Embodied robots are about to enter a period of capital consolidation: Who will become the acquirer?

On October 28, Sequoia China and Unitree Robotics jointly launched the "ShanYu · Creation Plan" to seek early-stage projects in embodied intelligence, robotic applications, and key upstream and downstream technologies.

Unitree provides engineering experience, Sequoia provides an investment system, and the two sides jointly screen projects.

DeepSeek is also looking for an entry point into the physical world.

According to an October 12 report by Bloomberg, DeepSeek plans to lead a tens-of-millions-of-dollars seed round in robotics AI company Winverse AI, which is proposed to have a post-investment valuation of about $400 million.

Winverse AI was founded only a few months ago, and its founder Hu Han previously led research on Tencent's vision-language models and embodied foundation models.

The deal has not yet been finalized, but the proposed valuation already reveals large-model companies' willingness to price scarce teams and robot "brains."

IT Juzi data shows that in the first half of 2026, 288 financing deals occurred in China's embodied intelligence and robotics sector, with financing exceeding 46 billion yuan; according to IT Juzi statistics cited by 36Kr, about 70% of the funds flowed to the top 20 companies.

Based on public reports and court announcements, TMTPost statistics show that from February 2025 to September 2026, more than 20 robotics companies underwent bankruptcy, dissolution, liquidation, or shutdown of business lines.

In the article "Embodied Robots, Winter Is Coming," Chaoyong AI specifically described the current situation of the embodied track. At the same time, Chaoyong AI believes that embodied robots are about to enter a period of capital consolidation. Next, scarce robot "brains," as well as companies with core component technology, data, engineering capabilities, or real-world scenarios, have the opportunity to become acquisition targets.

Leading companies still have funds and also have the need to fill capability gaps.

Although some companies have not yet achieved a closed commercial loop, their teams, technology, and data still have value.

For buyers, acquiring a formed team may be faster than rebuilding from scratch.

For companies still operating but finding it difficult to independently achieve a closed commercial loop, M&A offers another path: integrating teams, technology, orders, and data into the acquirer's products, factories, and customer systems, reducing duplicated construction.

Large-model and chip companies, industrial companies, and automakers are three types of potential buyers worth watching.

DeepSeek's two moves point respectively to robot bodies and "brains."

Previously, it had contributed about 141 million yuan to participate in Unitree Robotics' IPO strategic placement; the planned investment in Winverse AI is closer to the model and infrastructure side.

Winverse AI has a short history and no public products, making it difficult to value based on existing revenue and orders. Investors are mainly attracted by the team, technical direction, and future positioning.

Chip companies are also paying for similar capabilities.

On September 28, AMD announced an acquisition agreement with World Labs, valued at about $8.2 billion, paid entirely in stock.

Image source: AMD official website

World Labs was co-founded by Fei-Fei Li and mainly researches spatial intelligence and world models, with technology involving 3D environment generation, robot learning, and simulation. The deal is expected to be completed before the end of 2026 and still requires regulatory approval.

What AMD values is not robot production capacity and orders, but the model research team and the opportunity to understand next-generation physical AI workloads.

Not all tech companies need to acquire directly.

In the equity financing agreement already signed by XPeng's robotics business, Tencent and Alibaba each subscribed about $100 million.

In the future, large-model companies such as Zhipu, Alibaba, and ByteDance, or internet giants with large-model businesses among their leading operations, may all invest in or directly buy robotics and robot "brain" companies.

Contact does not equal investment, but it shows that large-model companies are already looking for specific targets.

02

Whether they invest or acquire depends on which capability their existing businesses still lack.

In June 2025, CATL led Galbot's B round of more than 1.1 billion yuan. At that time, Galbot's heavy-duty industrial robots had already entered CATL's battery factories.

Galbot robots working in a CATL battery factory

For CATL, the order matters: first verify whether the robots can be used in the production system, then deepen cooperation through equity investment.

Hangzhou Kelin chose a more direct approach: acquiring complete-machine manufacturer Kepler. Hangzhou Kelin mainly engages in intelligent perception and diagnosis of power equipment, and has already accumulated technology in six-axis force sensors, dexterous hands, and motion control systems; Kepler makes heavy-duty full-size industrial humanoid robots.

Through the acquisition, Hangzhou Kelin extends its robotics business from core components to complete machines, and its existing power and industrial customers may also become entry points for robot products.

Hangzhou Kelin invested in Kepler twice within five months, and the valuations corresponding to the two transactions were not the same.

In December 2025, Hangzhou Kelin acquired 9.43% of Kepler's equity for 100 million yuan, corresponding to a valuation of about 1.06 billion yuan.

In May 2026, Hangzhou Kelin announced it would continue to acquire Kepler equity for no more than 300 million yuan, raising its shareholding to 51%; the overall valuation corresponding to this transaction dropped to about 722 million yuan.

In August 2026, Kepler completed its industrial and commercial registration change and became a controlled subsidiary of Hangzhou Kelin.

Kepler's 2025 revenue was 4.3372 million yuan, with a net loss of 66.9390 million yuan, but its order backlog still exceeded 47 million yuan.

Kepler robots

The acquisition is not of an already profitable business, but of a set of assets that continues to lose money yet still has orders, a team, and technology.

For Hangzhou Kelin, the key is whether Kepler's orders can be delivered and whether the complete machines can be combined with its own components and customer channels into a business.

Kepler's team, orders, and embodied intelligence capabilities are thus being revalued within Hangzhou Kelin's business.

The Kepler control transaction is valued at about 722 million yuan, while Winverse AI is seeking a post-investment valuation of about $400 million. The two figures cannot be compared directly, but placed side by side, they reveal two different pricing logics: manufacturing assets versus scarce model teams.

Hesai, 51WORLD, and Horizon provide perception, simulation, and computing capabilities respectively; Momenta and Mobileye already have real-world experience in perception, decision-making, and scaled delivery.

Their approaches to entering the robotics business are not the same.

Hesai chose to build out its infrastructure.

In the second quarter of 2026, Hesai's robot lidar shipments reached 142,400 units, up 193.4% year on year.

Hesai 2026 Q2 financial report

Beyond lidar, it also launched the spatial intelligence platform Kosmo and robot power modules. Liu Xingwei, vice president of its robot perception business, stated explicitly that the company will not build general-purpose embodied large models or robot bodies, but will provide infrastructure.

51WORLD entered from digital twins and simulation.

In March 2026, it released the 51World Model world model; in August, it launched the AperData embodied data foundation and the AperOne embodied application platform.

AperData covers data collection, governance, and training, while AperOne connects scene reconstruction, simulation training, evaluation, deployment, and operations. 51WORLD sells training grounds and deployment foundations, and does not manufacture robot bodies.

If 51WORLD wants to build a solid simulation, data, and deployment platform, it is also very likely to fill in the body downstream through investment or acquisition.

Momenta invested in data first, rather than buying a body first.

In August 2026, it led the Series A round of Jianzhi Robotics.

Jianzhi does not make bodies; it mainly builds a bodyless data system, a Data Foundation Model, and real-machine evaluation infrastructure. What Momenta invested in is data and model capability, using it to extend its autonomous driving data flywheel experience into embodied intelligence.

Horizon, meanwhile, spun its robotics business off into D-Robotics, providing chips, algorithms, and a software development platform.

D-Robotics likewise does not make bodies; instead, it lets various robot customers use its general-purpose foundation.

D-Robotics

In April 2026, D-Robotics completed a $150 million B2 round, bringing its cumulative B-round financing to $270 million. Its customers already have bodies, and acquiring a complete-machine manufacturer could instead turn it into a competitor of its own customers.

Mobileye chose direct acquisition.

In 2026, it acquired the humanoid robotics company Mentee Robotics for about $900 million, with the latter operating as an independent business unit.

Mobileye already has visual perception, autonomous driving AI, and scaled mass-production experience, but no ready-made humanoid robot body or team. It directly bought a platform already iterated to its third generation, filling in the "body" it lacked to enter physical AI.

Such companies may all fill in capabilities upstream or downstream: if they lack a "brain," they add models; if they lack a body, they add a body.

They already have factories, supply chains, and manufacturing systems, and when entering the robotics business, they care more about which capabilities need in-house development and which can be filled in through investment or acquisition.

XPeng chose to make robotics its second core business and open its equity to external capital.

Its robotics business has signed equity financing agreements exceeding $900 million, and its body, brain, cerebellum, data, and infrastructure systems have also been built.

XPeng's accumulated AI chips, supply chain, manufacturing, and globalization capabilities can all be reused in robotics; IRON is planned to first enter XPeng stores and campuses, and then be delivered to external customers.

XPeng robots

For automakers, the key is not whether to buy a robotics company, but whether that company's capabilities can be integrated into their existing technology and product systems.

BYD is taking a "self-development + investment" route.

It has confirmed that it is developing humanoid robots and has invested in AgiBot and PaXini Tech.

AgiBot provides complete machines and embodied intelligence capabilities, while PaXini provides tactile sensors and dexterous hands. BYD can validate products in its factories and then fill in external technical capabilities through investment.

Geely is closer to binding external teams with capital.

In November 2025, Geely Capital led a nearly 1 billion yuan Series A+ round in Robot Era.

Robot Era works on both embodied large models and robot bodies, and is already seeking orders in logistics, manufacturing, and other scenarios. For Geely, investing in a company that already has models, bodies, and an engineering team is faster than building a full-stack team from scratch.

Robot Era robots

For Robot Era, an industrial shareholder like Geely brings not only capital but also the opportunity to enter a real production system.

Therefore, automakers will not naturally back struggling robotics companies.

Whether the invested company's capabilities can enter factories, supply chains, or sales networks determines whether the investment can create synergies with the automaker's own business. Otherwise, automakers can choose to develop in-house or invest only in leading teams.

Sequoia and Unitree have begun jointly seeking early-stage projects, leading companies continue to raise funds and make acquisitions; Kepler is selling control at a lower valuation, and more than 20 robotics companies have gone bankrupt, dissolved, liquidated, or shut down operations within 19 months.

Capital is not simply retreating; it is re-stratifying.

The value of M&A is not simply clearing struggling companies off the list, but allowing still-useful teams, technology, data, and orders to enter larger business systems.

For embodied intelligence companies with long-term losses, being acquired by a company with scenarios, products, and channels may not be worse than struggling to complete the next financing round; for the industry, this can also reduce investment in homogeneous bodies and duplicate production lines.

For mid-tier robotics companies, the real question is no longer "who can save me," but "who am I useful to."

Whether orders can be verified, whether the team can be embedded, and whether the technology can fill the buyer's shortcomings will determine whether it is invested in, acquired, or liquidated.

The industry's dividing line is also shifting from "who can get the next financing round" to "who can enter another real business."

This article is from the WeChat public account "Chaoyong AI," author: Yan, published by 36Kr with authorization.