HomeArticle

Nvidia and AMD have teamed up to make a joint bet for the first time, Genesis has completed its Series A financing with a valuation exceeding 23 billion.

量子位2026-11-18 11:28
Embodied Full-Stack Player

Just now, US embodied intelligence company Genesis AI has completed a new round of financing —

A $500 million Series A round, bringing its post-money valuation to $3.5 billion.

The investor lineup is extremely star-studded:

NVIDIA, AMD, former Google CEO Eric Schmidt, Zoom founder Eric Yuan, Yahoo founder Jerry Yang... A host of industry giants are on the list.

It is worth noting that this marks the first joint investment by AMD and NVIDIA in an embodied intelligence company.

Against the backdrop of a generally cold financing market for embodied intelligence, this round of financing is undoubtedly particularly prominent.

What exactly makes Genesis different?

After in-depth research on its official website and technical documents, we found that the answer may lie in Genesis's own positioning —

A full-stack embodied player.

And all of this dates back to half a year ago.

One Sword Forged in a Year

Genesis AI was founded in March 2025, with headquarters spanning the United States and France, co-founded by Chinese scholar Zhou Xian and Théophile Gervet.

Before officially founding the company, its team had already gained extremely high popularity in the academic and engineering circles with their open-source physical simulation project.

This May, Genesis AI unveiled its first embodied foundation model: GENE-26.5.

Upon release, a series of demo videos showcasing its capabilities drew widespread attention.

The main battlefield is the kitchen.

In an unaccelerated four-minute-long video, the robot smoothly completed more than 20 subtasks:

Cracking eggs with one hand, cutting tomatoes with a knife, using an egg beater, a salt grinder and a spatula.

The next scenario is the laboratory.

Requiring the same millimeter-level precision and the same long-range one-shot task, GENE-26.5 can independently complete pipetting, sealing test tubes, and finally placing the test tubes into a centrifuge.

In addition, it can also make milkshakes, restore Rubik's Cubes, and play the piano, fully demonstrating the general operation capability of its dexterous hand.

The most critical point is that these capabilities are not pieced together by multiple dedicated small models, and the vast majority of operations rely on the same set of GENE models with shared weights.

However, in today's robotics field, demo videos are nothing new, and the methodology behind them may be more worthy of exploration.

One special point of Genesis is that the official stated that the model is not all they want to showcase.

In their view, robots are highly coupled engineering systems. Data, evaluation, ontology... Shortcomings in any link will, like dominoes, ultimately limit the upper limit of the overall performance.

What to do?

Genesis's answer is:

I want all of them.

Disassembling the Full-Stack "Trinity"

In Genesis's system, data, model, evaluation and ontology are designed as a tightly coupled closed-loop system.

1. Data Engine: Solving the "Chicken-or-Egg" Problem

One of the biggest bottlenecks in robot learning is the scarcity of high-quality data, especially hand operation data.

Human hands can generate the richest and most natural interaction data, but the huge morphological difference between human hands and robotic hands makes it difficult to transfer the data directly.

But what if this ontological gap does not exist?

Genesis's solution is quite straightforward: they directly bridge this gap at the physical level —

1:1:1 full-link mapping.

Human hands: the source of all interactions.

Data glove: adopting electromagnetic field (EMF) finger tracking technology and high-density tactile sensing to capture the signals of human hands at work with high fidelity.

Genesis dexterous hand: 1:1 adapted to human hands, with 20 degrees of freedom, and its surface is covered with flexible materials simulating human skin.

The effect is that when workers wear data gloves to do daily work, their hand movements and tactile signals can be collected with high fidelity and transferred to robots for learning almost without loss.

Working is data collection.

In this way, by this May, Gene had used more than 200,000 hours of multimodal data. We learned that the current data volume has reached the scale of millions of hours.

This is exactly the charm of the full-stack approach.

Many seemingly intractable problems at the model level may find more fundamental solutions when approached from the system level.

2. Genesis World: Turning Iteration into a Computing Power Problem

Data is the raw material, and the progress speed of the model depends on the efficiency of "processing" the raw material.

If every improvement of the model has to be verified in the physical world, the R&D cycle will be greatly slowed down. Genesis's solution is: Go all in on simulation.

We regard simulation as the evaluation and iteration engine for robot foundation models, not just a data generator.

Real-world experiments limit the speed at which we can score candidate solutions; simulation removes this limit, turning the development cycle from a wall-clock problem into a compute problem.

This is Genesis World, a "spiritual time house" for robots to train and practice.

The effect is remarkable.

A single model evaluation may involve hundreds of tasks, each repeated hundreds of times. In the real world, this requires an operator and a robot to work continuously for more than 200 hours.

But in Genesis World, the same tens of thousands of tests can be completed in less than half an hour without any human or hardware intervention.

Of course, the premise of all this is "credibility".

In order to narrow the Sim-to-Real gap, Genesis has invested tremendous efforts to self-develop the rendering engine Nyx, cross-platform compiler Quadrants and a unified physics engine.

Finally, in the controlled experiments conducted by the official, the simulation evaluation and real robot test results showed a strong correlation, with the Pearson correlation coefficient reaching 0.8996.

Beyond the simulation engine, Genesis has internally built a complete pipeline that automatically generates tasks, environments, objects and reward functions, allowing robots to self-evolve in large-scale automatically generated simulation environments.

3. Eno: Born for "Getting Things Done"

But up to this step, only the preparatory work has been completed.

Data, simulation, and models all ultimately need a product form to carry them and build a bridge to the physical world.

For Genesis, this product is Eno.

Eno's minimalist and timeless design has been called "the iPhone moment of robots" by Forbes magazine.

It is worth mentioning that Eno has just won the Gold Award of the German Design Award. It is not the cool bipedal humanoid robot we imagined. On the contrary, it has a wheeled base, and its torso is made of several folded plates.

Aircraft do not need to have bird wings. Similarly, many features of human beings may be burdens for machines.

From the very beginning, Eno's design goal was not to imitate the "appearance of human beings". But this instead gives it the most "human-like capabilities", while encapsulating these capabilities in an elegant and minimalist body.

Wheels ensure movement efficiency and battery life; the liftable body allows it to adapt to workbenches of different heights.

Every design serves functionality, efficiency and cost. Behind this is also the embodiment of that first principle — full-stack embodied native.

As Genesis CEO Zhou Xian said in the financing statement:

Our mission is to eliminate the limitations imposed on human ambition by the physical world. To achieve this, we are building the world's best general-purpose robot, a robot that can work reliably day after day in the real world.

This requires intelligence, hardware, and everything needed to train and test robots to work together as an integrated system. By building all of this ourselves, we can obsess over every detail and improve the entire robot faster.

Deployment First

With $500 million in hand, what is the next step?

Mass production.

Genesis said that this round of financing will be mainly used to promote the large-scale mass production and deployment of Eno in the real world.

Specifically, they plan to start production and customer deployment by the end of 2026, first entering industrial environments such as manufacturing, logistics and laboratories, then expanding to service industries such as hotels and hospitals, and finally entering consumer household and outdoor applications.

From structured to unstructured, taking one step at a time.

But to be honest, even taking this first step is extremely difficult.

Demo videos can have unlimited room for imagination, but when it comes to deployment, "reliability" is the only measure.

The link of embodied intelligence is extremely long, and any error in any link may lead to a total collapse in real scenarios.

At the same time, the long-tail problems in the real world are numerous and complicated. Only by going through hardships in the scenarios can the delivery standards be gradually met.

Technology and products must be advanced together. It is not enough to just sit in the laboratory to train models. Getting hands dirty is a must-do, and only at the front line can the closed loop of model-deployment-data be realized.

From this perspective, mass production deployment itself may also be an indispensable part of the embodied intelligence R&D process.

In the past few years, the debate on the route of general-purpose robots has often revolved around "humanoid" and "non-humanoid".

Now, Genesis AI provides a new frame of reference: the real dividing line may not lie in the shape, but in the R&D paradigm.

AGI in the physical world is a much more arduous and longer-lasting tough battle than large language models.

When a system is complex enough, local optimization will eventually hit a bottleneck. Only by taking all variables into our own hands and thinking, designing and iterating as a whole can we achieve a breakthrough.

Computer pioneer Alan Kay left a maxim decades ago:

People who are really serious about software should make their own hardware.

This is true for Tesla's car manufacturing, and the same for Apple's consumer electronics. Now, more and more embodied intelligence companies like Genesis are proving that:

Building robots must also be like this.

"Full-stack native."

Against the background that OpenAI and Anthropic are also starting to bet on robot brains today, this may be the only path to physical AGI.

This article is from the WeChat official account QbitAI (ID: QbitAI), author: Jay, published with authorization from 36Kr.