DeepSeek has agreed to lead the first funding round for Winverse AI, a months-old stealth startup building artificial intelligence for robots, Bloomberg reported on October 12, citing people familiar with the matter. The seed round amounts to tens of millions of dollars and would value Winverse at about 400 million dollars after the investment. Meituan's Long-Z Investments has also committed capital, according to the reports. One caution belongs up front: the deal has not closed, and the terms and the investor list could still change.
Winverse was founded this year by Hu Han, who most recently led Tencent's vision-language and embodied foundation model teams. His background helps explain the valuation. He holds a PhD from Tsinghua University, previously worked at Microsoft Research Asia and Baidu, and co-authored the Swin Transformer, which won the best paper award at ICCV 2021. Startups at this stage are rarely priced on revenue. They are priced on scarce people, and a researcher who has already built influential vision systems and run embodied-AI teams at scale is exactly the kind of scarcity investors pay for.
The company itself is focused on systems that let robots perceive, understand, and act in the physical world. That is a different problem from the one DeepSeek is famous for. A language model predicts the next word from patterns in text. A robot has to look at an unfamiliar room, work out what it is seeing, and turn an instruction into physical movement that does not fail halfway through. Perception, planning, and control have to work together, in real time, with real consequences when they get it wrong.
The strategic signal is as interesting as the check. Bloomberg has reported that DeepSeek is closing in on more than 12 billion dollars in fresh financing, much of it intended for model training and data centers. Some of that capital is now flowing into a hardware-adjacent bet instead. That makes DeepSeek look less like a pure model lab and more like an investor building positions around its own technology, in the same way other frontier labs have backed the application and infrastructure layers they expect to grow around their models.
For builders, the practical lesson is to watch what is not yet known. No product, launch timeline, or technical collaboration between DeepSeek and Winverse has been disclosed, so there is nothing here to integrate, buy, or benchmark today. What the round does confirm is where smart money thinks the next bottleneck sits: not in generating fluent text, but in giving models a reliable grip on the physical world. Teams working on robotics, vision, or agent systems should treat embodied data and control loops, not bigger chat models, as the contested ground.
A 400 million dollar valuation for a months-old, still-stealth company is a bet, not a verdict. It could look prescient if Winverse turns elite research talent into robot systems that actually work outside a lab demo, and it could look frothy if perception and control prove slower to crack than the funding cycle assumes. Either way, DeepSeek has made its direction clear: the race after language is about machines that can act, and it intends to own a piece of that race, not just supply models to it.
