Sora lost to Google and ByteDance
Why Sora failed
AI video is shifting from a frontier model demo into a scale business, which favors companies that can pay huge compute bills while also owning distribution. Google already pipes Veo into Gemini, Vertex AI, Flow, YouTube Shorts, and Google Vids, while ByteDance has kept improving Seedance into a multimodal model for text, image, audio, and video inputs. That leaves smaller winners to specialize in workflows, not foundation models, and leaves OpenAI prioritizing coding and enterprise where revenue density per GPU is higher.
-
Google is not just training Veo, it is turning it into a product family. Veo 2 launched in Gemini in April 2025, Veo 3 reached Vertex AI in June 2025, and by early 2026 Veo 3.1 was spread across Gemini, API, Vertex AI, Flow, and Google Vids. That kind of bundled distribution is hard for a standalone lab to match.
-
ByteDance is playing the same game from the other side, pairing deep cash flow with a fast moving model org. Seedance 1.0 launched in June 2025 with 1080p and multi shot generation, then Seedance 2.0 added a unified audio video architecture and broader editing controls in February 2026, showing continued investment rather than retrenchment.
-
The startups that remain strongest in video are selling finished jobs, not raw model access. Runway helps filmmakers and VFX teams cut the cost of a shot from about $350 to about $10, and reached an estimated $90M ARR by June 2025. Synthesia packages avatars, scripts, translation, and screen recording for enterprise training and sales videos, reaching an estimated $145.9M ARR by September 2025.
From here, AI video looks like cloud infrastructure underneath and workflow software on top. Google and ByteDance are positioned to keep pushing model quality and price down, while companies like Runway, Higgsfield, and Synthesia compete on the interface, templates, editing stack, and distribution into specific creator and business use cases.