Korea’s physical AI ambitions meet the factory’s bottom line

For startups training robots to work in factories, South Korea offers a practical advantage: Industrial customers are nearby.

Within a few hours’ drive, they can reach manufacturers across sectors, including automotive, batteries, displays, and semiconductors, says Terry Moon, co-founder and CEO of CarbonSix.

Based in South Korea and the US, the company develops models for manufacturing automation and software to train them on specific tasks. It raised US$40 million in July.

It helps that Korean manufacturers are already using robots extensively. The country has the highest density of industrial robots in manufacturing, while a shrinking workforce is driving further demand for automation, according to Young Jae Jang, professor at the Korea Advanced Institute of Science and Technology (KAIST).

In June, the government also designated physical AI as one of its three national megaprojects, alongside semiconductors and AI data centers.

But having access to manufacturers doesn’t necessarily make it easier for physical AI startups in South Korea to sell new tech. A manufacturer willing to provide data or host a trial still needs to be convinced that the resulting system can meet its production targets and justify its cost.

Still, their experiences offer insights into what physical AI startups in Southeast Asia can learn about commercialization.

From demonstrations to deployment

Training robots requires real-world data, something physical AI startups everywhere struggle to collect.

A manufacturer won’t “tolerate a useless robot” wandering around its facility for a year while it gathers data needed to become useful, says Moon of CarbonSix.

In South Korea, many manufacturers already own robotic arms from brands such as ABB, Fanuc, and Kuka. They want to add AI capabilities to their existing equipment rather than replace hardware they already trust, he adds.

Moon describes the “pretty scary spreadsheets” some prospective buyers bring, detailing their performance expectations and how quickly they expect to recoup their investment.

The case for repeat orders

The Korean approach to physical AI

What Southeast Asia can learn