DeepSeek is reportedly preparing for a listing on the Science and Technology Innovation Board, a move that could establish a new valuation benchmark for general-purpose AI models in China's A-share market.

Market sources indicate that DeepSeek has engaged CITIC Securities to handle its IPO preparations on the STAR Market. The key point of interest is not just the listing itself, but the choice of the STAR Market over the Hong Kong exchange, where peers like Zhipu and MiniMax have already completed listings. No pure-play general AI model company has yet listed on the A-share market, meaning a successful DeepSeek IPO would create an independent, localized pricing framework for such companies and could become a reference anchor for Zhipu and MiniMax should they consider returning to A-shares in the future.

The significance extends to revenue scale, commercialization capabilities, model competitiveness, and the valuation multiples the market ultimately assigns. Any of these metrics could serve as an important reference for future A-share AI model companies. Therefore, DeepSeek's true objective may not only be claiming the title of "first A-share AI model stock" but also securing pricing power for the entire sector.

The STAR Market positioning battle

Several factors are driving DeepSeek's decision to pursue a STAR Market listing now. The exchange has recently opened a window for AI model companies. On June 17, 2026, the Shanghai Stock Exchange issued guidelines extending the fifth set of listing standards to AI large model enterprises, a framework previously designed primarily for innovative drug companies. This provides a clearer path for model companies that have not yet established mature profitability but possess technological advantages and commercialization potential.

For DeepSeek, the timing aligns well with its current stage. The company has built a large user base, its model capabilities rank among the top domestically, and its commercialization strategy is evolving from free and low-cost APIs toward a more refined pricing structure. The API price adjustments made since August signal deeper commercialization efforts. Rather than waiting for the industry to develop a mature listing model, DeepSeek is positioning itself among the first batch of applicants.

Additionally, the A-share market lacks a general AI model listing. Hong Kong already hosts Zhipu and MiniMax, providing the first set of pricing references for the sector, but no comparable benchmark exists in A-shares. If DeepSeek lists on the STAR Market first, it would become the first true general AI model company in the A-share market and potentially set the initial valuation yardstick for the industry.

The company also has substantial momentum from user adoption and capital markets. According to QuestMobile data, DeepSeek's monthly active users reached approximately 130 million in June 2026, ranking among the top three AI-native applications in China. This scale provides a solid foundation for model iteration, brand influence, and commercialization conversion.

On the capital front, DeepSeek completed its first external financing round in April 2026, with closing in June, raising 50 billion yuan at a post-investment valuation exceeding 350 billion yuan. The round included 20 billion yuan from founder Liang Wenfeng personally, 10 billion yuan from Tencent, 5 billion yuan from CATL, with participation from NetEase, JD.com, and IDG Capital. A second round, still underway, is raising approximately 50 billion yuan at a pre-investment valuation of 500 billion yuan. If successful, cumulative fundraising would exceed 100 billion yuan.

However, as competition intensifies in the domestic AI model space, talent acquisition becomes increasingly critical. Compensation is only one factor; equity incentives and exit expectations are equally important for core researchers. Meanwhile, growing external funding creates clearer return expectations for early shareholders and strategic investors, making an IPO a natural liquidity option.

The regulatory window has opened, the A-share market lacks a local AI model pricing benchmark, and DeepSeek has accumulated sufficient users, capital, and technological assets. The answer to why the STAR Market and why now is becoming clear.

If DeepSeek becomes the first general AI model company on the A-share market, its impact will extend beyond its own fundraising trajectory. It could fundamentally repricing the entire industry.

Who faces repricing?

An IPO carries significance beyond a single company's fate. Previously, valuations for AI model companies were determined primarily in the private market, based on model capabilities, user scale, team strength, market potential, and commercialization projections. After listing, however, the dynamics shift. DeepSeek's valuation will no longer be negotiated but established daily through trading activity, naturally making it a comparison benchmark for other AI companies.

Kimi, reportedly also preparing for an IPO, would be among the most directly affected. Given similarities in model capabilities, open-source attributes, and API-based business models, DeepSeek and Kimi are highly comparable. Their current private market valuations sit in similar ranges: DeepSeek's second round pre-investment valuation is approximately 500 billion yuan (US$75 billion), while Kimi's parent company Moonshot AI has a latest round valuation of about US$50 billion.

Should DeepSeek list first, investors evaluating Kimi's IPO would likely compare the two directly. The first-day trading price of DeepSeek may not be fully representative, but the valuation center established over the following 6 to 12 months, after speculative premiums fade, would offer a more meaningful comparison based on revenue, growth, and profitability.

For Zhipu and MiniMax, already listed in Hong Kong, the impact is already reflected in current valuation differences. Market data shows DeepSeek generated approximately 475 million yuan in revenue during the first seven months of 2026, with projected full-year revenue of about 814 million yuan. Based on the 500 billion yuan pre-investment valuation, this corresponds to a price-to-sales ratio of roughly 614 times. Zhipu reported 954 million yuan in first-half 2026 revenue, implying full-year revenue of approximately 1.9 billion yuan. With a market capitalization of about 367.6 billion yuan as of September 9, its P/S ratio is around 193 times. MiniMax generated approximately 794 million yuan in first-half 2026 revenue, implying full-year revenue of about 1.6 billion yuan, with a market cap of roughly 96.8 billion yuan as of September 9, corresponding to a P/S ratio of approximately 61 times.

It is important to note that these comparisons involve different valuation calibers. Direct conclusions about relative value would be premature, as the three companies have distinct revenue structures, business models, and growth expectations. Nevertheless, the numbers indicate that the market is already pricing AI model companies using different approaches.

The true importance of DeepSeek's IPO lies in prompting the market to answer a more specific question: which types of AI models are worth more? Compared to MiniMax's "API + C-end application" model, Zhipu's business model shares greater similarity with DeepSeek, given its open-source approach and the rising share of MaaS platform API revenue, which reached 86.5% of first-half 2026 revenue. As both rely heavily on API calls, the market may begin scrutinizing revenue growth rates, customer retention, API call volumes, gross margins, and unit economics. The impact on valuations would be more direct.

DeepSeek does not need to outperform peers on every metric. Once trading begins, comparable data emerge, and each difference translates into valuation gaps.

DeepSeek itself faces repricing challenges

An IPO addresses financing needs but does not solve the business model question. After listing, investors will evaluate not just model performance metrics but concrete financial indicators.

The first question is sustainability of growth. Despite its large user base, converting ecosystem engagement into stable revenue is a separate challenge. API customers can migrate to competing models at low cost, with price, performance, and stability all influencing call volumes. The Information data indicates that 2026 first-seven-month revenue of 475 million yuan is ten times higher than 2025 full-year revenue. Whether full-year 2026 revenue can sustain this pace remains to be seen.

The second question is profitability. The company recorded a net loss of 715 million yuan in the first seven months of 2026. At current investment intensity, full-year losses could widen further. The industry is shifting from training-driven to inference-driven dynamics, with agents, long-context processing, and real-time interactions increasing computational demands on the inference side. DeepSeek is building a data center in Ulanqab and advancing domestic chip substitution to reduce long-term costs, but substantial investment will continue until the facility achieves full capacity.

Meanwhile, DeepSeek adjusted its API pricing on August 17. On September 10 at 12:00, the flash series API underwent another adjustment: input prices were reduced, while output prices remained unchanged (idle-time input adjusted to 0.02 yuan, cache-miss input at 1 yuan, output maintained at 4 yuan). The pricing structure suggests a strategy of "lower input prices to protect the ecosystem, maintain output prices to protect margins." Whether this tiered approach will prove effective, whether price cuts will drive sufficient call volume growth, and whether customers will remain after increases, all require subsequent data to validate.

The secondary market will not wait for gradual experimentation. With quarterly financial reports and earnings calls, if successive quarters reveal slowing revenue growth, expanding losses, or declining gross margins, the 500 billion yuan pre-investment private market valuation could face significant recalibration, even if model capabilities remain leading.

This article is for informational purposes only and does not constitute investment advice. Readers should make investment decisions with caution.