"Since the company was founded in 2023, we have accumulated 100,000 hours of data. By next year, that will surpass 1.5 million hours."
An Agibot official who met with the Seoul Economic Daily at the company's Shanghai headquarters on the 18th of last month said, "We will significantly expand our data collection scale to more firmly support the mass production system for humanoid robots." He explained, "For a robot to perform tasks delicately and accurately, it needs to accumulate at least 1,000 data points for each finely segmented motion," adding, "Real-world data is the key factor that determines success or failure in the robot market."
Agibot is the company ramping up mass production speed the fastest in the global humanoid robot industry. Last year, it became the first in the world to surpass 5,000 humanoid robot shipments. This year, the figure is projected to exceed 10,000 units.
In particular, the company plans to increase the volume of high-quality data by more than 10-fold year-on-year. While the target may appear ambitious, Agibot internally sees it as entirely achievable. Each robot collects roughly 8 to 10 hours of raw data per day, of which 2.5 to 3 hours of high-quality data remain after refinement. Based on this, a single robot can secure approximately 800 hours of high-quality data per year. If more than 1,000 robots are deployed daily, annual data accumulation on the order of one million hours becomes possible.
To realize this goal, Agibot broke ground on a second data center last year following its existing three-story warehouse-style data collection center and completed it in January this year. The entire 12-story building is being used as a humanoid robot training facility, and construction of a third center has already begun. The existing training center houses around 871 humanoid robots, but as center expansions ramp up, the number of robots deployed for data collection is expected to grow to the thousands.
The reason Agibot is expanding its data infrastructure so aggressively is that "how much real-world behavioral data can be secured and how quickly" is ultimately the key variable determining market dominance. With demand for humanoid robots rapidly broadening from automobile manufacturing to logistics and household assistance, the company has concluded that behavioral data accumulated across diverse work environments will inevitably dictate performance and commercialization. In recent public appearances, Agibot has also emphasized that the humanoid industry is moving beyond the technology demonstration stage and entering the phase of large-scale field deployment.
The data race is not simply a matter of volume. Quality matters just as much. Agibot's data collection centers are divided into dozens of compartments replicating various living and industrial scenarios including bedrooms, cafes, convenience stores, and kitchens. Each humanoid model performs different tasks depending on its purpose, and the movements and outcomes generated in the process are precisely captured through sensors and cameras.
The data accumulated this way is not only used for new product development but also holds significant potential as a new revenue model in its own right. In 2024, Agibot released "AgibotWorld," an AI training dataset, and began sharing more than 50,000 hours of data as open source. Going forward, there are also projections that the so-called "industrialization of data" — mass-producing real-world data like manufactured goods and selling it for other companies' model training — will begin in earnest. The official said, "Sharing data with partners encourages clients to more actively utilize Agibot humanoids, and the newly accumulated data from that process can in turn improve robot performance, creating a virtuous cycle."
What Is Core Power?
Core Power refers to the independent and irreplaceable capabilities a nation has secured across domains such as technology, security, diplomacy, administration, and culture.