DeepSeek restarts 500B yuan raise as China AI funding race intensifies

DeepSeek restarted its second funding round, seeking 500 billion yuan at a pre-money valuation of 500 billion yuan, a 43 percent increase from its first round, as China's AI model funding race intensifies.

Multiple dealmakers told Caijing that signing is targeted for late August, with both first-round investors considering follow-on stakes and previously waitlisted funds seeking entry.

The first round, which opened in April and closed in June, raised 500 billion yuan at a valuation exceeding 350 billion yuan — the largest first-round raise in Chinese AI history. Investor interest in the first round exceeded 100 billion yuan, leaving at least 500 billion yuan in unmet demand. The second round had opened in mid-July but was paused at the end of the month after founder Liang Wenfeng took issue with leaked investor meeting transcripts circulating online.

The two rounds combined would bring DeepSeek more than 100 billion yuan in fresh capital, dwarfing competitors. Moonshot AI, which completed a Series F at roughly 210 billion yuan ($31.5 billion) in late July, immediately launched a Pre-IPO round at a 500 billion yuan ($50 billion) valuation, with filing expected this year.

Investor demand outstrips supply

The first round drew interest exceeding 100 billion yuan, with investors including the National AI Industry Investment Fund, Tencent, CATL and Puquan Capital, NetEase, JD.com, Lishi Capital, IDG Capital, Zhengxingu Investment, and Shixiang Capital. The second round's waitlist includes both first-round "losers" — funds that missed out on initial allocation — and existing VCs and industrial capital considering increases.

Some investors who had been in active talks said they had not yet received word of the restart, with the channel still on hold. Dealers said DeepSeek and its negotiating investors want the round to proceed quietly this time, a departure from the first round's high-profile close.

The pause in late July was driven by Liang Wenfeng's dissatisfaction with widely circulated commentary about an alleged investor meeting transcript that leaked online. Caijing's reporting did not confirm the authenticity of the document, and DeepSeek did not respond to requests for comment.

Model release sharpens competitive edge

DeepSeek's funding momentum coincides with the July 31 release of V4-Flash, the formal version of its flagship model. The model retains the same architecture and parameter count as its predecessor but strengthens post-training, boosting agent capabilities. On Artificial Analysis's Intelligence Index, V4-Flash scores 50, second only to Zhipu's GLM-5.2 at 51 among Chinese models.

The model continues DeepSeek's low-cost strategy: output pricing at $0.28 per million tokens, compared with GLM-5.2's $4.29. On OpenRouter, V4-Flash consumed 7.1 trillion tokens this week, ranking first globally. The pricing pressure has pushed OpenAI to cut GPT-5.6 API prices, with its entry-level Luna model down 80 percent.

The V4-Pro formal version has not yet been released, with the public testing date still unannounced. A source close to DeepSeek told Caijing in late July that the company faced significant pressure training the new model, with the original mid-July release date slipping to July 31.

The funding round sets a valuation benchmark for Chinese AI model companies. DeepSeek's 500 billion yuan pre-money valuation, achieved in under a year since its first raise, reflects investor conviction that model capability and efficiency — not cash flow — determine pricing in this sector. "Pricing for large models is essentially an option, not a cash-flow-based financial model," one dealmaker told Caijing. The competitive pressure extends to Moonshot AI, whose $50 billion Pre-IPO valuation now faces direct comparison with DeepSeek's trajectory.

This article is for informational purposes only and does not constitute investment advice.