Key Summary
- 49.2% for large enterprises, 4% for SMEs
- A structure where data flows back, not just GPU count
- Start by dividing data that can be shared
- Where does our process data actually live?
In a survey of 504 manufacturing companies conducted by the Korea Chamber of Commerce and Industry (KCCI) in November 2025, AI adoption in management stood at 49.2% for large enterprises and 4% for SMEs. Sidae's September 22, 2026 feature series 'AI Full-Stack Korea's Leap,' in its fourth installment 'Korea's Manufacturing Floor: Where Physical AI Will Be Decided,' pointed to the manufacturing site as Korea's key differentiator, noting that expanded investment by large corporations has not sufficiently translated into AI adoption among smaller manufacturers.
How Are Large Corporations Investing in Manufacturing AI?
NVIDIA's October 2025 announcement of its GPU supply plan for Korea totals over 260,000 units, with the allocation structured as follows.
- Samsung Electronics, SK Group, Hyundai Motor Group: approximately 50,000 units each
- Naver Cloud: approximately 60,000 units
- Government-led AI infrastructure: approximately 50,000 units
Samsung Electronics is pursuing semiconductor manufacturing innovation, SK Group is building cloud infrastructure to support industrial AI and robotics, and Hyundai Motor Group is advancing manufacturing and autonomous driving AI development.
Hyundai Motor Group is pursuing plans to deploy humanoid robots on automobile production lines, with Atlas set to be introduced at Hyundai Motor Group Metaplant America (HMGMA) in Georgia, USA in 2028, expanding to Kia's Georgia plant in the second half of 2029. Supporting this effort, the Robotics Metaplant Application Center (RMAC) opened in the US in June 2026. Serving as a hub that recreates real manufacturing environments to train and validate robots, it is designed to use data gathered from factories to train robots and then reapply improved technology back to production sites.
HD Hyundai is targeting the realization of an intelligent, autonomously operated shipyard by 2030, and is developing a 'Shipbuilding AI Master Agent' that draws on the knowledge and experience of skilled technicians.
How Wide Is the Gap With Smaller Manufacturers?
- KCCI November 2025 survey (504 manufacturers): 82.3% of respondents said they do not use AI in management, with adoption rates at 49.2% for large enterprises and 4% for SMEs.
- 2024 Smart Manufacturing Innovation Survey: Only 0.1% of small and medium manufacturers had actually adopted manufacturing AI, rising to just 1.7% when including companies with adoption plans.
- Gap in generative AI adoption between large and small firms: 9.2 percentage points in services versus 24.2 percentage points in manufacturing — the largest gap of any sector.
What Support Are the Government and China Providing?
The M.AX (Manufacturing AI Transformation) Alliance, launched in September 2025, had over 1,500 participating organizations as of August 2026. The Ministry of Trade, Industry and Energy announced plans to invest 700 billion won in 2026 centered on the alliance, supporting areas including the following.
- Joint use of manufacturing data
- Development of sector-specific AI models
- Regional AI transformation
In China, eight ministries including the Ministry of Industry and Information Technology jointly released implementation guidelines for the 'AI+Manufacturing' special initiative in January 2026. The plan sets targets of building 1,000 industrial AI agents and 100 high-quality industrial datasets by 2027, along with disseminating 500 representative application cases and fostering AI service companies specialized for manufacturing.
Choi Soo-hwan, Executive Director at RealWorld, assessed that "while becoming the world's number one or two in LLMs is difficult, physical AI itself is an area where Korea is in a uniquely favorable position, with the potential, demand, and supply all in place to lead." However, he added that "unlike China, no single company in Korea can do everything," stressing the need to build a collaborative structure bringing together hardware companies, model companies, and chip makers producing processors like NPUs.
Implications for Korean Enterprise AX
What stands out in the large-corporation cases is not GPU volume but the structure by which field data flows back into AI. RMAC trains robots using factory data and reapplies improved technology back to the field, while HD Hyundai's 'Shipbuilding AI Master Agent' is being developed to leverage the knowledge and experience of skilled technicians. Smaller manufacturers considering a similar direction would do well to first check where and in what form their own process data exists, rather than starting with robots or computing resources.
The inclusion of joint use of manufacturing data among the M.AX Alliance's support areas can be read in the same light. SMEs considering participation should first sort out which data can be shared externally and which cannot. Who will handle data collection, refinement, and on-site staff training should also be decided before adoption.
There is one thing worth checking this week: make a list of which processes at your plant record and store process data, and who is currently using that data.
Source: [Sidae Report] Korea's Manufacturing Floor: Where Physical AI Will Be Decided