DeepSeek leads $400M seed round for ex-Tencent scientist's robotics-AI startup Winverse AI

DeepSeek is leading a seed round for Winverse AI, a stealth Chinese robotics-AI startup, in a deal reported to be worth tens of millions of dollars that would value the young company at roughly $400 million post-investment ^[1]. For a research lab barely three years old, writing a check at this scale into an outside startup - rather than building the capability in-house - is itself a tell.

Reporting frames this as one of DeepSeek's largest startup investments since the company's 2023 founding ^[2], and the scale matters because it marks a real shift in where DeepSeek is willing to spend. Winverse makes no language-model promises - it's a robotics-AI lab, full stop. Backing it with tens of millions of dollars says DeepSeek sees embodied AI as worth founder-level conviction, not just a side bet.

The timing is what makes this interesting rather than routine. DeepSeek is simultaneously closing in on its own roughly $12 billion (80 billion yuan) financing round, backed by Tencent and CATL ^[3], raised ahead of a planned early-2027 stock listing ^[4]. That capital was pitched to backers as fuel for model training and data-center buildout. Diverting a slice of it into seeding an outside startup, while still mid-raise on its own account, is an unusual sequencing choice for any lab.

Reporting suggests the capital raised for training and infrastructure may now also bankroll a new generation of co-development startups, of which Winverse is a case ^[5]. Read generously, this is DeepSeek using its balance sheet as a platform play - fund promising teams outside the building rather than hire every specialist internally. Read skeptically, it's a sign DeepSeek has more capital committed than it has immediate infrastructure spend to absorb, and robotics is where the overflow is landing first.

The more technical reason to care sits inside what Winverse is actually trying to build. By most accounts, Chinese humanoid robots still lack the autonomy to take a plain-language instruction and execute it reliably in an environment they haven't seen before ^[6][7]. That's a software gap, and it happens to be exactly the problem Winverse's founder, a former Tencent vision-language and embodied-AI scientist, spent his career working on before starting the company.

One sharper framing from a single AI-focused commentator on X cuts against the ex-Tencent-scientist headline entirely: DeepSeek's own models are strong on text and code, but training a frontier model on video, 3D scenes, and interactive environments requires exactly the kind of real-world data a robotics lab generates by default and a pure-text lab cannot manufacture on its own. Under that reading, the Winverse bet isn't just about robots - it's about plugging a data gap in DeepSeek's own training pipeline that no amount of additional text or code can fill.

This deal doesn't exist in isolation. Two months earlier, DeepSeek took a 2.31% strategic stake in Unitree Robotics ahead of its Shanghai IPO - a stake that came with a 36-month lock-up, roughly triple the typical hold period for a strategic investor ^[6]. Analysts read that lock-up as a sign DeepSeek wanted a durable partnership, not a quick return.

Put the two deals side by side, with Meituan's Long-Z Investments also committing capital to the Winverse round ^[8], and a structure emerges: Unitree supplies the physical hardware, Winverse supplies the AI layer that tells the hardware what to do, and DeepSeek sits across both positions instead of trying to own the entire stack itself. Morgan Stanley's China industrials team has separately flagged a broader wave of Chinese humanoid-robot IPOs moving closer to completion in the second half of 2026, igniting wider market interest in the sector ^[10]- a sign DeepSeek is positioning early in a category the rest of the market is only now starting to price.

None of this is locked in yet. Multiple reports note the round is not final and both the final valuation and the roster of participating investors could still shift before it closes ^[1][9]. That's worth sitting with rather than glossing over - DeepSeek is committing to lead a round whose terms it hasn't fully locked down, at the same time it's finalizing a $12 billion raise of its own.

That sequencing reads less like carelessness and more like confidence. A lab moving cash this freely into an unfinished deal, stacked on top of an already long-locked Unitree position, is signaling that it can afford to be patient about terms because it isn't worried about having the capital to close. For the broader embodied-AI market, that's the real headline: the gap between 'foundation-model lab' and 'robotics investor' is shrinking, and the labs with the deepest balance sheets are the ones doing the shrinking.