DeepSeek has raised 105.5 billion yuan, roughly $15.8 billion, in a second funding round led by Tencent Holdings and battery giant CATL, according to BigGo Finance. Nearly 30 investors took part, pushing the Hangzhou-based AI company's valuation to roughly 615.5 billion yuan, or $84.6 billion, and bringing its combined haul across two rounds to about 136.5 billion yuan, close to $16.3 billion.
Tencent put in 10 billion yuan in this round, matching its stake from the first round earlier this year. CATL, led by chairman Zeng Yuqun, doubled its commitment to 10 billion yuan after contributing 5 billion yuan the first time. Monolith Management reportedly wrote the single biggest check of the round. State-owned entities, including China Mobile, China Telecom, CICC and Citic Securities, accounted for about 27% of the capital, up sharply from just 2% in the first round.
Why is state money suddenly pouring in?
DeepSeek rattled global AI markets in early 2025 by showing it could train competitive large language models on a fraction of the compute budget of US rivals. That reputation now has a price tag. Investors told BigGo Finance that DeepSeek doubled its fundraising target mid-raise as black-market prices for Nvidia B300 servers climbed, a sign that chip scarcity, not model quality, has become the binding constraint on China's AI leaders.
Building a war chest ahead of possible supply shocks is the explicit goal here. With US export controls still limiting legal access to top-tier Nvidia silicon, Chinese AI labs are stockpiling capital so they can buy whatever compute becomes available, wherever it surfaces, before rivals do. That dynamic also explains the jump in state participation: Beijing has an interest in making sure its most capable model developer doesn't run short of chips at a critical moment in the AI race.
What strings come attached?
Founder Liang Wenfeng kept tight control over the company despite the inflow of outside capital. Most investors in this round accepted five-year lock-ups on their shares and gave up voting rights, an unusual structural concession for a round this large. Liang personally added 2 billion yuan on top of the institutional money, reinforcing that DeepSeek intends to remain founder-directed rather than board-directed as it scales.
The corporate roster reads like a cross-section of Chinese industry rather than a typical venture syndicate: CXMT and SMIC from chipmaking, vivo and Xiaomi-adjacent players from consumer hardware, Meituan and Trip.com from consumer internet, Geely and Midea from manufacturing, and Luxshare Precision from electronics assembly. Many of these companies are themselves building AI into physical products, from EVs to appliances, giving them a direct interest in DeepSeek's model roadmap.
What should the robotics and physical-AI industry watch next?
DeepSeek has stayed focused on language and reasoning models rather than robotics, but its low-cost training techniques have already influenced how embodied-AI labs approach model efficiency. A better-funded DeepSeek with deeper compute reserves could accelerate open releases that robotics startups lean on for perception and planning stacks.
The bigger signal is structural: Chinese AI companies are now raising war chests specifically to hedge against chip shortages, not just to fund research. Expect more of China's physical-AI and robotics players to follow the same playbook, locking in compute capacity years ahead of deployment rather than buying as needed.
