DeepSeek Wants to Be the Brain Inside the Robots

DeepSeek made its name proving a lean AI lab could compete with the best-funded names in the industry. Now it is spending its capital in a new direction: robot bodies. The Hangzhou-based lab has agreed to lead a seed financing round for Winverse AI, a months-old stealth startup building AI systems that let robots perceive, understand, and act in the physical world. The deal values Winverse at roughly $400 million post-money, with tens of millions of dollars changing hands.

Bloomberg first reported the transaction, citing people familiar with the matter. The deal has not closed, and terms and participants could still shift. But the headline number stands out: a company that has existed for just months is drawing a near $400 million valuation on its first outside round. That reflects intense demand for talent capable of bridging large models with real-world control systems.

The founder behind Winverse AI

Winverse AI was founded this year by Hu Han, whose resume reads like a map of the industry’s pivot from language to bodies. His last role was at Tencent, where he headed the company’s vision-language and embodied foundation model teams. Before that, he earned a PhD from Tsinghua University, spent time at Microsoft Research Asia, and did earlier work at Baidu. His research contributions include co-authoring the Swin Transformer, which earned the ICCV 2021 Best Paper award, also known as the Marr Prize.

It is a telling hire-by-investment. DeepSeek is not backing a generalist; it is backing one of the relatively few researchers in China who has led a major tech company’s embodied AI effort. The thesis of the investment is that robot intelligence is a model problem first and a hardware problem second.

The hard problem: from next word to next joint angle

The technical problem every humanoid robot company is chasing right now is the same one DeepSeek is approaching through Winverse. It needs a model that can look at an unfamiliar room and reliably turn plain instructions into physical action. Large language models are excellent at predicting the next word. They are much worse at predicting the next joint angle, and that gap is one of the biggest barriers holding humanoid robots back from general-purpose use.

Language models are excellent at predicting the next word. They are much worse at predicting the next joint angle, and that gap is what holds humanoids back.

This is why embodied AI has become the industry’s most heavily funded research frontier. A robot that can fold laundry or walk across a warehouse floor without supervision needs more than good vision. It needs a model that connects what it sees to what its motors should do, continuously, in environments that never look exactly the same twice. The companies that crack that mapping will own the intelligence layer of a hardware market that is already shipping at scale.

DeepSeek’s two-pronged push into robotics

The Winverse investment is DeepSeek’s second verified robotics move this year, and the two bets complement each other. In August, the lab paid 140.8 million yuan, about $20.8 million, for a 2.31 percent stake in Unitree Robotics, one of China’s best-known makers of legged and humanoid robots, according to a stock-exchange filing reported by Reuters. Unitree priced the offering at 150.80 yuan a share and raised about 6.1 billion yuan, valuing the company near $9 billion and making it the first mainland-listed humanoid robot maker in China.

That deal came paired with a procurement pact running both directions: Unitree gets priority when buying model-training services and technical solutions from DeepSeek, while DeepSeek gets priority when buying robots or pursuing embodied-AI applications with Unitree. Put the two investments together and the strategy is legible. Unitree supplies the bodies and the real-world deployment surface; Winverse supplies frontier research on the brains; DeepSeek supplies capital and the model-building discipline that made its name.

The money behind the strategy is substantial. In mid-June 2026, DeepSeek closed its first-ever external funding round, raising approximately 50 billion yuan, or nearly $7.4 billion, at a post-money valuation near 400 billion yuan, around $59 billion, with founder Liang Wenfeng personally putting in roughly 20 billion yuan. Bloomberg reported on October 6 that the lab was close to securing at least 80 billion yuan, about $12 billion, in a second round backed by Tencent and battery giant CATL, ahead of a potential IPO in 2027.

The US side of the race looks nothing alike

Compare that with how the American humanoid robot field is being financed. Figure AI announced in 2025 that it had exceeded $1 billion in Series C commitments at a $39 billion post-money valuation, with earlier backers including OpenAI, Nvidia, and Microsoft. Tesla’s Optimus has no standalone valuation, but Morgan Stanley has described Tesla as an important enabler in humanoid robotics, and the company has said first-generation Optimus production lines are being installed in anticipation of production in 2026. Elon Musk has described $20,000 to $30,000 as a long-term target price, not a confirmed commercial price.

The contours are clear. In the US, hardware bets are backed by Big Tech balance sheets and sovereign-scale venture capital. In China, the play is model-first: an AI lab building a portfolio of stakes and partnerships that route the intelligence layer through its own research. Both sides are making the same wager. For the last three years, the race that mattered was who builds the best chatbot. Now that is giving way to a race over who builds the best body and brain for a machine that can actually pick things up.

What to watch next

First, whether the Winverse round closes as reported, and who else joins the cap table. Meituan’s Long-Z Investments has already committed capital, and other backers may join before the ink dries. Second, what Winverse actually publishes: a months-old stealth company has a research record to establish, and the embodied AI field rewards demonstrated systems more than claimed roadmaps. Third, the Unitree partnership, which will be the first place DeepSeek’s robot-AI ambitions meet physical deployment at volume.

The stakes justify the attention. Morgan Stanley projects 1 billion humanoids in use by 2050, worth $5 trillion when supply chains, maintenance, and support join hardware sales. The International Federation of Robotics counted about 7,000 full-size humanoids sold worldwide in 2025, and Counterpoint estimates more than 22,000 shipped globally in the first half of 2026, up roughly 300 percent year over year, with Agibot holding more than 43 percent share. Humanoid startups have raised more than $13 billion cumulatively, with roughly $5.1 billion arriving in the first half of 2026 alone.

DeepSeek built its reputation showing that disciplined engineering could beat brute-force spending in language models. Whether that same discipline translates into motion control and real-world perception is the question the Winverse partnership now has to answer. The lab that made models cheaper is now betting it can make robots smarter, and the valuation it paid says it is not alone in that conviction.