Subvolts take fal builds the infrastructure for generative video, enabling faster model deployment than training labs. Dec 10, 2025 - Developers
- Learn about the specialized infrastructure and optimization techniques for building and deploying generative video models at scale.
- Business
- Understand the emerging applications and economic models for generative media across various industries, including entertainment and advertising.
This interview provides a deep dive into the specialized infrastructure for generative video, a key area for AI media.
The Rise of Generative Media: fal's Bet on Video, Infrastructure, and Speed
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Sequoia Capital's Training Data podcast features fal founders Gorkem Yurtseven, Burkay Gur, and Head of Engineering Batuhan Taskaya. They discuss building infrastructure for generative media, highlighting video models' unique compute demands. fal's technology allows them to run over 600 models simultaneously, faster than training labs. The episode explores demand from AI studios, education, advertising, and Hollywood, comparing generative video's rise to early CGI.
This episode explains how a company called fal is creating the behind-the-scenes technology that powers AI-generated videos. They discuss how making videos with AI is much harder than making text and how their system helps run many video AI models quickly, even faster than the original creators.
This episode offers insight into the specialized infrastructure required for generative video, a rapidly evolving field. It's for developers and businesses exploring AI-driven media creation. Listeners learn how companies like fal are tackling the unique challenges of video generation, which differ significantly from LLMs. This work supports advancements seen in frontier labs experimenting with multimodal AI and video output, indicating a shift toward more complex AI media generation.
- HostSequoia Capital
- GuestGorkem Yurtsevenfounder, fal
- GuestBurkay Gurfounder, fal
- GuestBatuhan TaskayaHead of Engineering, fal
- Episode Title: The Rise of Generative Media: fal's Bet on Video, Infrastructure, and Speed
- Show: Training Data (an official AI show)
- Length: 62 minutes
- fal's capability: Run more than 600 image and video models simultaneously
- fal's speed claim: Often faster than the labs that trained them
fal is building the infrastructure layer for the generative media boom. In this episode, founders Gorkem Yurtseven, Burkay Gur, and Head of Engineering Batuhan Taskaya explain why video models present a completely different optimization problem than LLMs, one that is compute-bound, architecturally volatile, and changing every 30 days. They discuss how fal's tracing compiler, custom kernels, and globally distributed GPU fleet enable them to run more than 600 image and video models simultaneously, often faster than the labs that trained them. The team also shares what they’re seeing from the demand side: AI-native studios, personalized education, programmatic advertising, and early engagement from Hollywood. They argue that generative video is following a trajectory similar to early CGI—initial skepticism giving way to a new medium with its own workflows, aesthetics, and economic models.Hosted by Sonya Huang, Sequoia Capital
- LLMs
- Gorkem Yurtseven, Burkay Gur, Batuhan Taskaya
- Large Language Models, AI systems that understand and generate text.
- compute-bound
- Gorkem Yurtseven, Burkay Gur, Batuhan Taskaya
- A task limited by the processing power available, rather than memory or other factors.
- tracing compiler
- Gorkem Yurtseven, Burkay Gur, Batuhan Taskaya
- A tool that optimizes code by tracking its execution path.
- custom kernels
- Gorkem Yurtseven, Burkay Gur, Batuhan Taskaya
- Specialized, highly optimized code routines for specific hardware operations.
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
Questions people ask
- What makes video models different from LLMs?
- Video models present a compute-bound optimization problem that is architecturally volatile and changes rapidly, unlike LLMs.
- What kind of demand is fal seeing for generative video?
- Demand comes from AI-native studios, personalized education, programmatic advertising, and early engagement from Hollywood.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.