Might as well sell it to Liang Wenfeng
Just after hiring a CFO, Liang Wenfeng steps into the venture capital sector.
The two people he intends to invest in have surfaced: one is Hu Han, a former scientist at Tencent, and the other is Yu Tao, former head of foundation models at Exzeo Dynamics. Both of their startups are reported to focus on the embodied intelligence track. A source close to Hu Han also told us that he has indeed started his own business, with the direction related to embodied brains, and is currently recruiting new employees.
At this moment, the coding capability is about to peak and enter a period of slow growth, and GPT-6 Astra has just demonstrated powerful foundational general model capabilities. Timely, DeepSeek has shown strong interest in embodied intelligence. On the other side, the embodied intelligence industry is facing tightening financing and blocked IPO processes.
The sentiment of "selling the company to Liang Wenfeng" has begun to spread in the embodied intelligence circle. Even a joke has gone viral: it is better to set up the company opposite DeepSeek, so that at least Liang Wenfeng can see it in a physical sense.
When large model companies are preparing to step into the embodied intelligence era, some opportunities will fade away, and some enthusiasm will repeat itself.
Liang Wenfeng Enters the Game, Embodied Intelligence Founders Line Up to Seek Audience
When doing VC, Liang Wenfeng seems to prefer to arrive a little earlier — both of the two companies he invests in are still operating under the radar.
Bloomberg revealed that DeepSeek has agreed to lead the multimillion-dollar seed round financing for Winverse AI, a startup founded by former Tencent scientist Hu Han, with a post-money valuation of about 400 million US dollars. However, people familiar with the matter said that the transaction has not been finally concluded, and the financing amount, valuation and investor participation are still subject to change.
As for what kind of company Winverse AI is, public speculation comes from the entrepreneurial direction Hu Han revealed when he left Tencent in July: focusing on embodied intelligence and world models.
A researcher close to Hu Han told us that his entrepreneurial direction is related to embodied brains, and he is currently recruiting new employees, with recruitment requirements aligned with the large model industry.
In addition to Hu Han, Yu Tao, former head of foundation models at Exzeo Dynamics, is also reported to have started his own business, with the direction of LLM + embodiment. According to relevant sources, the valuation of his company is 70 million US dollars in the first round and 200 million US dollars in the second round. In addition to a leading institution and a top-tier fund, DeepSeek is also among the investors.
There have been previous market reports that Liang Wenfeng has recently paid close attention to the embodied intelligence field intensively, communicated and studied with many entrepreneurs in this field, and discussed directions such as robotics, interaction between large models and the real world.
In addition, Liang Wenfeng is currently considering whether DeepSeek should further enter the embodied intelligence field. The plan may include forming a team to conduct in-house R&D, or cutting into the track by investing in related enterprises.
The grand occasion where investors once lined up to meet Liang Wenfeng for financing shares may now be reproduced by embodied enterprises. At the same time, some job seekers in the embodied intelligence field have also begun to look forward to it: is there a chance to work on robotics at DeepSeek?
Early Foreshadowing
What is also worth noticing is the timing of DeepSeek's entry into the embodied intelligence track.
In the past, there was actually a fear of difficulties in doing embodied intelligence in the large model circle. But the powerful foundational general model capabilities of GPT-6 Astra released in September are obvious to all, which has refreshed the perception of large model companies: OpenAI's large model team does not understand embodied intelligence, but they still developed a model that shocked the entire embodied intelligence circle.
Another fact is that the coding capability has visibly reached its peak, and will enter a slow growth stage afterwards. For large model companies, they need to find new outlets to sustain continuous impressive breakthroughs.
"In the past, large model companies may have thought that VLA had no possibility of achieving general intelligence, and they had more important things to do such as coding. But GPT-6 Astra has brought a qualitative change, letting all large model companies see that they can develop such a generalized embodied model without even understanding embodied intelligence."
A researcher in the embodied intelligence field told us that entering the embodied intelligence track is an inevitable choice for large model companies and large tech groups. Across the globe, whether it is OpenAI, Anthropic or Meta, all have formed their own robotics teams.
In fact, when Liang Wenfeng disclosed DeepSeek's AGI roadmap in May this year, he also released the rhythm of embodied intelligence: language model → Chain of Thought → Agent → continuous learning → self-iteration → embodied intelligence, breaking down the grand AGI goal into progressive, implementable and verifiable technical steps.
DeepSeek has now accelerated to the step of embodied intelligence.
Earlier, the AI circle guessed from DeepSeek's recruitment dynamics whether it would lay out the embodied intelligence business — the job description for its recruited multimodal understanding researcher shows that AGI will never stop at text. A key step towards AGI is to enable the model to have the ability to understand the physical world.
However, it is not simple for large model companies to enter the entire embodied intelligence field. "Robotics is a matter of software and hardware coupling, and large model companies have limited landing scenarios and data, so it is more practical to cut in from the embodied brain," an investor told us.
Therefore, it is conceivable that in the future, we will very likely see more outward investment and even acquisition of embodied intelligence companies by leading large model companies including DeepSeek.
"At present, most embodied teams still think in the logic of VLA and WAM. If DeepSeek invests in and acquires companies, embodied teams with large model background and spatial intelligence capabilities will be more favored," the aforementioned researcher in the embodied intelligence field mentioned.
Embodied Intelligence Enterprises No Longer Wait for IPO
"Hoping to be favored by Liang Wenfeng", an emotion is spreading in the embodied entrepreneur circle.
After all, the current reality is too harsh. Since the second half of the year, the embodied intelligence industry has cooled down significantly. It is first reflected in financing — over the past year and more, the embodied intelligence track was extremely hot. In the first half of 2026 alone, the total financing amount of the domestic embodied intelligence track exceeded 900 billion yuan, 5 times that of the same period in 2025, and more than 20 embodied companies with a valuation of 10 billion yuan were born. But the trend changed suddenly in the second half of the year. In many leading institutions, the number of embodied intelligence projects submitted for review has decreased significantly. Financial advisors in the circle also clearly feel that the embodied projects in their hands are difficult to promote.
"The leading embodied companies in the market have raised all the money they can raise, and their valuations are already very high, so it is difficult for institutions that have not invested to get in later." An investor said.
At the same time, top-level guidance has also returned to rationality. In August this year, Li Chao, Deputy Director of the Policy Research Office of the National Development and Reform Commission, said that the robotics industry must prevent blind following and swarming. A public piece of news is that The Information reported that regulators have issued informal "window guidance", which lists the conditions of having regular revenue capabilities, being on the track of narrowing losses, or having achieved real technological innovation as the scope of IPO approval for humanoid robots.
"IPO is getting harder" has become an indisputable fact in the industry.
But outside the Hong Kong Stock Exchange, there are about 40 to 50 robotics and embodied intelligence related enterprises queuing up for listing, which is the most crowded track in the 18C Specialised Technology Company channel. We recently learned that a number of embodied intelligence investors have begun to seek exit through selling old shares.
The law of history is that the prosperity of the industry often appears first with universal optimism, and finally ends with differentiation. The key point that determines differentiation, in any case, will never be "how much money has been raised before".
As Xia Zhijin, Managing Partner of Vertex Ventures China, mentioned at the 2026 Investment Media SuperLink Conference, at this stage, there is no huge technical gap between the 10-billion-yuan valuation embodied intelligence companies and startup teams. "The only thing that can be called a threshold is the financing threshold, some people have raised more money."
Getting money is only the first step. When the IPO prospect is unclear, "being acquired" is also an acceptable destination — everyone is starting to wait for Liang Wenfeng.
This article is from the WeChat official account "Investment Media" (ID: pedaily2012), written by Feng Yuchen, authorized for release by 36Kr.