Interview, Fireside Chat
Inside fal’s $1.65B Series E & F Led by Sequoia & Bessemer
- Sorcery identifies three primary AI product-market fits: search/chat interfaces, coding agents, and generative media, with the latter being the fastest-growing sector.
- Generative media has reached a scale where individual professionals spend $10,000–$20,000 monthly on cloud compute, comparable to coding agents.
- The company was 100% surprised by the magnitude of demand for high-cost video models, specifically noting that even a jump to $5 per 5-second clip faced "essentially infinite demand."
- Sorcery's revenue mix has fluctuated significantly: video models initially rose to 70% of revenue, while the emergence of image editing models revitalized image generation to a 50/50 split.
- Current best-in-class video models generate 10–12 seconds of content, with new iterations targeting 30-second durations; costs increase exponentially with resolution and duration due to attention mechanisms.
- Advertising is identified as the largest use case for generative media, specifically for UGC-style content where rapid iteration and A/B testing of thumbnails and hooks drive ROI.
- The company expects AI-driven reinforcement learning for automated ad optimization (testing thumbnails, titles, and copy) to become an industry norm within a few years.
- Sorcery evolved from a compute infrastructure provider in the 2021 data era to a specialized "generative media" platform, pivoting after realizing image and video models had superior commercial potential.
- The team size is approximately 150 people, with a flat organizational structure limited to two levels of reporting, relying on "Super ICs" (individual contributors) who can lead teams without pure management roles.
- Hiring criteria prioritize "agency," "hunger," and technical excellence, with early team members largely recruited from Coinbase and Amazon to leverage pre-existing trust and domain knowledge.
- Sorcery operates as an infrastructure partner for model providers and an API partner for applications, managing compute procurement ahead of model releases to mitigate scarcity.
- The generative media market is highly fragmented with 5–10 companies across multiple modalities (image, video, 3D, audio), preventing immediate market consolidation.
- Open-source models remain a critical marketing tactic for smaller labs, allowing studios to fine-tune models for proprietary IP, while closed-source models are preferred for security and trust in enterprise settings.
- Hollywood adoption has not yet impacted labor significantly; legal frameworks and quality concerns have slowed integration, though usage is accelerating.
- Sorcery projects that within 12 months, "timeless AI content" (e.g., feature films or animations) will emerge that retains quality and entertainment value years after release.
- The founders are bullish on "world models" (interactive, persistent 3D environments) potentially becoming a fourth major product-market fit in AI gaming.
- Sorcery's name derives from "Features and Labels," traditional machine learning terms, while also referencing Turkish coffee fortune telling.
- Hot Take 1: Margins are shifting from the application layer to the compute path; application-layer margins are near zero due to marketing costs and competition, while compute innovation is the only viable path to profitability.
- Hot Take 2: Coding is evolving into a subset of generative media, as developers increasingly build "programmable video" and content-rich applications rather than traditional websites.
- Speculation: A merger between OpenAI and Meta is deemed strategically plausible to dominate the consumer market, though no such deal is confirmed.
- Jack Dorsey and Ali Koç (Chairman of Anadolu Efes) are cited as key leadership inspirations for their reinvention capabilities and range.