
MakinaRocks (477850.KQ), a physical artificial intelligence company, said it aims to build a fully autonomous manufacturing plant by developing AI that works reliably in unpredictable factory environments.
"Real sites such as semiconductor plants never run on frontier AI models alone," CEO Yoon Sung-ho said at ATTENTION 2026, an AI conference held on the 3rd at the Westin Josun Seoul Parnas in Seoul's Gangnam district. "To genuinely raise a company's productivity with AI, you have to deploy AI that reflects the variables and context of the site and secure your own sovereignty."
Yoon said a recent survey by a U.S. economic research institute found that 90% of companies reported no improvement in their income statements despite using AI for three years. He attributed the gap to what he called a long-tail problem in industrial settings: more than 80% of issues are simple enough to handle easily, while the remaining 20% — less frequent but thorny and complex — remain beyond AI's reach.
"In the end, frontier AI models that do not know the data or the way work is done on the ground cannot break through the 80% barrier, and corporate productivity does not improve either," Yoon said. "AI systems that analyze an individual company's processes and data and actually run at the site will decide who succeeds and who fails."
Corporate AI sovereignty is essential to spreading such site-specific AI, he said. The term refers to a company's capacity to fully own its data, knowledge and know-how, and to internalize both the AI developed from them and the resulting output. "Independence and control are also core elements of AI sovereignty," Yoon said. "Each company must also have the ability to choose and change the technologies it uses at any time, so that it is not locked into a particular foundation model or infrastructure."
MakinaRocks is upgrading its own AI operating system, Runway, along those lines. The approach sends forward-deployed engineers (FDEs) responsible for on-site implementation into each location, applies each company's proprietary data and work processes to the AI, and then accumulates the results as internal corporate assets. The company plans to identify AI transformation projects across manufacturing, defense and other settings and to achieve a fully autonomous factory within three years.
"Our FDEs have visited more than 80 sites in 32 cities across five countries to date," Yoon said. "They put in roughly 70,000 hours in the field this year, and we have deployed more than 6,000 AI models at actual sites so far." He added that the company plans to build a fully autonomous factory and achieve hyper-productivity by 2028 in partnership with Enertork, an electric actuator maker.






