NewsFlash Articles Data Fundraising BlockBeats Pro

DeepSeek accelerates toward A-share IPO: has approached CITIC Securities, targeting the STAR Market

Read this article in 12 Minutes

DeepSeek has reportedly engaged CITIC Securities to prepare for its STAR Market IPO, while also advancing new financing to support model R&D, service expansion, and talent incentives.

Original Title: "DeepSeek Accelerates Toward A-Share IPO: Appoints CITIC Securities, Targets STAR Market"
Original Author: 动察Beating

On September 9, 2026, Reuters reported, citing two sources familiar with the matter, that DeepSeek has hired CITIC Securities to prepare for an initial public offering on the Shanghai Stock Exchange's STAR Market, aiming to kick off the IPO process this year. However, the listing timeline, fundraising scale, and issuance valuation have yet to be determined.

At the time of the report, neither DeepSeek nor CITIC Securities immediately responded to requests for comment.

As early as July, Bloomberg reported that DeepSeek was working with accounting firms and investment banking advisors to prepare the financial materials required for listing.

According to the corporate listing process published by the Shanghai Stock Exchange, before an official issuance, a company must complete application document preparation, internal decision-making, and counseling acceptance, followed by exchange review, CSRC registration, and pricing for issuance. The entry of a securities firm marks a key preliminary step; counseling filing, application acceptance, and final listing correspond to different milestones in the process.

Running parallel to the IPO preparations is a substantial round of private fundraising.

In June this year, DeepSeek completed its first external financing round, raising approximately $7.4 billion, with a post-investment valuation exceeding $50 billion. Shortly thereafter, discussions for a new round of financing surfaced. On August 26, the South China Morning Post, citing sources familiar with the matter, reported that the company planned to raise an additional approximately 50 billion yuan, corresponding to a pre-investment valuation of around 500 billion yuan, with expectations at the time to close by the end of August. That projected date does not necessarily mean the financing has been finalized.

There are currently differing figures regarding the valuation for this round. Reuters' September 9 report cited approximately 500 billion yuan, or about $75 billion; a public podcast transcript from the Financial Times published the same day mentioned around $71 billion.

Within just a few months, completing its first external financing round, launching a subsequent round, and advancing toward an IPO—DeepSeek's demand for capital is becoming more concrete.

This easily evokes the low-cost narrative that accompanied its rise to fame. However, to understand today's fundraising, one must first place that widely circulated training cost back into its proper context.

The approximately $5.576 million disclosed in the DeepSeek-V3 technical report is the estimated formal training compute cost, calculated at $2 per H800 GPU hour. The report explicitly states that this figure does not include prior research and ablation experiments on model architecture, algorithms, and data. It measures a specific training process and cannot represent the full cost of operating a frontier model company.

From an operational perspective, efficiency gains can lower the cost of a single training run or inference call, but the company still needs to continuously experiment with new models, maintain live services, allocate resources for growing request volumes, and pre-provision compute capacity for next-generation products. The research team spending money more effectively and the company requiring a larger long-term budget can coexist simultaneously.

Recent job postings have given these needs a more concrete shape.

On September 8, DeepSeek planned to add approximately 150 new positions, focusing on server-side development, Agent elastic computing R&D, and other roles, primarily targeting experienced backend engineers. A company representative explained that as data, machines, training and evaluation tasks, and user requests increase, existing systems need to be upgraded, maintained, or even rewritten.

These roles connect models with real-world services. After a model performs well in testing, someone still needs to ensure developers can reliably call it, and that task execution, resource scheduling, and data processing can withstand greater scale. This portion of engineering investment will continue to grow as services and research activities expand.

Reuters also offered a motive directly tied to going public. Another source said Liang Wenfeng hopes IPO proceeds will help DeepSeek strengthen incentives for core employees and researchers.

This makes the significance of listing easier to grasp. Capital can both purchase compute resources and support compensation and talent incentives. For a company that relies on sustained breakthroughs from a small core team, whether it can retain researchers and recruit experienced engineers will directly affect the continuity of product iteration. A single funding round adds to the balance sheet, while going public could further broaden future financing channels and provide a clearer market pricing benchmark for equity incentives.

Meanwhile, DeepSeek appears to still want to preserve its existing decision-making autonomy. The Financial Times reported on September 9 that the new funding round includes conditions such as a five-year lock-up period and no voting rights, allowing current controllers to retain control. Even with strict terms, investor demand for funding allocations remains strong.

These arrangements show that while the company seeks capital, it is also choosing the way capital enters. Long-term model research requires tolerating failure and reserving budget for directions that cannot yet be monetized. External investors want returns, and the founding team wants to control the pace of research—these two demands ultimately need to be reconciled in specific governance structures. Financing terms can coordinate them, but ongoing information disclosure and shareholder relations after listing will bring new requirements.

The STAR Market emerging as the listing target at this time also has a clear institutional backdrop.

On June 17, 2026, the Shanghai Stock Exchange (SSE) issued special guidelines for AI large model companies seeking to list under the fifth set of listing standards on the STAR Market, supporting high-quality large model enterprises that have not yet generated a certain scale of revenue to proceed with IPOs. The guidelines set requirements around technological advantages, phased achievements, relevant approvals, and market space, and make the launch of products with scaled application a key requirement for phased achievements.

This change provides a clearer filing path for large model companies whose R&D spending outpaces their revenue. It allows the capital market to assess technological capabilities and application progress before a company's commercialization has fully matured. However, which set of listing standards DeepSeek ultimately chooses, and how it proves its compliance, will still depend on its subsequent filing materials.

Once it truly enters the public market, DeepSeek will need to provide investors with a more complete picture of its story.

Model capabilities, developer reputation, and application scale can demonstrate the competitiveness of its products. To understand the company's value, one also needs to see how these advantages translate into revenue, whether customers continue to pay, whether service income can cover computing costs, how fast R&D spending is growing, and how much of that spending can be funded by operating cash flow. Only when these metrics are viewed together can one see what kind of growth the financing is supporting.

The open-source model makes this analysis even more necessary. Widespread adoption of the model can expand technological influence, but usage within the ecosystem does not all convert into revenue for the model developer. In the future prospectus, the most worthwhile section to read closely will be the relationship between technology dissemination, paid services, and sustained R&D.

DeepSeek's IPO push this time presents the process of a model company gradually scaling up its operations. It has already begun arranging longer-term funding sources for continued research, service expansion, and talent competition.

Going forward, the public tutoring and filing documents, along with the business, financial, and governance information they contain, will give the outside world its first opportunity to more fully assess what kind of long-term business the company's technological advantages can support.