LLMs

DeepSeek: China’s AI Star Plans Multi-Billion Fundraising and IPO

co-written by eustella15. July 2026, 09:02

Just weeks after its first external funding round, DeepSeek is doubling down: The Hangzhou-based AI company has begun preparations for an initial public offering and could file the paperwork as early as this year — with a listing on a mainland exchange to follow in 2027, Bloomberg reports, citing people familiar with the matter. According to the report, the startup is already in talks with accounting firms and banks. The financial reports required for the IPO application are expected to be completed by the end of December. According to The Information, a listing in Shanghai is under discussion.

In parallel, talks about fresh capital are underway — and this is where the reports diverge. The Information reports that DeepSeek, buoyed by strong revenue growth, is targeting a second funding round of 50 billion yuan (around $7.4 billion), which would value the company at roughly $74 billion. Bloomberg, on the other hand, speaks of at least 10 billion yuan (approx. $1.5 billion) at a pre-money valuation of at least 480 billion yuan (around $71 billion) — with the final amount potentially coming in significantly higher depending on investor interest. In either scenario, it would be a hefty jump in valuation: It was only at the end of May that DeepSeek closed its very first external round, raising $7 billion at a valuation of $52 billion, as reported by the Financial Times.

Zhipu AI and MiniMax Already Listed

The high capital requirements are driven by infrastructure expansion: According to the FT, the rapid succession of funding rounds is linked to the anticipated capital needs for the company’s own data centers and additional AI chips — with the development of AI agents in particular driving up demand for computing power significantly. In June, DeepSeek posted job openings for its own data center team and plans to build a gigawatt-scale campus; according to Reuters, the company has also been working on its own AI chips for about a year. On the business side, momentum appears to be building: Annualized revenue is approaching the $500 million mark, according to The Information.

It would not be the first Chinese AI company to go public. Zhipu AI, maker of the GLM models, has been listed on the Hong Kong stock exchange since the beginning of the year and currently commands a valuation of a whopping $93 billion. MiniMax, also publicly traded, is valued at around $10 billion.

The Chinese State Is Already at the Table

The composition of DeepSeek’s existing investors is notable. Alongside Tencent, the $7 billion round included battery maker CATL, JD.com, NetEase, IDG Capital — and the National Artificial Intelligence Industry Investment Fund, Beijing’s state-backed AI investment fund. This fund is one of the state-backed vehicles the Chinese government has deployed to advance AI as a national strategic priority. The makeup of its backers thus raises the question of where DeepSeek sits on the spectrum between private tech company and strategic state asset — a question an IPO would force into the open. Founder Liang Wenfeng himself put around $3 billion of his own money into the first round; his net worth is now estimated at roughly $36 billion.

Liang has signaled to potential investors that DeepSeek will continue to prioritize fundamental AI research over short-term commercialization and stick to open-source models. The timing of the IPO plans fits into a broader trend: OpenAI and Anthropic are also reported to have filed confidential IPO documents in 2026, and SpaceX already pulled off a record listing in June.

DeepSeek Strong on Open Weights, Clearly Behind the US Frontier

So how do DeepSeek’s models currently stack up? Independent benchmarking service Artificial Analysis paints a two-sided picture. The top of the Intelligence Index (as of July 14) is dominated by proprietary US models: Anthropic’s Claude Fable 5 leads with 59.9 points, ahead of OpenAI’s GPT-5.6 Sol (58.9) and Claude Opus 4.8 (55.7). DeepSeek’s current top model, V4 Pro (Reasoning, Max Effort), scores 44 points, placing it well behind the US frontier models — at number 15 in the overall ranking.

Within the open-weight competition, too, DeepSeek is no longer unchallenged: China’s GLM-5.2 from Z.AI, at 51.1 points, now ranks ahead of all DeepSeek models, while MiniMax M3 sits on par with V4 Pro at 44.4 points. A year ago, DeepSeek’s R1 models were still the leading open-weight reasoning models in the index. On agentic tasks, however, DeepSeek holds its ground: On the GDPval-AA benchmark, which measures real-world work tasks, V4 Pro leads the open-weight models with 1,554 points — ahead of Kimi K2.6, GLM-5.1, and MiniMax-M2.7.

The strongest argument remains the price: On the intelligence-vs-cost curve, V4 Pro stands out at $0.02 per task — leading proprietary models cost 20 to 45 times as much; GPT-5.5 comes in at $1.04 per task, Claude Fable 5 at $2.75. However, Artificial Analysis also clearly documents one weakness: V4 Pro and V4 Flash exhibit a very high hallucination rate — when the models don’t know an answer, they almost always respond anyway.