Author｜Sun Ningyu

Editor｜Jiang Zhou

On July 15, news outlets such as Shenzhen Economic Daily and Cailian Press reported that DeepSeek, a prominent domestic developer of large-scale AI models, has officially begun preparations for an A-share Initial Public Offering (IPO). The company aims to submit its application materials within the year, with a target to go public by 2027.

DeepSeek recently raised over RMB 50 billion in its initial funding round, achieving a pre-money valuation of USD 71 billion (approximately RMB 480 billion). Key investors include Tencent, CATL, and the National AI Industry Fund.

With the anticipated listings of Zhipu and MiniMax on the Hong Kong Stock Exchange boosting the AI sector, there has been a surge in interest, prompting a closer look at DeepSeek's capital strategy, technological capabilities, and potential listing risks.

Capital Surge

DeepSeek's capitalization path reflects the rapid valuation growth in China's AI sector.

From an estimated valuation of USD 10 billion in April 2026, its post-money valuation soared to USD 52 billion after its first funding round in June and further climbed to USD 71 billion during second-round negotiations in July, marking a sevenfold increase in just three months.

The initial funding round exceeded RMB 50 billion, with founder Liang Wenfeng contributing RMB 20 billion as the largest investor. Tencent invested RMB 10 billion, CATL RMB 5 billion, while JD.com, NetEase, and IDG Capital each invested RMB 3 billion. The National AI Industry Fund also participated, creating a shareholder structure that combines industrial, financial, and policy-backed capital.

To maintain control over its technology, DeepSeek adopted a differentiated investment framework. Except for the national AI fund, other industrial investors receive only dividend rights without board seats or voting privileges. All external shares are subject to a five-year lock-up period. Through this funding and structural design, Liang Wenfeng retains nearly 78% control, reflecting institutional investors' focus on long-term technological execution rather than short-term financial gains.

For its listing venue, DeepSeek bypassed Hong Kong and selected Shanghai's STAR Market, taking advantage of the new AI-specific fifth listing criteria introduced by the Shanghai Stock Exchange (SSE) on June 17. This framework waives traditional profitability requirements, instead focusing on core technological strengths, scalable product deployment, and regulatory compliance, thus providing a fast track for loss-making AI technology firms. The company has initiated auditor due diligence to prepare financial documents by year-end.

Market sentiment was further bolstered by the wealth effects from Zhipu and MiniMax's successful Hong Kong listings earlier this year.

Zhipu's stock peaked at HKD 2,980, reaching a market cap of HKD 1.27 trillion, while MiniMax surged 109% on its debut, hitting a peak valuation of HKD 410 billion.

Spurred by news of DeepSeek's IPO, AI stocks rallied on July 15: MiniMax gained over 13%, Zhipu rose 6%, Tencent Holdings jumped nearly 4%, and Alibaba closed up over 2%, as capital intensified bets on leading independent large-scale model developers.

Technological Foundation

While capital fuels DeepSeek's IPO ambitions, technology remains its bedrock.

Spinning off from domestic quantitative trading powerhouse High-Flyer Quant in July 2023, DeepSeek initially relied on its parent's stable cash flow from RMB 70 billion in assets under management to fund R&D.

By 2025, High-Flyer Quant achieved a 56.55% return rate and RMB 8.6 billion in revenue. Founder Liang Wenfeng, holding 85% equity, secured billions in annual dividends, providing self-sustaining capital for high-intensity AI R&D—a stark contrast to startups reliant on external funding.

Technologically, DeepSeek prioritizes the Mixture of Experts (MoE) architecture, emphasizing low-cost, high-performance solutions. It has developed a comprehensive product suite—DeepSeek-V2, V3, R1, and V4—covering general dialogue, code generation, mathematical reasoning, multimodal interactions, and industrial lightweight deployment.

Benchmarked against OpenAI's o1, DeepSeek-R1's training cost was just USD 5.576 million, significantly lower than its overseas counterparts. Its open-source version gained 300% more GitHub stars over the years, earning the nickname "AI's Pinduoduo" for its cost-efficiency and global developer adoption.

Its commercialization strategy consists of three tiers: lightweight AI assistants for consumers (C-end), customized private deployments for governments and enterprises (B-end), and a global developer ecosystem via open-source API monetization.

According to industry insiders, foreign media reported DeepSeek's annualized revenue at USD 400–500 million, primarily from V4 series API subscriptions, which boast over 50% gross margins—its sole mature revenue stream. However, revenue concentration and lack of major clients have fueled valuation controversies.

To address challenges in autonomous innovation, DeepSeek plans to allocate substantial funding to GW-scale self-built AI computing centers and proprietary AI inference chip development, pursuing a dual strategy of "overseas high-end GPU procurement + domestic chip adaptation." The V4 model now supports Huawei chips, reducing reliance on overseas supply chains and fortifying computational autonomy.

Against the industry backdrop, China's general-purpose large model sector is transitioning from parameter races to commercialization. Digital transformation in government and enterprises, smart manufacturing upgrades, and AI-driven internet product enhancements are driving massive demand.

Industry research forecasts a 45%+ Compound Annual Growth Rate (CAGR) for China's large model market from 2026–2030, with model API calls, industry-specific solutions, and computational leasing as key revenue drivers.

Currently, fewer than 10 independent Chinese firms possess full self-developed capabilities and scalable commercialization. DeepSeek leads with its rapid iteration and cost advantages. Tencent and CATL's investments aim to unlock consumer internet and new energy industrial applications, creating synergistic ecosystems.

Critical Crossroads

Despite favorable capital, technological, and policy conditions, DeepSeek's ability to justify its RMB 480 billion valuation post-IPO depends on its post-listing performance.

Currently reliant on fragmented API orders without large-scale government/enterprise projects, management prioritizes R&D over short-term monetization, delaying profitability.

General-purpose models must undergo rigorous algorithm registration, training data copyright verification, and data security audits. The STAR Market's fifth criteria impose stringent disclosure requirements on AI firms, demanding full transparency on algorithm architecture, data sources, and IP ownership—any compliance gaps could derail the IPO.

Additionally, escalating salaries for top AI talent and potential core team attrition risk technical discontinuities.

In summary, DeepSeek exemplifies China's large model capitalization era, leveraging industrial upgrading, capital enthusiasm, and policy easing to craft a compelling listing narrative.

Investors should monitor three key performance indicators (KPIs): API conversion rates, progress in self-built computational infrastructure, and government/enterprise contract volumes. Only by transitioning from R&D to sustained profitability can DeepSeek shed speculative labels and emerge as a truly fortified AI leader.

Its IPO journey will also mark a pivotal inflection point for the entire large model sector, transitioning from capital hype to performance delivery.