TwelveLabs
Foundation model · US
Multimodal video-understanding models (Marengo, Pegasus) that index and reason over existing video — a perception layer for physical AI.
Marengo 3.0 (video embedding) + Pegasus 1.5 (video→structured data) via API.
Research mapping
Physical AI / Perception & Sensing (L7) / Foundation model
Evidence boundary
Original analysis and independent corroboration remain under development. Commercial scale and durable technical advantage remain under review.
What makes it defensible
Video-native foundation models built from scratch (Marengo embeddings, Pegasus video-to-structured-data) rather than a language model adapted to frames - which is why enterprise data platforms embed it as their video layer.
Position in the stack: Model and API layer sitting between raw video and enterprise data platforms. Hard to copy because it requires owned video training corpora and a purpose-built architecture; an API wrapper over a text LLM cannot reproduce the temporal grounding.
Who buys it
Named customers: Databricks, Snowflake, SK Telecom
Founding team
Jae Lee — Co-founder & CEO — Co-founder and CEO of TwelveLabs; leads the company's video-native multimodal model strategy.
SourceFinancing detail
- Last round
- Series B ($100M, Jul 2026)
Investors: NEA, NAVER Ventures, NVentures, Amazon, Radical Ventures, Index Ventures
Mapped funds and investors
Direct relationships supported by published evidence.
Reported holdings as-of
No sourced relationships mapped yet.
Disclosed investment history
No sourced relationships mapped yet.
Showing up to four relationships in each group. ETF weights describe the fund’s portfolio, not ownership of this company. Private investment history does not establish current ownership.
Timeline
-
Product
Jockey reaches general availability — adds ChatGPT MCP support alongside Claude, URL/Google Drive/device ingestion, and MP4 reel export
Source