Warum ist der Marktwert von Figure in drei Jahren auf 39 Milliarden US - Dollar gestiegen, obwohl es eine neue Finanzierung in Höhe von 1 Milliarde US - Dollar erhalten hat?

A super unicorn has emerged in the overseas embodied intelligence track.

Figure has secured a $1 billion Series C financing round, post - investment valuation reaching $39 billion.

This financing round was led by Parkway Venture Capital. Notable participating institutions include Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Align Ventures, Tamarack Global, LG Technology Ventures, Salesforce, T - Mobile Ventures, and Qualcomm Ventures.

In February 2024, it received a $675 million Series B financing round, with participants including Microsoft, OpenAI Startup Fund, Jeff Bezos, etc. The main participants in the Series C financing, Parkway Venture Capital, Intel Capital, and Align Ventures, also participated in the Series B investment.

Although embodied intelligence is one of the hottest tracks both overseas and in the Chinese market, Figure, which has received two consecutive rounds of huge financing, is still a phenomenon - level case, showing a momentum similar to that of OpenAI in the large - model field. So why do so many top - tier investment institutions favor it, and what are its competitive advantages and moats?

Vertical integration of software and hardware: The Tesla model in the embodied intelligence industry

In 2022, Brett Adcock founded Figure and elaborated on his ultimate goal for founding Figure: To solve the labor shortage and improve human well - being.

He once wrote: "The compensation for physical labor is the main driver of the prices of goods and services, accounting for about 50% of the global GDP (approximately $42 trillion per year). However, when robots 'join the labor force', labor costs will decline. Over time, when robots can manufacture other robots, humans may even completely withdraw from the production process - which will further drive down prices. Robots can eliminate the need for unsafe and unpopular jobs - ultimately allowing us to lead happier and more meaningful lives."

For this purpose, he also announced his own Master Plan following Tesla's example:

Step 1: "Build a fully functional mechatronic humanoid robot."

Step 2: "Achieve anthropomorphic manipulation ability."

Step 3: "Integrate humanoid robots into the labor force."

Brett Adcock is a successful serial entrepreneur. His founded company Archer went public successfully with a valuation of $2.7 billion, and Vettery was acquired for $100 million. The team he assembled consists of top AI and robotics experts from Boston Dynamics, Tesla, Google DeepMind, and Archer Aviation.

Brett Adcock, founder of Figure (Source: Figure)

In the early stage, the founder and the team are important criteria for investors. Brett Adcock's background as a serial entrepreneur, the powerful team he assembled, and his Master Plan are all factors that investors value.

The flywheel effect and closed - loop formed by vertical integration bring high barriers

However, Figure started to receive huge investments and cooperation from technology giants during the Series B and C rounds. At this stage, corporate investors focus on the competitive advantages of technology and products, not just the team. What are Figure's competitive advantages? It is its vertical integration strategy of combining software and hardware and the implementation of this strategy.

Compared with software, the vertical integration strategy is more likely to gain an advantage in hardware, as seen from Apple to Tesla.

From self - developed chips, to a dedicated closed - source operating system, and then to the App Store software ecosystem, Apple controls every key link from the underlying hardware to the top - level software. This creates a seamless and smooth experience integrating software and hardware and brings strong brand premium ability to Apple, making it extremely difficult for competitors to replicate.

Tesla has a fully self - developed autopilot system integrating software and hardware, self - manufactured electric vehicles, and Robotaxi services. The combination of its autopilot system and electric vehicles forms a self - evolving data flywheel, and vehicle sales and Robotaxi services provide strong cash flow for the evolution of the entire system. Equally important is that once this vertically integrated model scales up, it not only offers a better experience than competitors but also significantly reduces costs.

In the embodied intelligence industry, most companies focus on a certain part of the entire industrial chain. Whether it is the "brain" (model), perception, mobility, or dexterous hands, there are good companies or large - company laboratories working on them.

For example, companies or laboratories focusing on the "brain" include Google DeepMind, which has developed a series of VLA models such as RT - X and RT - 2. The core members of this series of models later founded the company Physical Intelligence and created the general robot brain Π0. Of course, there are also other powerful companies and laboratories in this track, such as Covariant AI (founded by AI expert Pieter Abbeel) and NVIDIA.

Companies focusing on perception include Zivid (industrial - grade 3D color cameras) and SynTouch (tactile sensors); those focusing on mobility include Agility Robotics, and those focusing on dexterous hands include RightHand Robotics.

Figure's robot body and model are both self - developed. Furthermore, it has its own computing power facilities and large - scale manufacturing capabilities, and now its robots are building other robots in its own factory.

This forms a complete flywheel effect and closed - loop: the self - developed robot body and model cooperate seamlessly to achieve optimal performance. The self - owned computing power facilities and large - scale manufacturing capabilities save costs and give control in the long run. And having robots build other robots not only verifies the robots' capabilities in scenarios but also provides data for further improving the model's capabilities.

This is Tesla's approach and also a necessary path for Figure to build its core barriers.

A body born for AI, a thinking brain, and a self - replicating factory

Figure's robot body: the current model is Figure 02; the design of the next - generation robot body, Figure 03, has also been completed. Figure 03 is a fully mechatronic humanoid robot, and its hands are also self - developed.

In addition to the improvement in functions, the biggest change lies in its manufacturing process. Figure 02 uses the high - complexity, high - tolerance, and low - efficiency computer numerical control (CNC) machining process, which allows only small - scale production and incurs high costs. Figure 03, on the other hand, has switched to mold - based processes such as injection molding, die - casting, metal injection molding, and stamping. Parts that previously took more than a week of CNC machining can now be manufactured in 20 seconds using complex steel molds. This makes large - scale manufacturing of robots possible.

Moreover, Figure has developed the F.03 battery system. The new battery has a capacity of 2.3 kilowatt - hours (kWh) and can support the robot to operate at peak performance for up to 5 hours. It is also the first humanoid robot battery undergoing dual battery safety standard certifications of UN38.3 and UL2271. And the cost of this battery has been reduced by 78% compared to the previous generation.

Figure robot's intelligence is provided by an intelligent system called Helix. This system is essentially a general vision - language - action (VLA) model, which has many features in terms of architecture and function.

Helix uses a "System 1 (fast thinking) + System 2 (slow thinking)" architecture; System 2 is a 7 - billion - parameter VLM model, mainly responsible for scene understanding and language comprehension. System 1 is an 80 - million - parameter Transformer model, responsible for executing fast reactive vision - motion strategies, converting the semantic representations of System 2 into precise and continuous robot actions.

Helix uses a fully end - to - end training method, learning all behaviors with only one set of neural network weights. Driven by Helix, Figure's robots can perform a series of tasks such as logistics sorting, folding clothes, loading dishwashers, and autonomous navigation, and it does not require fine - tuning for specific tasks.

In addition to the above features, Helix can achieve zero - sample multi - robot coordination, which makes it possible to deploy robots on a large scale and have them work collaboratively.

Compared with other overseas embodied intelligence companies, one of Figure's biggest features may be its self - owned large - scale manufacturing capabilities. This capability is realized by the BotQ factory. It is reported that the first - generation production line of BotQ will be capable of manufacturing up to 12,000 humanoid robots per year.

A key innovation in the BotQ factory is integrating the automation capabilities of Figure's humanoid robots into the assembly line. With the introduction of the AI system Helix, they will be able to use robots to assemble key components on the production line and act as material handlers. By using robots to manufacture more robots, this will accelerate the production process and ultimately lay the foundation for future autonomous manufacturing.

Further analysis shows that having robots manufacture robots on the production line directly allows robots to accumulate a large amount of practical data. This data can, in turn, optimize the Helix intelligent system. Figure has also partnered with Brookfield to collect data in its residential, commercial office, and logistics facilities, which is then fed back to the Helix intelligent system. Since the perspective and kinematic characteristics of humanoid robots are very similar to those of humans, it is possible to transfer knowledge directly from daily human activity videos. Based on these two data collection sources, Figure has the potential to build an ultra - large - scale robot dataset.

Meanwhile, compared with the traditional method of training models with human - annotated data, this method is programming - free and unsupervised, which can be said to unlock a new Scaling Law in the embodied intelligence industry.

The self - owned body, self - developed model, self - built factory with large - scale manufacturing capabilities, and the environment for scenario verification and data collection. Vertically integrating these elements is Figure's moat and differential competitive advantage, and also the reason for its $39 billion valuation.

AI hardware reshapes the physical world

In the field of embodied intelligence, to implement the vertical integration route of combining software and hardware, it is important not only to have design capabilities but also manufacturing capabilities, scenario, and industrial - chain integration capabilities. Coincidentally, the Chinese market has strong manufacturing capabilities and an industrial chain, as well as cost advantages, which provides a good foundation for Chinese embodied intelligence entrepreneurs.

Currently, many Chinese embodied intelligence companies have adopted the vertical integration route of combining software and hardware, such as Unitree Robotics, Galaxy Universal, Zilliz Robotics, and Zhujidongli.

The vertical integration route of combining software and hardware is not an easy path. Why do they choose it? Because to take this path, they need to overcome a series of difficulties in software, hardware, the industrial chain, scenarios, data, etc. Once these difficulties are overcome, they can build a high moat in the AI era.

This principle applies not only to embodied intelligence but also to Physical AI, which reshapes the physical world, and to AI - native hardware.

As Liu Gang, a partner at Alpha Startups, said: "From hardware to embodied intelligence. Traditional hardware is a passive tool, while AI - native hardware is an active intelligent agent. Hardware is just a starting point, the 'tentacles' for AI to perceive and influence the physical world. In the future, it will continuously evolve along the path of'single - point hardware → intelligent agent → embodied intelligence platform'. What we are seeing is not just product upgrading but the birth of a whole new computing paradigm - from Marc Andreessen's 'Software eating the world' to what we will see in the future, 'AI hardware reshaping the physical world'."

This article is from the WeChat official account “Alpha Startups” (ID: alphastartups). Author: Discovering extraordinary entrepreneurs. It is published by 36Kr with authorization.