NVIDIA recently launched Jetson Thor, which is like the physical intelligence core of robots. It has a Blackwell GPU and 128GB memory stacking 2070 FP4 TFLOPS AI computing power, which is 7.5 times that of the previous generation Jetson Orin. TrendForce said that this is not only a single digital increase, but also helps the end body to process large-scale test data and large language models (LLMs) in real time, allowing high-level humanoid robots to truly see, think and act. With the continuous adoption and construction of ecological circles among manufacturers such as Agility Robotics, Boston Dynamics, and Amazon, the scale of the human-type robot chip market is expected to exceed US$48 million in 2028.

The new software and hardware of humanoid robots have been launched, and the application scenarios that affect the key volume are still the focus of the industry. According to the latest article of the International Joint Robot Association (IFR) in the second half of the year, although human-type robots in various countries are different in technology and purpose, they mainly focus on trial-based replenishment in the short term, and enter manufacturing and service scale in the medium term. It is expected to popularize the daily family scene in the long term, and the high-level SoC utility is more important. TrendForce's "1H25 Robot In-depth Research Report" puts forward a view that global humanoid robots need to wait until they enter the family stably before and after 2032 to truly exceed 100,000 units, which is in line with the stage development mentioned by IFR.

The short- and medium-term application scenarios of humanoid robots are becoming more clear, and manufacturers are becoming more committed. Judging from the ending Taiwan robot and smart automation exhibition, R&D, Xinhan, Solomon and others all showcase products and services built with Jetson Thor as the core. In the fields of servo motors, precision speed reducers, advanced sensors, etc., Taiwan has also expanded from a single component to provide a complete solution; some manufacturers have launched functional module products, such as fast-integrated visual and sensing modules, allowing developers to assemble and adjust in the form of stacking wood, effectively reducing entry barriers, and thus as the core value of the supply chain entering the international factory.

TrendForce notes that despite the powerful performance of the NVIDIA Jetson Thor series, the price of the development kit is $3,499, a significant increase from the $1,499 of the previous generation Jetson Orin. However, the industry trend hopes to reduce the price of humanoid robots to facilitate promotion. For manufacturers who only plan humanoid robots to perform relatively simple operations in the short and medium term, relatively average chips can meet demand. Therefore, NVIDIA may once again make good use of software and hardware to launch more software platforms that can activate efficiency for Jetson Thor, achieving benefits such as shortening development time and executing specific tasks to increase the value of high AI computing power costs.