DeepSeek spent its first years building a reputation as the lab that gives away frontier models for almost nothing. Now it is spending like a venture fund. Bloomberg reports that the Hangzhou lab has agreed to lead the seed round of Winverse AI, a stealth robotics startup founded by a former top Tencent scientist, in a deal that values the months-old company at about $400 million. Meituan's Long-Z Investments is also participating.
The money trail
This is DeepSeek's second robotics move, and the pattern matters more than any single check:
| Date | Move | Amount | Why it matters |
|---|---|---|---|
| Early 2026 | First external financing | ~50B yuan (~$7.5B) at ~400B yuan valuation | Ended years of self-funding; founder Liang Wenfeng kept control |
| Aug 2026 | Unitree Robotics IPO investment | 140.8M yuan (~$20.8M) | First verified robotics bet, with a two-way procurement pact |
| Oct 6, 2026 | New raise sought | 80B yuan (~$12B), Tencent and CATL leading | War chest for whatever comes next |
| Oct 12, 2026 | Winverse AI seed (lead) | Tens of millions; $400M post-money valuation | Stealth embodied-AI startup, ex-Tencent founder |
The amounts tell the story: DeepSeek is raising in the tens of billions and deploying in the tens of millions, which means the spending has barely started.
Why robots, why now
The Unitree deal is the template. DeepSeek did not just buy shares in the humanoid robot maker's Shanghai IPO; it signed a procurement pact, Unitree gets priority buying model-training services from DeepSeek, and DeepSeek gets priority buying robots and pursuing embodied-AI applications. Capital buys a relationship, not just equity.
The technical bet is clear too. Large language models are good at predicting the next word and bad at predicting the next joint angle, which is the gap holding humanoid robots back from general-purpose use. DeepSeek's V4-Pro, released in late April, reportedly performs close to GPT-5.5 and Claude Opus 4.7 at a significantly lower API cost. If cheap, capable models are the commodity, the scarce asset is the physical platform they run on. The same month an AI agent designed a full RISC-V CPU in 12 hours, the models, the chips, and now the robot bodies are all being built by AI labs at once.
What it means
DeepSeek is becoming something new: a research lab that also acts as a strategic investor, steering its supply chain with capital. The Winverse round is small relative to the $12 billion being raised, which suggests it is the first of several deployments rather than a one-off. And the broader signal is about China: with investors pricing agent startup Manus at $4 billion and Tencent and CATL backing DeepSeek's own raise, the country's AI capital is moving from pure software into the physical world. The next frontier is not a bigger chatbot. It is a machine that can walk into a room.
By ByteHerald Staff. Deal terms via Bloomberg's reporting, citing people familiar with the matter; DeepSeek has not confirmed the fundraise. "Tens of millions" is the reported range, not an exact figure.
Comments
Post a Comment