DeepSeek reportedly engages CITIC Securities for STAR Market IPO

By Li Kunkun and Li Zhenghao, China Economic News | Beijing

Recent reports indicate that DeepSeek has engaged with CITIC Securities to prepare for an initial public offering (IPO) on the Shanghai Stock Exchange's STAR Market, with plans to initiate relevant procedures this year. However, specific details such as the listing timeline, financing scale, and target valuation remain undetermined.

DeepSeek's IPO preparations appear to be progressing alongside its ongoing new round of financing. The new round targets a valuation of 500 billion yuan ($69 billion), with investors comprising existing shareholders from the first funding round and backup investors who did not secure stakes in the initial round.

IPO Preparations

Industry insiders note that while DeepSeek initially relied on cash flow from High-Flyer Quant and could afford to reject external venture capital, the AGI sector is a typical "capital black hole" requiring continuous investment in computing power, chips, data centers, and talent. Relying solely on internal funds and early-stage financing makes it difficult to sustain long-term competition against OpenAI and Anthropic.

First Funding Round (June)

DeepSeek completed its first funding round in June, with market estimates placing the total raised at 50–51 billion yuan ($6.9–7.0 billion). Key investors include:

• Liang Wenfeng (Founder): Approximately 20 billion yuan — largest single investor
• Tencent: Approximately 10 billion yuan
• CATL system entities (including CATL and Puquan Capital): Approximately 5 billion yuan
• NetEase, JD.com, Monolith Capital, IDG Capital: Approximately 3 billion yuan each
• Zhengxin Valley Investment, Shixiang Technology: Approximately 1.5 billion yuan each

The post-money valuation is estimated at approximately $50 billion.

Timeline and Policy Support

• Late August: Reports surfaced that DeepSeek was preparing for a potential IPO, possibly filing applications as early as the end of 2026, with a listing target of 2027.
• June 17, 2026: The Shanghai Stock Exchange issued guidelines allowing AI large-model companies to apply the fifth set of listing standards on the STAR Market, specifically targeting self-developed large-model enterprises that have not yet achieved stable profitability.
• September 7, 2026: Cui Tianyi, technical lead of the DeepSeek Harness team, posted recruitment information on social media, seeking approximately 150 candidates globally — primarily engineering roles such as server-side development and elastic computing for Agents, rather than deep learning researchers.

Cui explained that large-scale expansion leads to a sharp increase in complexity across data volumes, machine and container counts, training tasks, evaluation tasks, Agent environments, user numbers, and request volumes. The existing backend systems face increasing complexity, necessitating team expansion to upgrade, maintain, and rewrite the backend infrastructure.

Senior financial columnist Ma Jipeng told China Business Journal that although DeepSeek's model versions have been continuously updated and their performance rivals the most advanced US models, its clear advantage lies in lower costs — both in training expenses and customer pricing — making it highly competitive compared to foreign counterparts.

Analysts believe DeepSeek's IPO preparation is closely linked to favorable policy developments. Launching IPO preparations less than three months after the STAR Market guidelines were released demonstrates efficient alignment between the company and industry policies and capital market rules.

The True Test of Commercialization

Industry observers view DeepSeek's IPO preparation as a critical step in transitioning from "technology-driven" to "capital-driven." This signifies entry into a new phase requiring substantial external funding for computing power, R&D, and talent, reflecting a shift in domestic large-model competition from "model capability" to "commercialization ability."

Revenue and Profitability

Listing does not imply immediate profitability. The Shanghai Stock Exchange's June 2026 guidelines support AI companies without scaled revenue using the fifth STAR Market standard. Post-listing, markets will scrutinize practical issues:

• Where does revenue come from?
• Are API calls driven by genuine production needs or subsidized traffic?
• While open-source strategies wield significant influence, high user numbers do not automatically translate to high official revenues.

Recent reports indicate DeepSeek generated approximately 475 million yuan in revenue during the first seven months of 2026, roughly ten times its full-year 2025 revenue. During the same period, gross margin stood at 44.6%, with API business margins reaching 82.9%. These figures are unofficial and subject to future audits or prospectus disclosures.

Key Challenges

Investment banking sources suggest DeepSeek may prioritize complete annual financial statements over the industry-standard Annual Recurring Revenue (ARR) metric for its IPO prospectus, which would anchor its listing valuation. Scaling up and recognizing more revenue are now urgent priorities.

Insiders argue that while DeepSeek's core strength lies in architectural innovations that reduce inference costs, low costs in technical reports differ from high margins in financial statements. R&D trial-and-error, peak computing redundancy, and enterprise services all impact profit and loss accounts.

Future price wars present a dilemma: lower prices expand the market but may reduce per-token profits. Efficiency only becomes a moat when call volumes, repurchase rates, and computing utilization improve simultaneously.

Talent attrition among the core team is increasingly prominent. From late 2025 to early 2026, several key technical staff left DeepSeek. Meanwhile, major tech firms' competition for top AI researchers has driven talent costs to new heights. More attractive equity incentive plans post-IPO may help retain critical personnel.

Broader Industry Impact

Industry experts believe DeepSeek's listing will reshape sector-wide valuation logic:

• Capital concentration: Capital will concentrate among leaders
• Foundation-layer players: Must combine model efficiency with commercialization capabilities
• Application-layer players: Must control industry data and customer access
• Mid-tier companies: Those lacking technological differentiation will feel pressure first

Computing procurement evaluations will change — markets will no longer applaud mere "ten-thousand-card clusters" but focus on GPU utilization, unit inference costs, and real business output.

Talent will be repriced: Merely training models is insufficient. Individuals capable of integrating models, inference systems, and customer needs onto a single profit statement will become scarcer.

Regulatory and Valuation Considerations

Notably, CITIC Securities has entered due diligence, though no formal tutoring agreement has been signed. No listing filing information for DeepSeek is currently available from the Zhejiang CSRC bureau. Given that the tutoring filing process requires at least three months, meeting a year-end submission deadline appears tight.

Market expectations for DeepSeek's valuation are high. With a pre-money primary market valuation of 500 billion yuan and first-half revenue of ~475 million yuan, the price-to-sales ratio exceeds 1,000x. Whether secondary markets will accept this valuation remains uncertain. Speculative investments via multi-layered SPVs have emerged, prompting Liang Wenfeng to personally review shareholder lists to prevent complex structures from hindering the IPO process.

Overall, DeepSeek's listing marks the industry's transition from "proving we can build it" to "answering whether we can generate sustained profits," warranting continued attention to its commercialization data and regulatory progress.

(Edited by Wu Qing; Reviewed by Li Zhenghao; Proofread by Yan Jingning)

Disclaimer: The above content (including images and videos if any) was uploaded and posted by a user of NetEase Hao, a social media platform that provides only information storage services.