LG Electronics will begin full-scale operation of its robot data factory in Seoul in the fourth quarter of 2026 and has raised its robot training-data target for the year fivefold, to about 500,000 hours, as it deepens its physical AI collaboration with NVIDIA. The plans were set out on LG’s third-quarter earnings conference call on October 28, where executives described the Seoul facility as Korea’s largest hub for robot training and data acquisition, and said the data it collects will first be used to improve LG’s own robots before any paid work for outside customers.
A Data Factory Built For Real-World Tasks
Chief financial officer Kim Chang-tae told the call that the LG Data Factory in Umyeon-dong, in Seoul’s Seocho district, will move to full-scale operation in the fourth quarter. The factory is designed to collect and train the data robots need to carry out tasks in real-world environments,. That data will feed LG’s robot foundation model, the AI system that governs a robot’s decisions and movements.
LG is collecting data in spaces configured to resemble real work environments, drawing on experience from its homes and manufacturing sites. The company combines data gathered in the physical world with data generated in virtual environments, so that robots can practise in simulation before they act in the real world. The combination, executives said, is what makes the data useful for decision-making and motion, not just for recognising objects.
Five Times The Original Target
The 500,000-hour figure is a sharp increase on the goal LG set in August, when it aimed to secure 100,000 hours of training data by the end of the year. Lee Jae-mo, LG’s head of robotics strategy, said on the call that the company is collaborating with NVIDIA across the full spectrum of physical AI technologies, including simulation, world models and robot development platforms.
“By combining real and virtual data, we are establishing a virtuous cycle that will secure about 500,000 cumulative hours of training data within this year,” Lee said. A world model learns how the physical world behaves so a robot can predict its actions’ outcomes, and more hours of combined real and simulated experience should sharpen that judgment.
Internal Use First, Outside Business Later
LG was careful to set expectations about commercialisation. Lee said the company is reviewing several business models for the data factory, including research-oriented products built around single devices, vertical hardware and software packages developed with service-sector partners, and data sales and dataset construction. But paid pilot projects for external customers remain a medium- to long-term consideration.
“We are reviewing paid pilot commercialisation for external customers over the mid to long term, but for now we are prioritising internal use to advance our robots’ performance,” Lee said. “Regarding external commercialisation, we will communicate when the business model is further defined.”
Actuators, The Parts Robots Move With
Alongside the data factory, LG is expanding the hardware side of its robot business. An actuator, which combines a motor and a reducer to control the movement of a robot’s joints, is one of the core components that determines how smoothly and precisely a robot can move. LG built a pilot production line for actuators in Changwon in the first half of the year and is now preparing for mass production, including equipment automation and stronger quality controls.
Kim said LG expects to gain an early advantage by helping to shape component standards while the robot market is still forming, and that it is already carrying out order-taking activities aimed at global clients. The plan is to build a business that spans finished robots, core components such as actuators, data platforms and operating systems, rather than relying on any single part of the stack.
Why It Matters
This matters because robot data, not just robot hardware, is becoming the scarce resource in physical AI. A robot that has only ever been trained in simulation will hesitate, or fail, when it meets a cluttered kitchen or a busy production line. A facility that generates hundreds of thousands of hours of combined real and virtual experience is an attempt to close that gap at industrial scale, and LG’s decision to raise its target fivefold in two months suggests the company believes the approach is working.
It also matters for how the robot industry will be structured. If LG succeeds in turning its factory into a data platform, it could eventually sell data and datasets to other robot makers, much as chipmakers sell components today. If it does not, the factory still serves LG’s own products, from home robots to factory machines. Either way, the October call marks a shift from talking about physical AI to measuring it in hours of training data, a metric that rivals will now be judged against, as in South Korea’s plan to put 1,700 AI robots into public service and LG Innotek’s humanoid vision work on NVIDIA’s Holoscan platform. The same scarcity logic is driving the wider hardware market, as in SK Hynix’s $75 billion AI memory windfall.
Background
LG Electronics has been building its robot business around its strengths in home appliances and manufacturing, using its factories and homes as sources of real-world data. The full-scale operation of the Seoul data factory in the fourth quarter is the point at which that work moves from preparation to production.
Source: Chosun Biz report and Aju Press report
0 Comments