding the job market,
  84. 03:04how people do internships because
  85. 03:06immediately people start school and
  86. 03:10prepare for their summer internship and
  87. 03:12then the next summer and like they have
  88. 03:14a whole plan on how their career is
  89. 03:16going to happen. I didn't know I should
  90. 03:18be doing that. So couple years I wasn't
  91. 03:21really planning my internships towards
  92. 03:24my career. So that was a big big shock.
  93. 03:26I actually did an internship at Oracle
  94. 03:28like during college. I was working on
  95. 03:30some like fairly boring things in the
  96. 03:32beginning to be honest and I had started
  97. 03:34my green card process. This is like a
  98. 03:36very typical thing for immigrants in the
  99. 03:39US. You could kind of like be stuck in
  100. 03:41jobs if you start your green card
  101. 03:42process. Around 2015 was a very
  102. 03:45interesting time. Deep learning was just
  103. 03:47like kind of starting to become popular.
  104. 03:49I started getting really into it. Around
  105. 03:51that time I had a few other friends at
  106. 03:53Coinbase and Coinbase was a very small
  107. 03:55company back then. It's like maybe 40 50
  108. 03:57people. One of my friends told me like,
  109. 03:59"Hey, we're building a machine learning
  110. 04:01team." And I was like, "Okay, this
  111. 04:03sounds very interesting. Like I can go
  112. 04:04like do some deep learning in this new
  113. 04:06company and there's a lot of things
  114. 04:08things I can learn there." But I was
  115. 04:10mainly excited about like starting my
  116. 04:12own thing. I had actually talked to a
  117. 04:13lot of my founder friends seeing their
  118. 04:15experience. I had a lot of encouragement
  119. 04:17from friends to actually go and start my
  120. 04:19own thing.
  121. 04:20In the beginning of co Burka and I rent
  122. 04:23a house in Palm Springs for a while. We
  123. 04:25were talking about potentially starting
  124. 04:27a company but we didn't have particular
  125. 04:30angle or idea to go after. So we knew
  126. 04:33that we want to do this we would have to
  127. 04:35go through period of exploration where
  128. 04:37we find something that we are both
  129. 04:39passionate about. Purai quit maybe 4
  130. 04:41months before me and then I joined them.
  131. 04:44It is liberating because all my life I
  132. 04:47also had to deal with immigration work
  133. 04:50visa and then green card. That's one of
  134. 04:52the reasons actually I stayed at working
  135. 04:55at a big company. I wouldn't say that's
  136. 04:56the only reason but that's definitely a
  137. 04:59factor. And at that time all my
  138. 05:01immigration process had ended as well.
  139. 05:03That was also liberating in the sense
  140. 05:06that I didn't have to work for a big
  141. 05:08tech company to stay in the country. I
  142. 05:11could do whatever I want and I took the
  143. 05:12opportunity then.

  144. Why We Bet on Gen AI Video Instead

  145. 05:17Starting with the posthatit era it was
  146. 05:19brand new to everybody. It was such a
  147. 05:22new environment that like nobody knew
  148. 05:24where things are going. We started
  149. 05:26running image workloads and we saw a
  150. 05:29tremendous growth in the companies that
  151. 05:31are working with us. That made us really
  152. 05:33excited about the space. We also sat
  153. 05:35down and thought like where could this
  154. 05:37be going? 2 and a half years ago people
  155. 05:39saw LLMs and and Chhatra PT and and they
  156. 05:41sort of like drew out where this
  157. 05:43technology go and you know they
  158. 05:45immediately said okay you know we're
  159. 05:46going to AGI. We felt similarly about
  160. 05:49image models. We thought like as the
  161. 05:51models get better, there's going to be
  162. 05:52more capabilities. Quality is going to
  163. 05:54increase and the resolutions are going
  164. 05:56to increase and the controllability is
  165. 05:57going to increase.
  166. 05:58I think finding a niche market that is
  167. 06:00fast growing is the key to startup
  168. 06:03success. There are a lot of niche
  169. 06:05markets that stay niche and never grow.
  170. 06:08But we were lucky that market we
  171. 06:10operated in was very niche and small but
  172. 06:13also was growing incredibly fast. What
  173. 06:16changed after these big models were
  174. 06:18released is that you didn't have to
  175. 06:20train it anymore. You can just pick it
  176. 06:21off the shelf and start building
  177. 06:23something useful. And that meant the
  178. 06:26number of users maybe 10xed, 100xed,
  179. 06:29maybe over a millionx. We saw this this
  180. 06:31change early on and we decided, okay,
  181. 06:34this changes everything. Now that these
  182. 06:36models are going to be used by millions
  183. 06:37of people, we have to build systems that
  184. 06:40are ready for this change. And that's
  185. 06:42why we decided to build an inference
  186. 06:44platform early on. Another decision we
  187. 06:46had to make when the revenue was
  188. 06:48constant for a couple of months. One
  189. 06:50tempting thing we could have done run
  190. 06:53inference for LLM models as well.
  191. 06:55Focusing on imu and video is going to be
  192. 06:58an important differentiator. We already
  193. 07:00have a technical advantage because we've
  194. 07:03been working on on this type of models
  195. 07:05for a while. If we are wrong, we can
  196. 07:07always revisit our decision, but it's
  197. 07:09going to be harder for us to go from
  198. 07:11general to specific. So we tried to stay
  199. 07:14specific. I I think if you focus on a
  200. 07:16specific market, you get to work with
  201. 07:19your users in a closer manner. You
  202. 07:21understand their problems better. For
  203. 07:23us, this was image models and
  204. 07:25fine-tuning image models. In the
  205. 07:27beginning, all of our customers were
  206. 07:29doing very very similar things. So we
  207. 07:32were able to focus on it, get really
  208. 07:34good at it and differentiate ourselves
  209. 07:36from others. So our ultimate vision is
  210. 07:38basically we want to be the
  211. 07:39infrastructure layer for this new
  212. 07:41technology. Chat GPT moment for video is
  213. 07:43that I don't think we've hit it yet. I
  214. 07:45think there's a lot of signs like we're
  215. 07:47getting very close to it. Like if you've
  216. 07:48seen V3 it's close to the chat moment.
  217. 07:52It's a very capable model but I think I
  218. 07:54think we're still not there yet. But
  219. 07:56interestingly like you know now if you
  220. 07:58go to your Instagram Tik Tok feed like
  221. 08:00third half the the videos are AI
  222. 08:03generated right? it is already happening
  223. 08:04in a way. It's just happening in like a
  224. 08:06little bit of a slow motion. There may
  225. 08:08be a point this year is that we see like
  226. 08:11even better models that can actually
  227. 08:13like be edited uh real time and you can
  228. 08:15interact with the characters that are in
  229. 08:17the video and and you know generate very
  230. 08:19very interesting content. And we want to
  231. 08:21be the place where like all of this
  232. 08:24infrastructure is being hosted and all
  233. 08:26the builders that are building with this
  234. 08:28technology we want them to do that
  235. 08:30through fall.

  236. Kill Your AI Product If It Doesn't Sell on Day1

  237. 08:34I think there are two things happening
  238. 08:36with AI. People are willing to pay but
  239. 08:38there are questions about the quality of
  240. 08:40that revenue or how durable that revenue
  241. 08:43is going to be. I think AI markets are
  242. 08:45are incredible markets. Generative media
  243. 08:48is is one of those things where that it
  244. 08:51can be monetized right away. In the
  245. 08:53previous versions of internet
  246. 08:55businesses, people waited years and
  247. 08:57years to monetize their their business.
  248. 09:00First built a user base and then maybe
  249. 09:02try to monetize with subscription or
  250. 09:05ads. But with AI, people are willing to
  251. 09:07pay for it right away. The MVP you are
  252. 09:10building should be good enough for
  253. 09:12people to start paying. And it is really
  254. 09:14easy to get signs. the revenue numbers
  255. 09:17are increasing or not. Now monetization
  256. 09:20should be something a priority from day
  257. 09:22zero and it's actually easier for the
  258. 09:25founder to see if this is a good idea or
  259. 09:28if this is a good product by the revenue
  260. 09:31they are making from from the first day.
  261. 09:34Um we are very particular about what
  262. 09:36models we want to put because there's a
  263. 09:38lot of models out there. There's a lot
  264. 09:40of projects out there, research
  265. 09:42projects, even things that like big
  266. 09:44funded companies that put out there that
  267. 09:46are cherrypicked. Basically, like you
  268. 09:48take the results and you look at the
  269. 09:50good ones and you just use those for
  270. 09:52your demo or like for your launch. So
  271. 09:54that's called cherrypicking. So there's
  272. 09:56a lot of cherry picking happening in in
  273. 09:58models. When you when we look at the
  274. 09:59model, first thing we do is we take the
  275. 10:01model, we run the model and we run bunch
  276. 10:04of queries to understand like is it
  277. 10:06actually doing the thing that is
  278. 10:07advertised and then we we will go and
  279. 10:09optimize it and make sure like it can
  280. 10:11run faster and faster especially if
  281. 10:12there's a lot of demand. Developers like
  282. 10:15they spend so much time optimizing their
  283. 10:17like iterative loop, right? Like making
  284. 10:19sure that like once they do something
  285. 10:21they can see the result, they can see
  286. 10:22the tests and go iterate. So no one
  287. 10:25wants to like sit and wait around 5
  288. 10:27minutes for a video to generate in the
  289. 10:29future. This is going to be seconds.
  290. 10:31It's going to be real time and we're
  291. 10:32we're preparing ourselves from
  292. 10:34infrastructure standpoint for that
  293. 10:36future.

  294. Stay Small to Grow Big

  295. 10:40Scaling the company has been one of the
  296. 10:43like most exciting things about this job
  297. 10:45to be honest. We were like very small
  298. 10:47team like six people for the first two
  299. 10:50years almost. I think small teams is
  300. 10:53very important before product market
  301. 10:55fit. You actually do want to have like
  302. 10:57the smallest team that you can, right?
  303. 10:59And and and experiment and like have a
  304. 11:01small group making decisions and move
  305. 11:03really really fast. I think this like
  306. 11:05alignment with the company's mission is
  307. 11:08very important. This is something people
  308. 11:09talk about. This is another thing like I
  309. 11:11really learned from Coinbase. Like
  310. 11:13Coinbase early days like everyone was a
  311. 11:16crypto head. you would not find anybody
  312. 11:19that is not, you know, just insanely
  313. 11:21excited about crypto and and that
  314. 11:23created the foundation for the company
  315. 11:25that that is just so it's just so
  316. 11:28specific and so like, you know, just by
  317. 11:31default people are just excited about
  318. 11:32what they're working on, you know. But
  319. 11:34one of my criteria was that like I had
  320. 11:36to love I had to love it, you know. That
  321. 11:39was like super important to me. like the
  322. 11:41intersection of creativity and AI. I
  323. 11:43mean, there's like unlimited fun there.
  324. 11:47At least for me, I wake up every day,
  325. 11:49I'm very excited about like the next
  326. 11:51models that are released, what this tech
  327. 11:53where this technology is going, like
  328. 11:54what amazing things are people building.
  329. 11:56Uh it it is it is literally like the
  330. 11:58most fun thing I could be I could be
  331. 12:01doing. If I wasn't doing this, I would
  332. 12:03probably go play with these models
  333. 12:04myself. I I love this technology and and
  334. 12:07that's the thing that's like that that
  335. 12:09you know gives me a lot of drive
  336. 12:20[Music]