Key Takeaways
- • Jeff Herbst, co-founder of GFT Ventures and former NVIDIA vice president of business development, told the Chosun Ilbo that AI investing is "a market where speed determines victory," arguing that startups failing to scale quickly are likely wrong investments.
- • GFT Ventures, founded in 2021 by Herbst and CEO Jay Eum, manages approximately $160 million (214 billion Korean won) and has invested in more than 20 companies, with Korean backing from Mirae Asset and SK Hynix.
- • Herbst disputes AI bubble concerns, stating demand for high-performance computing exceeds supply by at least 10 to 100 times, with data center, power, and memory bottlenecks potentially lasting up to a decade.
- • Herbst names agentic AI and physical AI (robotics) as next-generation investment themes and sees more promise in AI applications built on existing models in South Korea than in proprietary model development.
- • Herbst says founder resilience, shaped by his NVIDIA experience, is a key criterion when evaluating founders.
Summary
GFT Ventures was founded in 2021 by Herbst and CEO Jay Eum, formerly of Translink Capital. According to the report, the firm manages approximately $160 million (214 billion won) and has invested in over 20 companies. It was launched with Korean capital from Mirae Asset and SK Hynix. Eum attended venture investment meetings during President Lee Jae Myung's July visit to Silicon Valley. Herbst noted that a $1 billion exit was once considered a success, but investors now expect early-stage AI companies to grow into multi-billion-dollar businesses, and that current fund structures allow entire funds to be returned even at higher valuations and larger check sizes.
Herbst rejected AI bubble concerns, arguing that supply constraints are the greater issue. He said demand for high-performance computing exceeds supply by at least 10 to 100 times, and that bottlenecks in data center sites, power supply, and memory could take up to a decade to resolve. He praised NVIDIA CEO Jensen Huang as someone who anticipates developments "10, 20, or 30 steps ahead" of the industry, and said NVIDIA's culture treated failure to learn and improve as a greater problem than failure itself. That experience, he said, shaped his emphasis on founder resilience.
Looking ahead, Herbst identified agentic AI, where systems autonomously execute tasks, and physical AI such as robotics as next-generation investment themes. For South Korea, he favored AI applications built on existing models in areas including video, healthcare, and security over developing proprietary models, saying "Now is the time for genuine AI app companies to emerge." Over a 10-year horizon, he pointed to personalized AI that understands individual memories, experiences, and traits, and questioned whether platforms or individuals will hold sovereignty over that data.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.