Productivity Permit-Ready Blueprints in Minutes: Blueprints AI Turns a Sketch Into Code-Checked Construction Documents at $3,200 a Month Business DeepSeek Nears a Tens-of-Millions Seed Bet on Robot-AI Startup Winverse at Roughly $400 Million, Bloomberg Reports News 32,768 CPU Cores in a Single Rack: AMD Pitches Helion, an Open OCP Reference Design, as the Missing Half of Agentic AI Infrastructure Meta Meta Launches Muse Wonder and Muse Wonder Full, Claiming Frontier-Level Performance Against GPT-6 Astra and Fable 5.1 Reviews 45 Percent Ready, by Its Own Admission: ArtCraft AI-Built Adobe Clones Pass 86,000 Stars and Twelve Apps - Here Is Where to Try Them Claude HCLTech Signs a Global Premier Partnership With Anthropic and Stands Up a Dedicated Claude Business Unit Security Six Major Agent Frameworks Fall to a 'Repeat-After-Me' Prompt, and 118 Open-Source MCP Servers Ship Verified Bugs News NVIDIA and Cisco Introduce USFP Optics: 400 Gb/s Per Lane and 12.8 Tb/s Per Module for AI Factories Productivity Permit-Ready Blueprints in Minutes: Blueprints AI Turns a Sketch Into Code-Checked Construction Documents at $3,200 a Month Business DeepSeek Nears a Tens-of-Millions Seed Bet on Robot-AI Startup Winverse at Roughly $400 Million, Bloomberg Reports News 32,768 CPU Cores in a Single Rack: AMD Pitches Helion, an Open OCP Reference Design, as the Missing Half of Agentic AI Infrastructure Meta Meta Launches Muse Wonder and Muse Wonder Full, Claiming Frontier-Level Performance Against GPT-6 Astra and Fable 5.1 Reviews 45 Percent Ready, by Its Own Admission: ArtCraft AI-Built Adobe Clones Pass 86,000 Stars and Twelve Apps - Here Is Where to Try Them Claude HCLTech Signs a Global Premier Partnership With Anthropic and Stands Up a Dedicated Claude Business Unit Security Six Major Agent Frameworks Fall to a 'Repeat-After-Me' Prompt, and 118 Open-Source MCP Servers Ship Verified Bugs News NVIDIA and Cisco Introduce USFP Optics: 400 Gb/s Per Lane and 12.8 Tb/s Per Module for AI Factories

DeepSeek Nears a Tens-of-Millions Seed Bet on Robot-AI Startup Winverse at Roughly $400 Million, Bloomberg Reports

DeepSeek Nears a Tens-of-Millions Seed Bet on Robot-AI Startup Winverse at Roughly $400 Million, Bloomberg Reports

DeepSeek has agreed to lead a seed round of tens of millions of dollars in Winverse AI, a robotics foundation-model startup founded in 2026 by ex-Tencent distinguished scientist Hu Han, Bloomberg reported, citing people familiar with the deal. Meituan-backed Long-Z Investments has also committed. The deal, which would value Winverse at about $400 million, has not closed and terms may still change.

DeepSeek is preparing its largest external startup investment yet, and it is aimed at robots. According to a Bloomberg report published Monday, the Chinese AI lab has agreed to lead a seed round of tens of millions of dollars in Winverse AI, a company founded this year to build AI models that let robots understand voice instructions and handle messy real-world tasks rather than just execute preset motions. Bloomberg cited people familiar with the matter; the deal has not closed and the valuation and investor list could still change.

The round would value Winverse AI at roughly $400 million, a striking figure for a months-old startup, and Meituan-backed Long-Z Investments has made a commitment to participate, according to the report. For DeepSeek, whose models are best known for text reasoning, the move is the clearest signal yet that it intends to push beyond language models into embodied intelligence.

Winverse founder Hu Han brings heavyweight credentials. A Tsinghua University PhD, he was a distinguished scientist at Tencent, where he led the Hunyuan vision-language and embodied foundation model teams and founded its speech team, after serving as a principal researcher at Microsoft Research Asia and stints at Baidu. He is best known as one of the four co-authors of the Swin Transformer paper, which won the Marr Prize for best paper at ICCV 2021 and became a backbone of modern computer vision.

DeepSeek has edged into robotics before, buying a small stake in Unitree Robotics and participating in the humanoid makers August IPO with about 141 million yuan, with an eye to co-developing embodied models. The Winverse deal, if completed, would be several times larger and its most significant bet on an independent startup. It also fits the funding picture: Bloomberg has separately reported that DeepSeek is close to raising at least 80 billion yuan, about $12 billion, ahead of a planned early-2027 IPO, and that some of that capital may fund joint development with younger companies.

The context is a funding race that has gone global. American peers Skild AI and Physical Intelligence, which market general-purpose robot brains, are valued at about $14 billion and $5.6 billion respectively with backing from NVIDIA and Alphabet, and PitchBook estimates robotics and physical AI startups have raised roughly $48 billion this year. China leads the world in humanoid hardware manufacturing, with Unitrees $900 million IPO in August as proof, but the open question everywhere is the same one Winverse is chasing: giving robots the perception and reasoning to operate outside scripted environments.

Neither DeepSeek nor Winverse has publicly confirmed the investment, and Bloomberg notes the terms remain fluid. But the direction of travel is hard to miss: the most cost-efficient model lab in China is starting to spend its new capital downward into the physical world, and the first test of that thesis will be whether a Swin Transformer veteran can give robots a brain worth $400 million.

Comments (0)

Log in to join the discussion

Log In

No comments yet