DeepSeek's latest model upgrade and its initial public offering plans on the STAR Market have captured significant market attention, raising questions about which A-share companies stand to gain from the AI firm's rapid expansion.

The company unveiled its next-generation V4.1 Flash model on September 9, with a scheduled launch around September 10, claiming superior performance across all key metrics compared to its predecessor V4 Pro. Simultaneously, market speculation was confirmed that DeepSeek has engaged CITIC Securities for its STAR Market IPO preparations, with due diligence already underway.

Upgrade and Price Cut Shake the Industry

DeepSeek announced via its open platform that the V4.1 Flash model would launch around 12:00 Beijing time on September 10. According to internal and external testing, the new model outperforms the flagship V4 Pro in performance, cost, speed, and total processing time. The V4.1 Flash employs an entirely new architectural design with native multimodal support, and beta testers report generation speeds of approximately 284 tokens per second, nearly triple the 97 tokens per second achieved by V4.

Pricing adjustments accompanied the launch, with idle-time cache-hit input costs dropping to just 0.02 yuan per million tokens and output at 4 yuan, while peak-hour rates double. Cache-hit input prices have fallen by 60%. Notably, until the V4.1 Pro release, all API requests directed to V4 Pro will be automatically routed to the V4.1 Flash at lower rates, allowing users to benefit from improved performance and reduced costs without any code changes. This strategic U-turn is particularly striking given that less than a month ago, the V4 Pro launch saw peak-hour cache-miss input prices surge by up to 11 times.

CITIC Securities Initiates Due Diligence

Reuters first reported on September 8 that DeepSeek had engaged CITIC Securities to prepare for a STAR Market listing on the Shanghai Stock Exchange, potentially launching the IPO process within the year. The next day, 21st Century Business Herald confirmed that due diligence had commenced, though a formal listing guidance agreement has yet to be signed.

According to The Information, citing informed sources with unaudited data, DeepSeek generated approximately 475 million yuan in revenue during the first seven months of 2026, nearly ten times its full-year 2025 figure. The company reported a comprehensive gross margin of 44.6%, with its API business achieving an impressive 82.9% margin. However, the same period saw a net loss of 715 million yuan, alongside computing infrastructure expenditures of approximately 11 billion yuan, underscoring the company's aggressive expansion strategy.

Who Benefits in the A-Share Market?

DeepSeek's technological advancements and IPO preparations are illuminating the entire Chinese AI supply chain. Upstream computing infrastructure stands as the most direct beneficiary: Hygon Information Technology and Cambricon Technologies have completed model chip adaptations, while Inspur Information and Sugon provide AI servers and liquid cooling solutions, and Zhongji Innolight supplies optical modules. On the downstream application side, Unisplendour has launched a DeepSeek-enabled all-in-one device, and China Telecom's Tianyi Cloud has completed domestic deployment.

It should be noted that most of these are technical adaptations and product collaborations rather than confirmed large-scale orders. From an investment perspective, while computing infrastructure presents potential order opportunities, equity-linked targets such as Jacom and By-Health hold only minuscule stakes—ranging from ten-thousandths to thousandths of a percent—making them more of a theme-based valuation re-rating play. Following Monday's batch limit-up moves on September 7, technology stocks have shown renewed divergence this week, yet PCB concepts remain active, with storage and HBM sectors taking the lead on September 10.

Policy Window and Rational Expectations

DeepSeek's path to capitalization reflects the surging valuations across China's AI landscape. In June, the company completed its first funding round, raising over 50 billion yuan at a post-investment valuation of nearly 400 billion yuan, setting a domestic record for AI single-round financing. Founder Liang Wenfeng personally contributed approximately 20 billion yuan, while Tencent invested around 10 billion yuan. In July, the company initiated a second round with a pre-investment valuation of approximately 500 billion yuan.

The opening of the listing window follows regulatory reforms: in June, the Shanghai Stock Exchange issued guidelines extending the fifth set of STAR Market listing criteria to AI large-model companies, which do not mandate profitability. However, the pace of the listing process requires measured expectations. Senior investment banker Wang Jiyue noted that based on the standard counseling filing period of no less than three months, DeepSeek would likely not be able to submit its IPO application materials until the second quarter of next year. As of September 9, no listing counseling filing information for DeepSeek appeared on the Zhejiang Securities Regulatory Bureau's official website, indicating the IPO remains in early-stage preparation with inherent uncertainties.