DeepSeek backs $400M robot AI round, pushing China’s AI race into the physical world

The headline check: DeepSeek backs a $400M robot-AI bet

DeepSeek’s latest reported move is not another chatbot launch, benchmark claim or cloud-price cut. It is a step toward robots: the Hangzhou AI lab has agreed to lead a seed financing round for Winverse AI, a months-old startup building artificial intelligence for robots, in a deal that would value the company at about $400 million after the investment [1]. That distinction matters. Several headlines frame the story as a “$400 million round,” but the more precise reported structure is a seed round of tens of millions of dollars at a roughly $400 million post-money valuation [1].

Winverse is still operating largely in stealth, yet the profile of its founder explains why the round has attracted attention. Bloomberg’s report, republished by Yahoo Finance, identifies the company as founded in 2026 by Hu Han, a former Tencent scientist who led vision-language and embodied foundation-model teams and whose prior career included Microsoft Research Asia and Baidu [1]. Quartz likewise reported that Hu previously led Tencent’s vision-language and embodied foundation-model work before creating Winverse [2].

For DeepSeek, the proposed deal marks a shift from model prestige to ecosystem control. The company is reportedly drawing on capital from its own financing of more than $12 billion, money originally expected to support model training and data-center expansion but now also available for startup investments and co-development [1]. If the transaction closes as reported, DeepSeek would be using its balance sheet not only to train foundation models, but also to shape where those models are deployed.

Why Winverse matters

The story is important because robotics is becoming a new test of foundation-model economics. Chatbots generate enormous usage data, but they still live inside screens. Robots need models that can interpret speech, see the environment, plan actions, adapt to uncertainty and convert decisions into safe motion. A capable “robot brain” has to bridge perception, reasoning and control.

That is why Winverse’s focus is strategic even though the company has not publicly disclosed a product. Dealroom describes Winverse as a 2026 China-based startup developing AI-powered models for humanoid robotics, with DeepSeek and Meituan’s Long-Z Investments listed as investors in an October 2026 seed round at a $400 million valuation [3]. The same profile places the company in categories such as robot autonomy, motion control, AI model layers and humanoid robotics [3].

The investors are also significant. The Bloomberg report says Long-Z Investments, affiliated with Meituan, has joined commitments to the round alongside DeepSeek [1]. Meituan’s presence hints at the breadth of possible demand: delivery, logistics, warehouses, local services and commercial automation all depend on turning AI reasoning into reliable physical actions.

There is still uncertainty. The reported transaction has not been finalized, and valuation and investor participation may change [1]. DeepSeek, Long-Z and Hu Han did not respond to Bloomberg’s requests for comment, according to the report [1]. That makes this a live financing story rather than a closed corporate announcement.

From language models to embodied intelligence

DeepSeek’s interest in Winverse fits a broader Chinese push into embodied intelligence: AI systems that act in the physical world. Quartz reported that DeepSeek previously bought a strategic stake in Unitree Robotics before Unitree’s Shanghai listing, with analysts interpreting the position as a partnership rather than a purely financial bet [2]. Bloomberg’s report also said DeepSeek had bought about a 2% stake in Unitree to collaborate on embodied-intelligence models before the Winverse move [1].

This pattern suggests DeepSeek is building options across the robotics stack. Unitree gives exposure to robot hardware and humanoid deployment. Winverse could give exposure to the model layer that lets robots understand instructions and operate in unpredictable settings. Together, those pieces point toward a strategy in which the frontier model company is no longer content to sell APIs or release weights; it wants influence over the next interface for AI.

A fresh Chinese report from Phoenix Technology added another layer, saying DeepSeek founder Liang Wenfeng has recently been meeting embodied-intelligence entrepreneurs and weighing whether DeepSeek should enter the field directly through an internal team or through investment [5]. The same report cited industry sources saying Winverse may be closer to lower-level infrastructure and model acceleration than a strict embodied-intelligence company, a useful caveat for investors who assume every robotics-AI startup is building a full humanoid brain [5].

Unitree’s GPT-6 forecast raises the stakes

The other part of the story is timing. Unitree founder and CEO Wang Xingxing said on October 12 that OpenAI’s latest GPT-6-related advances could move the robotics industry closer to a breakthrough within one or two years [4]. South China Morning Post reported that Wang suggested a GPT-7 model arriving next year could potentially solve difficult embodied-intelligence problems, while framing the comment as an industry forecast rather than a disclosed OpenAI roadmap [4].

Wang’s comments matter because Unitree is one of China’s best-known robot makers. According to SCMP, he had previously said in August that embodied intelligence might need two to three years at the fastest, or as long as a decade, to reach its “ChatGPT moment” [4]. His latest remarks shortened the optimistic timeline to one or two years after testing and observing newer AI model capabilities [4].

Tech Edition reported the same broad message: Wang said advanced AI models had changed his expectations for how quickly robots could become more independent in real-world environments [6]. He described stronger model reasoning and multimodal understanding as possible accelerants for robots that must interpret instructions, understand scenes and perform physical tasks [6].

Still, the caveat is essential. Wang’s prediction is not proof that OpenAI has a robotics-ready system, nor is it a guaranteed GPT-6 or GPT-7 launch schedule. It is a market signal from a major robot maker: the hardware side increasingly expects general models to reduce the amount of custom engineering needed for each robot task.

The China angle: capital, talent and deployment

China’s advantage in this race is not only model talent. It also has hardware supply chains, manufacturing density and a rapidly expanding humanoid-robot ecosystem. The problem is that good motors, joints and sensors do not automatically create useful autonomy. Robots still struggle with generalization, changing environments, safety constraints and the long tail of physical tasks.

That is where capital is now moving. A $400 million valuation for a months-old stealth startup suggests investors believe the scarce asset may be the software layer between foundation models and machines. DeepSeek’s own reported financing of more than $12 billion gives it the ability to fund expensive model training while also buying strategic stakes in companies that can supply real-world data and deployment channels [1].

If the Winverse round closes, the company could become one of the more closely watched robotics-AI startups in China. But the story should be read with precision: this is not yet a public product launch, not a confirmed $400 million cash injection and not a guarantee of near-term humanoid autonomy. It is a large strategic signal at the intersection of AI models, robotics software and China’s manufacturing base.

What to watch next

The next milestone is whether the Winverse financing is finalized, at what valuation, and with which investors. A second question is whether DeepSeek’s role becomes passive capital or active technical collaboration. A third is whether Winverse discloses what part of the robot-AI stack it actually targets: visual-language planning, robot-control policies, model acceleration, simulation, data pipelines or full embodied agents.

The bigger question is whether foundation-model companies can escape the screen. DeepSeek’s reported Winverse bet and Unitree’s shortened breakthrough timeline both point in the same direction: the next AI contest may be fought not only in chat windows, but in factories, warehouses, labs and homes. If embodied models become the next training frontier, the most valuable dataset may be the messy physical world itself.

Sources from the last 72 hours

1. [1]DeepSeek Backs Ex-Tencent AI Scientist’s $400 Million Robot LabOct 12, 2026, 1:42 PM
2. [2]DeepSeek is moving to lead a $400 million seed round for a Chinese robotics AI startupOct 12, 2026, 2:00 AM
3. [3]Winverse AI company information, funding & investorsOct 12, 2026, 2:00 AM
4. [4]OpenAI’s GPT-6 may lead to robotics breakthrough as soon as next year: Unitree founderOct 12, 2026, 3:30 PM
5. [5]独家｜DeepSeek考虑入局具身智能，梁文锋正密集与相关创业者交流Oct 12, 2026, 3:32 PM
6. [6]OpenAI’s GPT-6 could accelerate robotics breakthrough by next year, says Unitree founderOct 12, 2026, 2:00 AM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.