$100 Million Seed Round Sets New Record for Silicon Valley Embodied Intelligence Funding; Chinese Co-Founders Lead the Charge}

This company aims to free humans from physical labor.

Recently, the rarely seen Ilya Sutskever made a rare appearance at the University of Toronto, delivering a speech where he mentioned that someday AI will complete all human tasks. This presents the greatest challenge in human history but also huge opportunities.

In recent years, AI progress has been remarkable. However, many joke that "we wanted AI to do chores, cook, and take care of children, while we write poetry, sing, and cultivate ourselves; now it’s the other way around." This highlights a paradox: AI’s 'brain' and 'body' are progressing at different speeds, with only some mental labor unlocked and physical tasks still limited.

The emerging field of embodied intelligence aims to solve this problem. Yet, a true 'OpenAI' for this domain or a universal architecture like Transformer has yet to appear.

Recently, a new Silicon Valley startup called Genesis AI attracted attention by raising $105 million in seed funding. According to TechCrunch, the round was led by top US venture capital firms Khosla Ventures and Eclipse. Khosla was an early investor in OpenAI, and Eclipse has a background in Tesla robotics.

"Genesis" sounds familiar? Yes, it’s related to the well-known embodied intelligence project and GitHub star Genesis. It’s a generative physics engine developed over two years by CMU and over 20 labs, capable of simulating 4D dynamic worlds and physical phenomena, designed for general robotics, embodied AI, and physics AI applications.

This project caused a sensation and was seen as a glimpse into the future of embodied intelligence (see 2-year effort by Chinese team, open-source physics engine Genesis, simulating everything in the world). It has become one of the most prominent embodied intelligence projects in recent years, and the Genesis community is now the largest open-source community in robotics simulation.

^{Physical world generated by Genesis. Tip: Mini Wukong with a stick runs on the desk for 3 seconds, then jumps into the air, landing with the right arm down. The camera starts close-up on his face, then follows and zooms out. When Wukong jumps, at the highest point, the action pauses for a few seconds. The camera rotates 360° around the character, then slowly rises and continues the motion.}

The core team’s entrepreneurial effort has attracted widespread attention. With such substantial funding, what are their plans? Here’s what we know about the company.

Genesis AI: Breaking Silicon Valley embodied intelligence funding records

The founding team of Genesis AI includes top talents from Mistral AI, NVIDIA, Google, Apple, CMU, MIT, Stanford, Columbia, and Maryland, with deep expertise in physics simulation, graphics, robotics, and large-scale AI training and deployment.

Unlike typical startups led by senior professors, this is a team of young PhDs from top research institutions, highly active in AI and robotics research.

Founder and CEO Zhou Xian. Zhou Xian graduated from Carnegie Mellon University last year with a PhD in robotics. His undergraduate was at Nanyang Technological University, where he led a robot project that was featured in Science Robotics. His research covers world models, imitation learning, and reinforcement learning. He proposed a new paradigm for generative simulation and initiated the open-source Genesis project.

Co-founder Théophile Gervet, a CMU graduate, early member of Mistral’s founding team, and head of multimodal models. He led the development of Mistral’s first multimodal large model Pixtral 12B, achieving 7x performance with only 1/7th the size of Llama-3.2 90B. He also led the development of Mixtral 8x7B, which outperforms GPT-3.5 and Llama 2 70B models, significantly reducing inference speed and costs. Before Mistral, Theo was a founding member of Skild AI.

Among the early team members listed on the company’s website are familiar Chinese faces:

Xu Zhenjia, previously focused on the Stanford UMI project and the popular Diffusion Policy robot network architecture. He graduated in 2024 from Columbia/Stanford with a PhD, advised by Shuran Song.

He previously led robot data collection and real-world system work for NVIDIA’s GEAR team, winning best paper awards at RSS and CoRL conferences.

Wang Zunxuan, recently graduated from MIT CSAIL with a PhD, advised by robotics expert Daniela Rus. His research spans data, brain, and ontology, key areas in embodied intelligence. Previously, he worked at Liquid AI on foundational models.

Yunlong Qiao, a drone reinforcement learning racing pioneer and main developer of the high-performance simulation tool Flightmare. He earned his PhD at the University of Zurich under D. Scaramuzza, specializing in robotics.

He once beat human world champions using reinforcement learning algorithms, with results published on Science Robotics.

^{Physical world generated by Genesis. Tip: Mini Wukong with a stick runs on the desk for 3 seconds, then jumps into the air, landing with the right arm down. The camera starts close-up on his face, then follows and zooms out. When Wukong jumps, at the highest point, the action pauses for a few seconds. The camera rotates 360° around the character, then slowly rises and continues the motion.}

The young team behind this project has attracted widespread attention. With such substantial funding, what are their plans? Here’s what we know.

Genesis AI: Breaking Silicon Valley embodied intelligence funding records

The founding team includes top talents from Mistral AI, NVIDIA, Google, Apple, CMU, MIT, Stanford, Columbia, and Maryland, with extensive experience in physics simulation, graphics, robotics, and large-scale AI training and deployment.

Unlike typical startups led by senior professors, this team consists of young PhDs from top research institutions, highly active in AI and robotics research.

Founder and CEO Zhou Xian. Zhou Xian graduated from Carnegie Mellon University last year with a PhD in robotics. His undergraduate was at Nanyang Technological University, where he led a robot project featured in Science Robotics. His research covers world models, imitation learning, and reinforcement learning. He proposed a new paradigm for generative simulation and initiated the open-source Genesis project.

Co-founder Théophile Gervet, a CMU graduate, early member of Mistral’s founding team, and head of multimodal models. He led the development of Mistral’s first multimodal large model Pixtral 12B, achieving 7x performance with only 1/7th the size of Llama-3.2 90B. He also led the development of Mixtral 8x7B, outperforming GPT-3.5 and Llama 2 70B models, significantly reducing inference speed and costs. Before Mistral, Theo was a founding member of Skild AI.

Among the early team members listed on the company’s website are familiar Chinese faces:

Xu Zhenjia, previously focused on the Stanford UMI project and the popular Diffusion Policy robot network architecture. He graduated in 2024 from Columbia/Stanford with a PhD, advised by Shuran Song.

He previously led robot data collection and real-world system work for NVIDIA’s GEAR team, winning best paper awards at RSS and CoRL conferences.

Wang Zunxuan, recently graduated from MIT CSAIL with a PhD, advised by robotics expert Daniela Rus. His research spans data, brain, and ontology, key areas in embodied intelligence. Previously, he worked at Liquid AI on foundational models.

Yunlong Qiao, a drone reinforcement learning racing pioneer and main developer of the high-performance simulation tool Flightmare. He earned his PhD at the University of Zurich under D. Scaramuzza, specializing in robotics.

He once beat human world champions using reinforcement learning algorithms, with results published on Science Robotics.