Just hired a CFO; Liang Wenfeng enters venture capital.
Two individuals have emerged as his investment targets: Hu Han, a former scientist at Tencent, and Yu Tao, former head of the foundational model team at ZhiJi Dynamics. Both of their startups are rumored to be in the embodied intelligence space. A source close to Hu Han also confirmed to us that he has indeed started a company focused on embodied intelligence and is currently hiring.
At this moment, as coding ability is nearing its peak and entering a phase of slow growth, and GPT-6 Astra has just demonstrated powerful foundational general model capabilities, perfectly timely,
DeepSeek
Expressing strong interest in embodied intelligence. On the other hand, the embodied intelligence industry is facing tightened financing and stalled IPOs.
The remark “sell the company to Liang Wenfeng” has begun to circulate within the community. A joke has even emerged: why not move the company to
DeepSeek
Opposite side, at least let Liang Wenheng see it in a physical sense first.
As large model companies prepare to enter the era of embodied intelligence, some opportunities will fade away, while others will reignite.
Liang Wenfeng has entered the scene, and founders of embodied AI companies are lining up to meet him.
As a VC, Liang Wenfeng seems to prefer getting in early—both companies were still below the radar.
Bloomberg reported that,
DeepSeek
Has agreed to lead the seed round of funding for startup Winverse AI, founded by former Tencent scientist Hu Han, with an investment amounting to tens of millions of dollars, resulting in a post-money valuation of approximately $400 million. However, sources familiar with the matter indicate that the deal has not yet been finalized, and the investment amount, valuation, and participation of investors may still change.
Public speculation about what Winverse AI specifically is stems from Hu Han’s disclosed entrepreneurial focus when he left Tencent in July: embodied intelligence and world models.
A researcher close to Hu Han told us that his entrepreneurial focus is on embodied intelligence, and he is currently hiring, with requirements aligned to those of the large model industry.
Aside from Hu Han, Yu Tao, former head of the pre-training foundational model at Zhujì Dynamics, has also launched a startup focused on LLM + embodied AI. According to insiders, the company’s valuation reached $70 million in its first round and $200 million in its second round, with participation from leading and top-tier funds.
DeepSeek
Also among the investors.
The market previously reported that Liang Wenfeng has recently been closely following the field of embodied intelligence, engaging in discussions and learning with multiple entrepreneurs in this area, exploring directions such as robotics, large models, and interaction with the real world.
In addition, Liang Wenheng is currently considering
DeepSeek
Should we further enter the field of embodied intelligence? Options may include building an in-house team to develop the technology directly, or investing in related companies.
The scene where investors once lined up to meet Liang Wenfeng for funding opportunities may soon be replayed by embodied AI companies. Meanwhile, some embodied AI job seekers are also beginning to harbor quiet hopes: the chance to join
DeepSeek
Do you work in robotics?
There were early signs
Worth checking out, also
DeepSeek
The timing to enter the embodied AI space.
In the past, there was a sense of apprehension within the large model community toward embodied AI. However, the powerful foundational general capabilities of GPT-6 Astra, released in September, have clearly reshaped the perception of large model companies: even OpenAI’s large model team did not fully understand embodied AI, yet they still developed a model that shocked the embodied AI community.
On the other hand, the reality is that coding ability has visibly reached its peak and will now enter a phase of slow growth; for large model companies, it is necessary to identify new emerging opportunities to sustain continued innovation.
Previously, large model companies may have believed that VLAs were not capable of achieving generalization and had more important tasks, such as coding. But GPT-6 Astra has brought a qualitative shift, showing all large model companies that it’s possible to create such generalized embodied models without needing to understand embodiment.
An embodied intelligence researcher told us that entering the field of embodied intelligence is an inevitable choice for large model companies and major corporations. Globally, companies such as OpenAI, Anthropic, and Meta have all established their own robotics teams.
In fact, in May this year, Liang Wenheng disclosed
DeepSeek
When outlining the AGI roadmap, they also revealed the progression of embodied intelligence: language models → chain-of-thought reasoning → agent systems → continuous learning → self-iteration → embodied intelligence, breaking down the grand AGI goal into a series of incremental, verifiable technical milestones.
DeepSeek
It has now accelerated to the stage of embodied intelligence.
Previously, in the AI community,
DeepSeek
In its hiring updates, there has been speculation about whether it plans to venture into embodied actions—the job description for recruiting multimodal understanding researchers indicates that AGI will not stop at text; a crucial step toward AGI is enabling models to understand the physical world.
But it’s not simple for large model companies to enter the entire field of embodied intelligence. “Robotics involves tight integration of software and hardware, and large model companies themselves have limited real-world applications and data—starting with the embodied brain is more practical,” an investor told us.
Therefore, it is reasonable to imagine that in the future, we will likely see including
DeepSeek
Leading large model companies are making more external investments, and even acquisitions, in embodied AI companies.
Most embodied teams today still think in terms of VLA and WAM. If
DeepSeek
An embodied team with expertise in large models and spatial intelligence will be more favored when investing in or acquiring companies," mentioned a researcher in embodied AI.
The entity is no longer going public.
There is a prevailing sentiment among embodied entrepreneurs that they hope to catch Liang Wenheng’s attention.
After all, the current reality is too harsh. Since the second half of the year, the embodied AI industry has clearly cooled down. This is first evident in funding: over the past year and a half, the embodied intelligence sector had been booming, with domestic funding in the first half of 2026 alone exceeding 90 billion yuan—five times the amount in the same period of 2025—and more than 20 embodied AI companies achieving valuations of over 10 billion yuan. But the tide turned sharply in the second half; within many top-tier institutions, discussions around embodied AI projects have significantly decreased, and insiders in the FA community have clearly noticed: their embodied AI projects are becoming much harder to push forward.
“The top embodied AI companies in the market have already raised all the funding they can, and their valuations are already very high—making it difficult for later-stage investors to get in,” said one investor.
At the same time, high-level guidance has also turned pragmatic. In August, Li Chao, Deputy Director of the Policy Research Office of the National Development and Reform Commission, stated that the robotics industry must avoid blind following of trends and rushed, uncoordinated entry. A publicly reported development is that, according to The Information, regulators have issued informal "window guidance" specifying that humanoid robot IPO applications must demonstrate recurring revenue potential, a path toward narrowing losses, or genuine technological innovation.
"It's become an industry fact that IPOs are harder now."
Outside the Hong Kong Stock Exchange, around forty to fifty companies related to robotics and embodied intelligence are queued up for listing, making it the most crowded sector under the 18C special technology channel. We recently learned that a group of embodied intelligence investors have already begun seeking exits through secondary share sales.
The historical pattern is that industry booms typically begin with widespread optimism and conclude with differentiation. The key factor determining this differentiation will never be "how much money was raised in the past."
As Xia Zhijin, Partner at Foresight Ventures, mentioned at the 2026 Investment World SuperLink Conference, there is currently no significant technological gap between embodied AI companies with billion-dollar valuations and startups—“the only possible barrier is funding access; some have raised more money.”
Getting the money is just the first step. With the IPO outlook uncertain, being acquired is still an acceptable outcome—everyone is now waiting for Liang Wenfeng.
This article is from the WeChat public account "Investment Daily" (ID: pedaily2012), authored by Feng Yuchen.