Twelve Labs Attracts $150 Million Investment with Video AI

Twelve Labs has secured a total of $100 million (approximately ₩150 billion) in Series B funding, co-led by NEA and Naver Ventures. Strategic investors include NVIDIA, Amazon, Snowflake, and Databricks, with Index Ventures and World Labs founder Alex Wang also participating.

In its early stages, Twelve Labs experienced severe financial difficulties. The CTO earned living expenses by tutoring for the SAT, and CEO Lee Jae-sung took out loans to purchase equipment and pay salaries for his colleagues. At the time, many were skeptical about investing ₩300 million in model training.

This AI model quickly searches and analyzes vast video data, helping with new content creation and being used for sports analysis and highlight generation.

Twelve Labs established an office in Itaewon to commemorate the founders' bond formed while serving together in the military in Yongsan. This location symbolizes their initial aspirations and serves as a strategic hub for fostering an international atmosphere where Silicon Valley and Korean team members can work together.

Twelve Labs operates dual offices in Seoul and San Francisco, clearly dividing roles. The Seoul office focuses on developing models and agents currently used by customers, while the San Francisco office builds wrappers on top of proprietary models, drives go-to-market (GTM) strategies, and handles business expansion.

It aims to be a next-generation platform that freely manipulates and expands data, expected to revolutionize how users access and utilize video data.

One of Twelve Labs' core technologies is making the semantic data layer more robust and usable. This layer is essential for robotics and vision-language-action (VLA) models to understand and process video data like humans.

From 2018 to 2020, during their military service, they encountered Transformer papers, recognizing the difficulty and potential of video understanding technology, which led them to co-found the company.

While many researchers at the time focused on text models, Twelve Labs chose video-based AI, an untouched field. They believed that to create AI that truly understands the world, learning must begin with video data, which is the signal most similar to how humans learn about the world. They also believed that challenging a difficult field would allow for a unique positioning.

For example, just as a person might watch Kim Yuna's long spin casually but focus intently on a triple axel, the AI system is designed to learn by concentrating on important parts of the video.

Park Yong-jung, a partner at Naver Ventures, presented two important criteria for founders to persuade investors. First is 'Founder-Market Fit,' assessing how well the founder aligns with the market. Second is whether the founder is tenacious and resilient, not simply following trends, but rather sticking with the problem and seeing it through even with limited resources.

The founders believed that Silicon Valley offered a stage where their challenging and difficult technology would be more generously supported and evaluated. Thus, they decided to conduct initial technology development in Korea but pursue their main competition in Silicon Valley.

In its early stages, Twelve Labs struggled for survival. The CTO earned living expenses by tutoring for the SAT, and CEO Lee Jae-sung used previously earned money and loans to buy equipment and pay salaries for his co-founders. At the time, many found it hard to understand investing ₩300 million in model training. They also participated in the tech startup accelerator 'Techstars'.

Park Yong-jung, a partner at Naver Ventures, stated that Index Ventures began to think video would follow text around 2022. Index Ventures is also well-known as an early investor in global unicorn companies like Entropic, Notion, and Cohere.

While this might seem small compared to current seed round sizes, it was a considerably large investment at the time.

The US perceives AI as a 'war,' taking an aggressive stance to seize hegemony, while Korea focuses on supplying global standards through technology utilization, adopting a somewhat defensive perspective.

Simultaneously, it also helps American companies it invests in enter the Asian market, contributing to strengthening the position of Korean companies in the global AI ecosystem.

Park Yong-jung emphasized that distinguishing between 'Korean AI' and 'American AI' is meaningless in today's global AI market. He explained that the moment a model or agent is released, people worldwide use it and it faces judgment in the market, so national borders cannot hinder the spread of technology. This suggests that even models developed in Korea can successfully compete in AI hubs like Silicon Valley.