EO StudioHow to Build $1.5B AI Startup in Just 3 Years | fal's co-founders
CHAPTERS
- 0:00 – 0:29
Betting early on image models while everyone chased LLM hype
The founders describe the moment when language models dominated attention, but they saw a similar inflection coming for image generation. Their thesis: niche today can become massive tomorrow if model capability and accessibility keep compounding.
- •LLMs were the main hype cycle; fal believed image models had similar upside
- •A deliberate choice to explore where image/video capabilities could go
- •Seeing “niche” as a launchpad rather than a limitation
- •Early conviction as a strategic advantage
- 0:29 – 1:05
Move 100x faster: day-zero releases and infrastructure urgency
fal’s operating principle is extreme speed—shipping models immediately and removing friction for builders. They argue that in fast-moving AI markets, slow iteration is indistinguishable from failure.
- •Day-zero model releases to stay ahead of the market
- •Speed as a competitive moat in rapidly evolving AI
- •Getting models “in front of people” with minimal usability barriers
- •Warning against spending months/years on ideas without traction
- 1:05 – 1:37
Series C, $1.5B valuation, and the platform vision for generative media
Burkay lays out fal’s current milestone—$125M Series C at a $1.5B valuation—and the broader ambition to become the infrastructure layer for generative media. The company is preparing for a coming “ChatGPT moment” in video and beyond.
- •$125M Series C and $1.5B valuation; readiness to scale
- •Vision: be the default platform for builders using generative media
- •Expectation that AI video’s breakout moment is near but not fully here
- •Expansion horizon: video, image, audio, and potentially games
- 1:37 – 2:07
What fal actually does: generative media APIs and a faster inference engine
fal positions itself as a developer platform that hosts difficult-to-run generative models behind simple APIs. A core differentiator is an in-house inference engine optimized for diffusion, aiming to reduce latency because latency suppresses creativity and iteration.
- •Generative media platform for developers (image, video, 3D audio)
- •Model hosting as APIs to make advanced models easy to consume
- •In-house inference engine optimized for diffusion (2–3x improvements)
- •Latency framed as a blocker to creativity and productivity
- 2:07 – 2:37
Credibility signals: enterprise customers and $90M run-rate revenue
They cite well-known customers and current revenue scale to show market adoption. The emphasis is that developers and large companies will pay when the infrastructure reliably turns cutting-edge models into usable products.
- •Customers mentioned: Adobe, Canva, Shopify, Perplexity
- •Reported $90M annual run-rate revenue
- •Developer-first positioning tied to reliability and performance
- •Scaling demand driven by model adoption across industries
- 2:37 – 3:26
Immigrant founder story: culture shock, career navigation, and constraints
Both founders recount moving from Turkey to the U.S. and adjusting to cultural norms and career planning. They highlight how immigration logistics (visas/green cards) can shape risk tolerance and early career choices.
- •Cultural adaptation after relocating to the U.S.
- •Surprise at how early and strategically peers plan internships/careers
- •Immigration and visa constraints influencing job decisions
- •Shared background and long-term friendship as co-founder foundation
- 3:26 – 4:20
From Oracle to ML curiosity to Coinbase: building toward entrepreneurship
Burkay traces the path from a conventional big-company start to discovering deep learning’s rise around 2015. Exposure to Coinbase’s early days and encouragement from founder friends pushed him toward starting a company.
- •Early career at Oracle and the ‘stuck’ feeling during green card process
- •Deep learning becoming salient around 2015
- •Coinbase as a small, fast-learning environment building ML capabilities
- •Desire to found a company strengthened by founder-network encouragement
- 4:20 – 5:17
Palm Springs exploration phase: quitting jobs and committing to the search
During early COVID, the founders created space to explore ideas without a fixed angle. Completing immigration processes reduced perceived downside and enabled them to fully commit to entrepreneurship.
- •Deliberate exploration period before choosing a thesis
- •Burkay quits first; Gorkem joins later
- •Immigration completion as a catalyst for taking risk
- •Using time/space to find shared passion and direction
- 5:17 – 5:58
Post-ChatGPT uncertainty to image workload traction: forming the core thesis
They describe the post-ChatGPT landscape as new and ambiguous, where conviction had to be built from observed demand. Running image workloads and witnessing rapid customer growth reinforced their belief that images/video would follow the LLM trajectory.
- •Early generative era felt directionless; few clear playbooks
- •Customer growth from image workloads validated the opportunity
- •Analogy: people extrapolated LLMs to AGI; they extrapolated image models to broad capability gains
- •Expectation of improvements in quality, resolution, and controllability
- 5:58 – 7:36
The niche that exploded: inference platform decision and staying specific
Gorkem explains why a fast-growing niche is ideal—and why generative media shifted dramatically once models could be used “off the shelf.” When usage could jump orders of magnitude, fal chose to build an inference platform and resisted the temptation to also serve LLM inference just to smooth revenue.
- •Success pattern: niche markets must be both small and rapidly expanding
- •Off-the-shelf models drove massive user growth (10x to potentially million-x)
- •Decision: build systems ready for large-scale demand via an inference platform
- •Strategic focus: image/video as differentiation; easier to broaden later than to re-specialize
- 7:36 – 8:34
AI video’s ‘ChatGPT moment’: slow-motion takeover and what’s next
Burkay argues video hasn’t yet had its defining breakthrough moment, though models are close and AI content is already saturating social feeds. The next phase may be real-time editability and interactivity, and fal wants to host the infrastructure powering those products.
- •Video breakthrough is approaching but not fully realized
- •AI-generated videos already common on TikTok/Instagram feeds
- •Near-term frontier: real-time editing and interactive characters
- •fal’s goal: be the hosting/infrastructure layer for builders in this era
- 8:34 – 9:34
Monetize from day zero: ‘kill it if it doesn’t sell’ mindset
They contrast AI with prior internet businesses: users are willing to pay immediately if the product delivers value. That makes revenue an early signal—founders should prioritize monetization and quickly abandon products that don’t convert.
- •AI products can monetize immediately vs. years-later internet models
- •MVP should be good enough to trigger real payments
- •Revenue as a fast feedback loop on product viability
- •Monetization treated as a day-zero priority
- 9:34 – 10:35
Model quality, anti-cherry-picking checks, and the race to real-time generation
fal evaluates new models skeptically, testing beyond curated demos and then optimizing for performance under real demand. They emphasize that developer iteration loops require low latency, moving from minutes today to seconds and ultimately real time.
- •Widespread ‘cherry-picked’ demos make independent evaluation essential
- •fal runs extensive queries to verify model claims
- •Optimization focus: faster inference as demand increases
- •North star: real-time generation to support rapid developer iteration
- 10:35 – 12:40
Stay small to grow big: tiny teams, alignment, and founder-level obsession
They argue small, aligned teams outperform before product-market fit because decisions are faster and experimentation is tighter. Burkay also frames mission-fit and genuine enthusiasm for the tech as a key hiring and culture ingredient, drawing a parallel to Coinbase’s early crypto-native team.
- •Team stayed ~6 people for nearly two years to maximize speed pre-PMF
- •Small teams improve alignment, decision-making, and iteration cadence
- •Mission obsession as culture moat (Coinbase ‘crypto-head’ analogy)
- •Founder motivation rooted in the creative possibilities of AI models
