Faced with increasingly fierce competition in the field of AI (artificial intelligence) chips, AI chip leader NVIDIA sees humanoid robots as the next huge market with great potential.
On December 29th local time, according to foreign media reports, Nvidia will officially launch the next generation of small computer dedicated to humanoid robots, Jetson Thor, in the first half of 2025. Deepu Talla, Vice President of Robotics Technology at NVIDIA, stated that the "ChatGPT moment" for physical AI and robotics technology is approaching, and the market has reached a "turning point".
At the NVIDIA GPU Technology Conference (GTC) in March this year, NVIDIA CEO Huang Renxun announced that the company will launch the Jetson Thor chip platform. As a system chip specifically designed for robots, Jetson Thor is built on the Nvidia Blackwell architecture and can provide 800 trillion 8-bit floating-point operations per second in terms of AI performance. It can run multimodal AI models that support humanoid robots, including Nvidia's Project GROOT Foundation model launched at the same time. Prior to the release of Jetson Thor, Nvidia had already released multiple chip solutions in the robotics field, including Jetson Orin, Jetson Orin Nano, and Jetson AGX Xavier.
Talla believes that two technological breakthroughs will drive the transformation of the robotics market: the explosive growth of generative AI models and the ability to train robots using simulated environments on these models. Talla said that the latter helps to address the "gap between robot simulation performance and real-world performance," which has gradually narrowed over the past 12 months: "We can now conduct simulation experiments and combine them with generative AI models, which we couldn't do two years ago. NVIDIA provides a platform for relevant companies to accomplish these tasks
Since the beginning of this year, Nvidia has extensively expanded its presence in the field of robotics. Previously, Talla stated in an interview that Nvidia does not intend to directly compete with robot manufacturers such as Tesla, but rather to provide them with a 'bottom-up outsourcing business'.
In February of this year, Nvidia, along with tech giants such as Microsoft and OpenAI, participated in the B-round financing of humanoid robot startup Figure AI. Figure AI raised approximately $675 million in this round of financing, with the company's valuation reaching $2.6 billion. In November of this year, robot company Apptronik, which uses Nvidia technology in its research and development process, announced a partnership with Google Deepmind. The two companies will jointly develop a new generation of humanoid robots to tackle complex operational challenges in dynamic real-world environments. Amazon, Toyota, and Boston Dynamics have also adopted technology provided by NVIDIA in the training of their robots.
As of now, Nvidia has not separately disclosed sales data for its robot products, and GPUs remain the main driving force for its revenue growth. After the market closed on November 20th local time, Nvidia released its third quarter financial report for the fiscal year 2025 ending on October 27th, achieving revenue of $35.082 billion during the period, a year-on-year increase of 94%; The data center revenue, including AI chips, was $30.8 billion, a year-on-year increase of 112%, accounting for about 88% of Nvidia's total sales in the third quarter.
According to data from BCC, a market research firm in the United States, the global robotics market is currently valued at approximately $78 billion and is expected to reach $165 billion by 2030. Goldman Sachs has also significantly raised its expectations for the size of the humanoid robot market in its latest report, predicting that by 2035, the market size of humanoid robots will reach $38 billion, more than six times the previous forecast of $6 billion.
However, David Rosen, head of the Robust Autonomy Lab at Northeastern University in the United States, said that the robotics market still faces significant challenges in training models and verifying their actual safety: "Currently, especially in the field of robotics, we have not developed very effective tools to verify the safety and reliability of machine learning systems, which is an important scientific challenge