DeepSeek raises $15B — and bets it all on robots

DeepSeek has closed a funding round of roughly 100 billion yuan — about $15 billion — according to Bloomberg and Reuters, making it one of the largest fundraising rounds in Chinese history. Tencent and CATL led as cornerstone investors, each putting in roughly 10 billion yuan. The round values DeepSeek at approximately 500 billion yuan ($75 billion) pre-money.

What makes this round unusual isn't just the size. It's how fast the target grew. DeepSeek initially set out to raise 50 billion yuan, then raised the target twice during negotiations, ultimately doubling to 100 billion. Combined with its 51 billion yuan first round in late June, the company has absorbed more than 150 billion yuan ($22.5 billion) in under four months — a capital intake velocity without precedent in the AI industry.

Who's in the room

The investor list tells a story about where DeepSeek thinks the next decade is going.

Sources: Bloomberg, 36Kr, October 22, 2026 (UTC+8).

The shift is striking. DeepSeek's early backers were largely internet companies and financial investors. This round pulls in semiconductor capital (SMIC's fund), automotive (Geely), energy hardware (CATL), and gaming (Perfect World, Shiji Hualong). That's not a random collection — it maps directly onto the three things DeepSeek is now spending on: domestic chip adaptation, embodied intelligence/robotics, and AI-generated game content.

Two special-purpose funds channeled public-market money into the round. "Tianjin Lisi Xingshen" raised 7.52 billion yuan and "Shixiang Shangshi" raised 3.65 billion yuan, both fully closed. Lisi Capital — founded by former Sequoia China partner Cao Xi and former Boyu Capital partner Tim Wang — has become the key conduit connecting A-share listed companies to DeepSeek's equity.

The IPO is real now

DeepSeek has retained CITIC Securities to prepare for a listing on Shanghai's STAR Market, according to Bloomberg and Chinese financial media. The company plans to file its application as early as the end of 2026, targeting a 2027 debut. Former Hillhouse Ventures partner Yan Wentao has come on board as CFO — a hire that only makes sense if a public filing is months, not years, away.

A STAR Market listing at a $75 billion pre-money valuation would make DeepSeek one of the most valuable IPOs in Chinese tech history. For context, that's roughly the market cap of Xiaomi at its 2018 IPO, and it would immediately place DeepSeek among the top five most valuable AI companies globally, alongside OpenAI, Anthropic, and xAI.

Why it matters

The $15 billion number is large, but the more interesting figure is the doubling — from 50 billion to 100 billion yuan mid-negotiation. When a company can double its fundraising target because demand outstrips supply, it tells you the investor consensus has shifted. DeepSeek is no longer seen as a cheap also-ran to OpenAI. It's now viewed as the Western-aligned alternative that can actually compete on cost, and investors are paying for that optionality.

The embodied intelligence pivot is the real strategic signal. DeepSeek has already invested 141 million yuan in Unitree Robotics (宇树科技), is poised to lead a seed round in Winverse AI (founded by former Tencent AI scientist Hu Han), and has backed a new company led by Yu Tao, formerly of Tencent's AI Lab "Juewu" team and a humanoid robot foundation model lead at LimX Dynamics. Founder Liang Wenfeng has been meeting personally with embodied-intelligence founders. This isn't a casual exploration — it's a coordinated expansion from cloud-based code and language models into physical-world interaction.

That pivot explains the investor mix. CATL brings battery and energy storage for robotics. Geely brings automotive manufacturing and supply chain. SMIC Poly Source brings domestic chip connections. Perfect World and Shiji Hualong bring game content pipelines and 3D rendering expertise — DeepSeek posted high-paying job listings for game development and Web 3D rendering back in August.

The critical lens

A few things deserve scrutiny. First, this is still based on "people familiar with the matter" — DeepSeek has issued no official statement. Bloomberg and Reuters both cite anonymous sources, and Chinese media outlets are aggregating those reports. The round may be "substantially closed" rather than formally signed.

Second, a $75 billion pre-money valuation for a company that gives away its models for free is aggressive. DeepSeek's revenue is not publicly disclosed, and its entire commercial strategy rests on the premise that ultra-low inference costs will eventually convert free users into paying enterprise customers. That thesis hasn't been proven at scale.

Third, the concentration of Chinese state-linked and strategic capital — SMIC's fund, CATL, Geely — raises questions about how this plays with the U.S. DHS ban on DeepSeek for federal contractors and the broader export-control environment. A STAR Market listing insulates DeepSeek from U.S. capital markets, but it also locks it into a Chinese regulatory framework that may complicate its global ambitions.

What to watch

• Whether DeepSeek officially confirms the round and valuation before filing its IPO prospectus
• Whether the STAR Market filing happens in Q4 2026 or slips into 2027
• How much of the $15 billion goes to embodied intelligence versus data center buildout
• Whether domestic chip adaptation (Huawei Ascend, SMIC connections) produces a custom silicon roadmap
• Whether Unitree or Winverse AI shipments begin in 2027, validating the robotics thesis

The money is the headline. The robot bet is the story.