Korean-founded, US-based TwelveLabs raises US$100m Series B co-led by Korea's NAVER and US firm NEA
Amazon joins with a multiyear AWS Trainium deal; total funding tops US$207m for video-native foundation models Marengo and Pegasus
Add to a list
No lists yet.
Summary
TwelveLabs, founded by Korean engineers and headquartered in San Francisco, raised a US$100m Series B on 1 July 2026 co-led by US firm NEA and Korea's NAVER Ventures, with Amazon, Radical Ventures, Index Ventures, Korea Investment Partners, Quadrille Capital and Red Bull Ventures participating. Total funding now tops US$207m. The startup builds video-native multimodal foundation models, Marengo for search and Pegasus for text generation, and is expanding into an agentic video-reasoning system. It signed a multiyear commitment to run inference on Amazon's Trainium chips, with new models debuting on AWS. The prior US$50m Series A in 2024 was co-led by NEA and Nvidia's NVentures.
The split
US outlets frame the round through Amazon: a hyperscaler locking in a video-AI supplier and steering it onto its own silicon. Korean coverage frames the same deal as a national win, leading with NAVER and Korea Investment Partners and denominating it in won. The two readings capture the company's real structure, a US-domiciled firm with a heavily Korean cap table and an Amazon compute dependency, an axis distinct from the US frontier labs.
By the numbers
- US$100m, Series B size.
- US$207m+, total funding to date.
- US$50m, 2024 Series A (NEA and Nvidia NVentures).
- 150bn won, the round as reported in Korea.
- July 1, 2026, announced.
Why it matters
Video is the least machine-readable large data type, and whoever makes it searchable captures surveillance, media and robotics-training markets. The Korea-US structure and Amazon's chip tie-in show hyperscalers buying strategic stakes in narrow foundation-model firms to feed proprietary silicon, a quieter consolidation than the headline frontier-lab rounds.
What to watch
- Whether AWS Trainium can serve video models competitively against Nvidia.
- NAVER's push to fold TwelveLabs capability into its own Korean and Japanese products.
- Whether video-native models generalise into robotics and world models.