"It is difficult to succeed in the physical AI business by merely selling robots. Ultimately, companies that provide services while accumulating on-site data will dominate the market," said Choi Hong-seob, CEO of Mind AI, in a recent interview at the company's office in the Pangyo IT Center. He emphasized that the core competitiveness of physical AI, which is gaining attention as the next growth driver after generative AI, lies in its 'service-based business model.' Mind AI aims to evolve into a full-stack service company that integrates everything from data collection to learning and operation, rather than just selling robots and AI models.
Choi joined Minds Lab (now Mind AI) in 2017 and has led the shift to a physical AI strategy. He currently serves as co-CEO alongside Yoo Tae-jun, overseeing technology development and commercialization.
Founded in 2014, Mind AI became a publicly traded AI company on the KOSDAQ in 2021. Recently, it has rapidly expanded its business, focusing on applying AI in real industrial settings as a future growth pillar. In 2023 and 2024, over 95% of its total revenue came from agent AI services, including enterprise AI platforms and AICC. However, last year, revenue from physical AI surged from 42 million won to 2.648 billion won, increasing its share of total revenue to 27%.
Choi identified the most important milestone of the year as the government project to develop a physical AI world model in collaboration with LG Electronics. This project aims to develop a domestic physical AI foundation model applicable to manufacturing sites and validate its practical use in real industrial environments.
He stated, "We plan to unveil our own physical AI model by the end of the year and are preparing a large-scale event related to it. We also plan to introduce our own dual-arm robot that can operate in actual factories within the year."
Mind AI's decision to develop its own robots stems from recognizing the limitations of the traditional robot sales model in the physical AI business. Simply delivering robots does not mean they can be immediately deployed in production environments; data collection, AI learning, and iterative performance improvements must follow.
Choi explained, "Just bringing one robot into a factory does not mean it can start working right away. We need to collect process data and use GPUs to train the model, and when new processes arise, we must repeat the same steps. It is virtually impossible for customers to carry out all these processes themselves."
As a result, Mind AI plans to shift to a service model starting next year, providing robots and AI without the initial investment burden for customers. Mind AI will handle data collection, remote operation, model training, and maintenance.
The market atmosphere for physical AI is also changing rapidly. Choi noted, "Last year, we were bidding for small-scale PoCs (proof of concepts) worth millions of won, but now project prices have increased by three to four times. Inquiries are coming in from large manufacturing projects, construction, and defense sectors, and contract sizes are expanding significantly."
The government research and development (R&D) environment has also evolved. While government projects focused on LLMs until last year, they have expanded into the physical AI sector this year, although the number of companies capable of executing these projects is extremely limited. He remarked, "This year, the number of government projects is the highest since our founding, and while thousands of companies compete in LLMs, the supply in physical AI is insufficient, leading to continuously rising project prices."
Choi pointed out that the high entry barriers in physical AI stem from the need to secure not only AI but also hardware, simulation, and on-device technologies. He stated, "We are now in an era where hardware is designed to ensure AI operates effectively, rather than AI determining hardware. It is challenging to secure commercialization competitiveness by developing AI and robots separately."
Initially, Mind AI plans for humans to remotely operate robots while accumulating data, and in the long term, as AI performance improves, one person will manage multiple robots. Choi said, "At first, one person will manage one robot, but as that number increases to two, five, or ten, profitability will improve significantly. Even conservatively, we aim for a business model with an operating profit margin of over 50%."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.