NVIDIA Unveils 'New Brain' for Robots with 750% AI Compute Surge; Boston Dynamics' Spot Performs Back-to-Back Backflips; UBTECH Releases Mid-Year Results | Wave Weekly Report

1. NVIDIA Unveils 'New Brain' for Robots, AI Compute Power Surges by 750%

On August 25, NVIDIA launched the Jetson AGX Thor Developer Kit and production modules, a computing platform designed specifically for robotics applications. The NVIDIA Jetson AGX Thor Developer Kit is priced starting at $3,499, while the Jetson T5000 production-grade module can be ordered through global distribution partners.

As the 'new brain' built for physical AI and robotics, the AI compute capability of the Jetson Thor is 7.5 times greater than that of the previous generation NVIDIA Jetson Orin, with energy efficiency improved by 3.5 times. NVIDIA CEO Jensen Huang described it as the supercomputer driving the true arrival of the era of physical AI and general-purpose robots.

Based on the latest Blackwell GPU architecture, the Jetson Thor features 128 GB of memory and delivers 2070 FP4 teraflops of AI compute power. It can easily run the latest AI models within a power envelope of just 130 watts.

Furthermore, it is compatible with NVIDIA’s software stack from cloud to edge, including NVIDIA Isaac for robotics simulation and development, the humanoid robot foundation model Isaac GR00T, NVIDIA Metropolis for visual AI, and NVIDIA Holoscan for real-time sensor processing. It can also run various generative AI models, including VLA, LLM, and VLM.

Additionally, the Jetson Thor enables real-time inference, addressing a critical challenge in current technology: running multiple AI workflows to allow robots to interact intelligently and in real time with humans and the physical world.

2. Boston Dynamics Quadruped Robot Unlocks Continuous Backflip Skill

August 27, Boston Dynamics' quadruped robot Spot showcased a new skill: consecutive backflips. Additionally, Spot demonstrated large-amplitude side flips, landing steadily in each instance.

During training, reinforcement learning was the core method. Through continuous trial-and-error and feedback adjustments, Spot gradually achieved more precise body control. Furthermore, Boston Dynamics stated that after mastering the high-difficulty consecutive backflip, Spot's walking posture became even more natural, closely resembling that of real quadrupedal animals.

Robot engineer Kumar explained that the team initially trained the backflip motion in computer simulations, but applying these results to real robots almost always resulted in failure. To reduce the risk of damage, tests were initially conducted on gymnastics mats, with difficulty gradually increased until Spot could perform the actions in higher-risk environments.

3. UBTECH Announces Mid-Year Financial Report for 2025

On August 29, UBTECH announced its mid-year financial report for 2025.

In terms of financial performance, UBTECH achieved revenue of 621 million yuan in the first half of this year, a year-on-year increase of 27.5%; net profit loss was 440 million yuan, compared to a loss of 540 million yuan in the same period of 2024; gross profit reached 217 million yuan, a year-on-year increase of 17.3%; gross margin was 35%, down by 3 percentage points from the same period in 2024. R&D expenses amounted to 218 million yuan, accounting for 35.1% of revenue. As of June 30, 2025, the balance of cash and cash equivalents was approximately 1.157 billion yuan, and the number of authorized patents reached 2,790.

Regarding specific business segments, both the company's AI education business and consumer-grade robotics and other hardware equipment businesses achieved nearly 50% year-on-year growth during the reporting period. Specifically, revenue from educational intelligent robots and intelligent robot solutions grew by 48.8% to 239.8 million yuan; revenue from consumer-grade robots and other hardware devices grew by 48.9% to 260.1 million yuan.

Revenue from logistics intelligent robots and intelligent robot solutions decreased by 5.7% to 56.2 million yuan; revenue from other industry-customized intelligent robots and intelligent robot solutions fell by 29.8% to 63.8 million yuan.

In terms of products, UBTECH developed the third-generation industrial embodied AI humanoid robot Walker S2. Its hot-swappable autonomous battery-swap system enables the robot to autonomously swap batteries in 3 minutes and operate continuously for 7×24 hours.

At the commercialization level, in April, UBTECH signed a procurement contract for large bipedal humanoid robots with customers; in July, UBTECH won a bid for MiYi Auto's 90.5115 million yuan robot equipment procurement project.

Additionally, on August 31, UBTECH announced the formal signing of a $1 billion strategic partnership agreement with the internationally renowned investment firm Infini Capital. The cooperation between the two parties will revolve around financial support and industrial synergy. According to the agreement, the high-tech fund under Infini Capital will provide UBTECH with a total strategic financing credit line of $1 billion. Financing methods include participating in placements, subscribing to convertible bonds, and granting cash drawdown rights. In particular, UBTECH has been granted a rapid financing channel that allows it to draw up to $1 billion in cash when needed.

4. Ruoyu Technology Announces Completion of New Round of Hundred-Million-Yuan Angel+ Financing

On August 25, Ruoyu Technology announced the completion of a new round of hundred-million-yuan Angel+ financing. This round was jointly led by existing shareholder Dongfang Precision Machinery and followed by Shenzhen Hanqingda Investment Development Co., Ltd., an industrial investor. The proceeds from this financing round will be mainly used for the industrialization and application of embodied AI robot brains in vertical scenarios.

Previously, in March 2024, Ruoyu Technology completed an angel-round financing of over 50 million yuan, led by Dongfang Precision Machinery, followed by Kunzhong Capital, with Yuanhe Capital serving as the exclusive financial advisor.

Ruoyu Technology was founded in 2023 and incubated by Harbin Institute of Technology (Shenzhen). In terms of products, Ruoyu Technology has already launched the multimodal large model "Ruoyu Jiutian," the humanoid robot "Ruoyu Lanyue," and the data acquisition kit "Ruoyu Tiangong."

"Ruoyu Jiutian" collaborates with robot body manufacturers and scenario companies across the industry chain through an API + SDK approach, utilizing a cloud-edge architecture. Its application areas include food processing, sorting, assembly, the 3C industry, and traditional manufacturing.