Winverse AI: DeepSeek Leads Seed Round for Ex-Tencent Scientist Han Hu's Robot Brain Startup

The deal

According to Bloomberg, first reported on October 12, 2026 and confirmed by Caixin, Sina Finance, 36Kr, Pingwest and Yicai, DeepSeek has agreed to lead Winverse AI's seed round at a post-money valuation of roughly $400 million. The round size is described as "tens of millions of US dollars" — most likely in the $30-50M range. Meituan's venture arm Long-Z Investments (龙珠资本) has also committed capital. People familiar with the matter cautioned that the deal has not closed and valuation and investor list may still shift.

Who is Han Hu?

Han Hu is not a typical robot startup founder. He is the lead author of the Swin Transformer paper (2021), a landmark in computer vision whose hierarchical transformer architecture reshaped modern perception models. A Tsinghua PhD, he previously worked at Microsoft Research Asia and Baidu, and at Tencent he led the Hunyuan vision-language and embodied foundation model team as chief scientist. He left Tencent in 2026 to found Winverse AI.

What does Winverse AI actually build?

The company is still in stealth mode and has not publicly disclosed a product. The positioning, per Pingwest and 36Kr, is to build robot AI models — the software that lets a humanoid understand spoken instructions, reason about unseen environments, and execute complex multi-step tasks, as opposed to the low-level motion control that robot makers have historically handled in-house. In short, Winverse AI is aiming to be the "brain" layer, not the body.

One nuance from 36Kr: some investors view Winverse AI less as a full-stack embodied-AI company and more as a model efficiency and training acceleration play — which is exactly DeepSeek's home turf. The bet is that Han Hu, with his CV foundation-model background, can build VLA (vision-language-action) stacks that train and infer cheaply, the same way DeepSeek cut LLM inference costs.

Why this matters for the Chinese robotics stack

The DeepSeek bet is a structural signal: rather than competing with OpenAI's Figure AI by building a humanoid itself, DeepSeek founder Liang Wenfeng has reportedly been meeting intensively with embodied-AI founders across China. The strategy is to supply the general-reasoning backbone (DeepSeek LLMs) and back the VLA application layer (Winverse AI) for the whole Chinese humanoid supply base — Unitree, Fourier, UBTech, etc.

What to watch

• First product reveal. Winverse AI has not shown any robot demo. Expect a 2027 launch tied to a hardware partner — likely a Chinese humanoid maker rather than a DeepSeek-branded robot.
• DeepSeek's embodied strategy. Liang Wenfeng's reported meetings with multiple embodied founders suggest more investments may follow.
• Talent retention. Han Hu's team is drawn from Tencent Hunyuan, MSRA and Baidu; retention through the stealth phase is the near-term risk.
• Closed round confirmation. The deal has not formally closed as of press time.

Frequently asked questions

What is Winverse AI?

Winverse AI is a stealth-stage Chinese embodied-AI startup founded in 2026 by Han Hu (胡瀚), former chief scientist at Tencent for the Hunyuan vision-language and embodied foundation model team. The company builds AI models that let humanoid robots understand spoken instructions and handle complex real-world tasks, rather than only executing pre-programmed motions.

Who is funding Winverse AI?

DeepSeek has agreed to lead Winverse AI's seed round at a valuation of roughly $400 million, according to Bloomberg and Chinese financial media reports on October 12, 2026. Meituan's Long-Z Investments (龙珠资本) has also committed. The round has not yet closed and terms may still change.

Who is Han Hu, Winverse AI's founder?

Han Hu is the author of the Swin Transformer, one of the most influential computer-vision papers of the 2020s, and a Tsinghua PhD who previously worked at Microsoft Research Asia and Baidu. At Tencent he led the vision-language and embodied foundation model team before leaving to found Winverse AI in 2026.

Why is DeepSeek investing in robotics?

The DeepSeek bet is on the robot 'brain' rather than the body. Rather than building its own humanoid hardware, DeepSeek founder Liang Wenfeng is reportedly meeting with embodied-AI founders across China and backing the model-layer companies — like Winverse AI — that will supply the reasoning and VLA (vision-language-action) stacks for Unitree, Fourier, UBTech and other Chinese robot makers. This mirrors how OpenAI backed Figure AI.

How does Winverse AI compare to other Chinese embodied AI startups?

Winverse AI is a software-only 'robot brain' company, positioning it against Pudong-based Galbot, Beijing-based Physical Intelligence (PI) rivals and Tsinghua-founded AgiBot's internal model team. Its $400M seed valuation puts it immediately in the top tier of Chinese embodied-AI startups, on par with 1X and Figure AI's early rounds.