CNBC International Live · Sep 28, 2026
With Daegwon Chae, Partner at Bond Capital
Daegwon Chae said about 90% of the white-collar work people do needs no frontier-level intelligence, and that cheaper AI models could make that part of the work almost free.
The debate on air was whether a slowdown at the leading AI labs spells trouble. Chae said it marks a new chapter instead: competition between models, and possibly their commoditization, as companies find cheaper ways to buy intelligence.
"The future is bright because there will always be a portion of the market that will pay a reasonable amount for high-quality and differentiated intelligence."
Daegwon Chae, Partner at Bond Capital, on CNBC International Live. He leads Bond's investments in OpenAI, Databricks and Cognition, among others, sits on the boards of Halter, inDrive, Suno and Toss, and was previously a principal at Kleiner Perkins.
The full segment is covered here so you can skip it.
Here are the 6 arguments that matter.
Key Takeaways
AI is moving from frontier-model dominance to model competition, and maybe commoditization
About 90% of white-collar tasks do not need PhD-level intelligence
Cheaper models could push the cost of that work toward the price of computing and electricity
The margin will sit with frontier models and with application companies, not with basic tokens
AI labs have a commercial reason to regulate themselves, because rogue models hurt their own customers
Orbital data centers work on paper, but whether they survive a capital market cycle or two is open
China has real advantages in supply chains, grid access and robotics, but Bond stays focused on the US
1. AI's Next Chapter
The interviewer said a slowdown at some of the frontier AI labs might cause concern but could also create dislocations and opportunities, and asked where Bond Capital sees them.
He describes AI's history as a set of chapters
Chae said the prologue was the research phase, when researchers at DeepMind and OpenAI built AI that beat humans at the board game Go and the video game Dota. The ChatGPT launch came next, when AI reached the mass market. Then came enterprise adoption.
"That then transitioned towards what I'd call the enterprise adoption moment where you had Anthropic take the helm as the most important company in the world as we realized the commercial value of AI was so enormous" — Daegwon Chae
He said the industry is now heading into a chapter defined less by the dominance of frontier models, the most capable systems from the leading labs, and more by competition between them. Companies are finding cheaper ways to run AI, including open-source models.
AI agents that act, not just recommend
The second chapter Bond watches is AI agents that take actions for people. He said a product that orders groceries sounds trivial, but that moving AI from recommending flights, hotels and restaurants to booking them on a person's behalf would rewire the internet, which was built on recommendation engines.
The interviewer said that is exactly where regulation, over-regulation and rogue AI agents become a problem.
