On August 25 local time, the Thor chip, referred to by NVIDIA as the “new brain” for robots, was officially launched in the form of a developer kit, priced at $3,499 (approximately 25,000 RMB). This chip is designed to support embodied intelligent robots in real-time, intelligent interaction with the physical world.

“Robots require rich sensor data and low-latency AI processing to support real-time operation, which must rely on powerful AI computing power and memory to parallelly process data from multiple sensors,” NVIDIA stated.

The issue of insufficient computing power faced by embodied intelligent robots has attracted industry attention. NVIDIA’s previous generation Orin chip is used by many robot manufacturers, but its maximum computing power is only 275 TOPS (275 trillion operations per second). In the view of some industry insiders, if robots operate autonomously in scenarios, this level of computing power is far from sufficient.

Official data shows that the new generation Thor chip, based on the Blackwell architecture GPU, provides a peak computing power of up to 2070 TFLOPS (2070 trillion floating-point operations per second) at FP4 precision, which is 7.5 times higher than the previous generation Orin chip, and energy efficiency (computing power output per unit power consumption) is improved by 3.5 times. Additionally, its CPU performance is 3.1 times higher than the previous generation, with 128GB of memory, a 2-fold increase.

NVIDIA claimed that the improvement in chip performance enables robots to process large amounts of sensor data and run robot AI models in real-time at the edge, minimizing reliance on cloud computing power. “The Thor chip provides critical real-time inference capabilities for high-performance physical AI applications such as humanoid robots, agricultural robots, and surgical assistant robots.”

NVIDIA introduced that renowned foreign robot companies such as Boston Dynamics and Figure AI, as well as domestic manufacturers like Ubtech, Galaxy General, Unitree Robotics, Zhongqing Robotics, and Zhiyuan Robotics, have率先 deployed the Thor chip.

Wang Xingxing, founder and CEO of Unitree Robotics, stated in a press release issued by NVIDIA that the Thor chip can bring stronger agility, faster decision-making, and higher levels of autonomy to robots, which is crucial for robots to achieve navigation and interaction in the real world.

NVIDIA founder Jensen Huang is optimistic about robots becoming the next wave of AI, but his company does not directly engage in the development of humanoid robots, still playing the role of a “shovel seller.” In addition to computing chip hardware specifically designed for robots, NVIDIA also provides several software platforms, such as the robot foundation model Isaac GR00T and the robot simulation development platform Isaac Sim.

Since 2014, NVIDIA has quickly bound developers in the robotics field. The company disclosed on August 25 that over 2 million developers have joined the NVIDIA robotics technology ecosystem, covering numerous industries such as manufacturing, food delivery, agriculture, healthcare, and facility maintenance.

However, the robotics business currently accounts for a very small portion of NVIDIA’s overall revenue. According to the company’s Q1 2026 financial report released in late May, the robotics and automotive businesses were combined into one category, with total revenue of $567 million, accounting for only about 1.29% of overall revenue, but growth is significant, with a year-on-year increase of 72%.