Robots that fold laundry and wash dishes have arrived. Robots that move boxes and assemble products in manufacturing processes have already reached commercialization. The era of physical artificial intelligence is rapidly approaching, centered on robotics.
Physical AI differs from online chatbots like ChatGPT and AI agents—it possesses physical form and operates in the real world. For physical AI to function properly in actual environments, understanding physics is essential. Even robots still struggle to freely adjust arm movement direction, force intensity, and pressure like humans. Sophisticated calculations and learning based on physical laws are necessary to determine how firmly to grip objects without damaging products.
SolverX is the company developing physics AI technology that learns and internalizes these physical laws through AI. In a recent interview with Seoul Economic Daily in Gangnam, Seoul, SolverX CEO Choi Yun-young said, "Technology is evolving toward handling larger amounts of information, from large language models to robotics. Among these, physics AI that helps understand reality will become the next-generation AI."
SolverX possesses technology that trains AI on complex physical laws operating in reality and stably predicts outcomes. Based on this, the company is developing AI solutions specialized for industrial engineering in automotive, aerospace, and battery sectors.
For example, when designing new vehicles, automakers run simulations in virtual spaces to check for physical problems according to new designs. The software used is computer-aided engineering (CAE) solutions, supplied by global companies including Siemens. The problem is that receiving simulation results can take up to two weeks.
"With the transition from internal combustion engines to electric vehicles, there's more to design and more to check," Choi said. "Product trends are changing rapidly with technological advancement, but long simulation times inevitably burden manufacturers."
SolverX solved this bottleneck with AI. The company creates customized solutions by combining physics-trained AI with each company's proprietary data. Running simulations on new designs with this solution produces results in seconds. Companies can receive AI-generated reports immediately.
This allows companies to move beyond testing ideas one at a time over several days with existing solutions. Using SolverX's AI solution, they can rapidly test multiple ideas without time and cost constraints.
"Companies can freely design and create better products by running simulations in real time," Choi explained. "This process can also increase company margins."
Training AI on physical laws to create industry-specific engineering solutions is no easy task. A key constraint is that manufacturing site data is often insufficient for AI learning. Technology verification (PoC) work particularly relies on past product data, but few companies systematically manage historical product data. Corporate culture also remains reluctant to share data with external companies.
"We needed to receive component information and detailed design data from clients, which was difficult," Choi said. "I went to academic conferences to meet and persuade corporate officials, made phone calls, and sent emails offering free tests to convince them."
Through this process, SolverX conducted its first PoC in 2024 with a domestic auto parts company. "This was a company that had previously wanted to adopt AI and pursued work with global companies but failed," Choi said. "Through the PoC, they were satisfied seeing simulation results different from global companies."
"LLMs are models that produce better results with more data, making them difficult to apply directly at manufacturing sites with limited data," Choi added. "By adding mathematical concepts and characteristics, we achieved accurate results even without much data, improving accuracy."
To date, SolverX has conducted PoCs with 15 companies. Auto parts companies represent the largest segment by sector. Other companies in home appliances, semiconductors, and heavy industry have also tested SolverX's solutions. Some companies signed actual contracts after seeing PoC results, recognizing the company's technology and business model. This year, SolverX plans to conduct PoCs with foreign auto parts companies.
"These days, companies are very interested in AI transformation, which makes business favorable," Choi said. "I have about six client meetings per week."
The company is also pursuing expansion of SolverX's solutions to other manufacturing fields including semiconductor manufacturing processes. Current capabilities cannot apply to all physical phenomena—development must enable understanding of difficult and complex physical phenomena. The more complex the physics, the greater the synergy when AI successfully learns them.
"Domestic robot companies are at the research stage after first proposing methods to utilize physics AI in robot learning," Choi said. "Currently our solution operates in cloud environments, but if it can run in internet-restricted situations, defense contractors could also use it."
Recently, the company added generative AI conversation capabilities to its solution. This allows employees to run simulations and receive results through conversation alone, without learning new solution features.
"Korea is a country with diverse manufacturing processes from simple components to cutting-edge rockets—only Germany, Korea, Japan, and China have such capabilities worldwide," Choi said, arguing this makes Korea an ideal environment for testing physics AI based on manufacturing processes.
SolverX is also actively knocking on overseas market doors. The company participated in CES in Las Vegas this year. Collaboration discussions continue through foreign company officials who visited the booth. This month, Choi will participate as a startup speaker at the Microsoft AI Tour in Seoul.
From the business's early days, Choi has prepared for overseas markets including the United States. SolverX has received investments centered on unrestricted capital for entering overseas markets or signing contracts with global companies. In the early business stage, the company also received investment recommendation letters from global big tech officials.
"Just as people think of 'Anthropic' as the representative of AI coding, I want to make 'SolverX' the company representing physics AI," Choi emphasized. "Our goal is to build competitiveness in the global market."