Gasgoo Munich-Genesis AI announced on Nov. 17 that it has closed a $500 million (approximately 3.3 billion yuan) Series A round, pushing its post-money valuation to $3.5 billion (approximately 23.3 billion yuan).
The round was co-led by Premji Invest, Khosla Ventures, and Eclipse, with participation from Durable Capital Partners, France’s Bpifrance, HSG, NVIDIA’s NVentures, AMD Ventures, Zoom Ventures, Samsung Next, and Ericsson Ventures.
Adding a previously closed $105 million seed round, Genesis AI has now disclosed total funding of $605 million.
Founded in December 2024, Genesis AI focuses on full-stack proprietary development of general-purpose robotics and dexterous manipulation technology. It has already launched products including a dexterous hand and data collection glove, the general-purpose robot Eno, the robot brain model GENE, and the simulation platform Genesis World.
Proceeds from this round will primarily target four areas: expanding data collection and large-scale model training to broaden the scope and complexity of tasks robots can handle; accelerating the realization of human-level physical manipulation capabilities; recruiting R&D and engineering teams; and driving the large-scale deployment of its general-purpose robot, Eno, in customer environments.
Turning the Human Hand Directly into a Data Terminal
Large language models can learn from the vast expanse of internet text, but robots lack similarly ready-made training materials.
A video can show a model how a hand picks up an egg, but it struggles to fully capture the force applied, when fingers slip, or how the contact surface changes.
Teleoperation can record more complete actions and states, yet the collection process relies on specialized equipment, space, and operators, making scale difficult to achieve. Furthermore, different robotic hands vary in size, degrees of freedom, and sensor configurations, meaning data accumulated on one hardware set doesn't necessarily translate directly to another.
Therefore, the data challenge in embodied intelligence isn't simply a matter of insufficient volume. What to collect, how to collect it, and whether it can be transferred to robots are the factors that ultimately determine if data is viable for training.
To address this, Genesis AI chose to start with the hand.
In May, Genesis AI released its first robotic foundation model, GENE-26.5, along with an accompanying anthropomorphic dexterous hand and data collection glove. In official demos, the robot completed tasks such as cracking eggs, chopping vegetables, solving a Rubik's cube, playing the piano, and performing lab pipetting. While this demonstrates the model's task coverage, success rates and consistency in real customer environments still require verification.
The company's data glove features tactile-sensing electronic skin and is designed to coordinate with a robotic hand modeled on human dimensions. This approach aims to establish a mapping between the human hand, the glove, and the robot. Unlike specialized teleoperation setups, this method embeds data collection into real labor: workers simply wear the gloves and supporting collection gear during their regular jobs to gather data.
Genesis AI claims its glove offers significantly better hardware cost and collection efficiency than the teleoperation solutions it benchmarked against, though public materials haven't fully disclosed the comparison targets, task samples, or statistical metrics.
Beyond glove data, Genesis AI also utilizes first-person video, internet footage, and simulation data, training and evaluating them through Genesis World.
However, a correspondence doesn't mean differences disappear entirely. Human skin, muscle, and joints remain distinct from robotic motors, reducers, and tactile sensors. As data scales, determining which experiences transfer and which actions require recalibration still demands verification on physical machines.

Image Credit: Genesis AI
Beyond the Model: Why Build the Whole Robot?
If it only provided models, data collection tools, and training platforms, Genesis AI could leave the manufacturing and deployment to robot makers. Instead, the company chose to go deeper, launching its own complete robot, Eno.
Officially, Eno abandons the common bipedal humanoid structure for a wheeled base and a collapsible, multi-segment torso. It adjusts height for work and compresses for storage, while retaining a dexterous hand close in size and form to a human's.
This design reduces the emphasis on "looking human" and focuses on operational range, mobility, and storage. After all, for factories and laboratories, whether a robot has legs is less important than whether it can use tools designed for humans and complete tasks within existing workstations—factors that directly impact deployment costs.
Eno also provides a landing point for Genesis AI's data loop: human hand data trains GENE, while Genesis World supplements simulation training and evaluation; as Eno undergoes customer testing and subsequent deployment, real-world task data can continue to feed back into model and hardware R&D.
Of course, whether this process functions effectively depends on deployment scale, data quality, and whether that data can be used for subsequent training.
For Genesis AI, getting a robot to complete tasks in a demo is just the first step. Entering customer scenarios requires solving production, durability, and continuous operation issues.
When the company launched Eno in June, it planned to initiate production and targeted customer deployments by the end of 2026. The latest public statements indicate Eno is still in small-scale testing with customers, with a broader product launch expected in 2027.
Currently, Genesis AI has partnered with LG CNS to test Eno in manufacturing and logistics scenarios.
Hardware durability is another hurdle. In a TechCrunch demo report, the Genesis AI robot was seen dropping boxes and struggling to peel off labels; the report also noted that existing dexterous hands may need replacement due to mechanical wear after operating for several weeks in standard industrial work.
Such issues illustrate that as robots enter long-term use, hardware maintenance and failure data feedback are equally important. Dexterous hands must withstand repetitive operations, and the problems encountered must be used to improve the next generation of hardware.
Moreover, full-stack proprietary development implies heavier investment. Model training requires computing power, data collection requires networks, and mass-producing complete robots requires supply chains, manufacturing, and after-sales systems.
$500 million can support continued investment in these areas, but it cannot allow the company to skip reliability verification.








