The AI and capital markets were abuzz in July 2026 with news that DeepSeek, a leading domestic large language model unicorn, has fully initiated preparations for an A-share IPO, planning to submit an application to list on the STAR Market within the year, with the goal of completing the listing in 2027.

Concurrently, the company is advancing a second round of financing with a pre-money valuation of approximately $71 billion (around RMB 480 billion).

From a rumored valuation of $10 billion in April to a post-money valuation of $52 billion after its first funding round in June, and now to a pre-money valuation of $71 billion in July, DeepSeek's valuation has multiplied roughly sevenfold in less than three months.

Once a company that was "financially self-sufficient and not seeking funding," it has now conducted two intensive fundraising rounds and is rushing towards an IPO in under two months.

This sharp turn reflects the reality that the "hundred-model war" in the AI industry has entered a critical phase—as computing power becomes the new oil and capital the new ammunition, the former "lone ranger" has finally decided to join the capital frenzy.

The Capital Surge: From "No Fundraising" to a 7x Valuation Jump in Three Months

DeepSeek's capitalization process epitomizes the soaring valuations in China's AI sector.

In late May, DeepSeek completed its first external funding round since inception, raising over RMB 50 billion, setting a record for a single funding round by a domestic large model company.

The investor lineup was impressive: founder Liang Wenfeng personally contributed RMB 20 billion as the largest single investor; Tencent invested RMB 10 billion; the CATL ecosystem invested RMB 5 billion; JD.com, NetEase, and IDG Capital each invested RMB 3 billion; the National Artificial Intelligence Industry Investment Fund also participated through a subsidiary.

The post-money valuation was approximately $52 billion (around RMB 350 billion).

However, just six weeks after closing the first round, DeepSeek swiftly began discussions for a second funding round.

According to reports, the company has started engaging with new investors, targeting a pre-money valuation of about $71 billion (around RMB 480 billion), a roughly 37% increase over the post-money valuation of the first round, with plans to raise at least RMB 10 billion.

Is DeepSeek so "cash-strapped" that it requires such a rapid fundraising pace? The answer is far more complex.

An analysis suggests that fundraising is normal, not primarily due to a lack of funds, but more crucially to retain talent.

Without a market-recognized valuation, employee equity lacks a concrete value representation.

Over the past year, at least five core R&D members at DeepSeek have reportedly left, including a key R1 researcher who joined ByteDance's Seed team.

A more fundamental driver is computing power.

As the AI industry shifts from training-driven to inference-driven, with agents requiring continuous reasoning and long contexts demanding more memory, nearly every new AI direction voraciously consumes computing resources.

For DeepSeek to accelerate its infrastructure expansion, it must leverage capital to lock in computing resources in advance.

Reports indicate that in early April, after seeing breakthroughs in Anthropic's models, Liang Wenfeng further recognized the importance of massive computing power investment—keeping up with competition at that scale requires immense investment in compute.

The company's recent moves clearly show where the funds are directed: building its own GW-scale data center—DeepSeek has begun recruiting IDC design and planning engineers, planning to construct an ultra-large-scale AI computing center; developing its own AI inference chips—reports suggest the project started about a year ago and is now engaging with chip design companies, foundries, and memory suppliers; and massive hiring expansion—after the first funding round, DeepSeek launched its largest recruitment drive since founding, planning to at least double the size of all departments.

Furthermore, the "oversubscribed" nature of the first round is a significant factor.

Investment interest in DeepSeek reportedly exceeded RMB 100 billion, while the first round closed at over RMB 50 billion, meaning at least another RMB 50 billion in capital sought entry.

Investment institutions that missed the first round are now actively vying for a stake in the new round.

Regarding the listing venue, DeepSeek has forgone the Hong Kong route, firmly choosing the domestic STAR Market.

In mid-June, the Shanghai Stock Exchange announced at a financial forum the expansion of the STAR Market's fifth set of listing criteria to cover the artificial intelligence field.

The core feature of this standard is that it does not require profitability or even significant revenue, focusing instead on technological strength.

The exchange subsequently issued specific guidelines detailing eight refined provisions for business scope, technological advantages, staged achievements, industry standing, and market potential.

This regulatory framework opens a green channel for hard-tech AI companies still in the investment and loss-making phase to go public.

The timing for the IPO also has clear reference points.

Other AI companies listed in Hong Kong earlier this year saw their share prices and market capitalizations surge significantly post-listing.

Stimulated by news of DeepSeek's IPO plans, shares of several AI-related companies in Hong Kong rallied strongly in mid-July.

This robust market performance undoubtedly provides a powerful reference for DeepSeek's listing ambitions.

Technological Foundation and Commercial Validation: The Logic Behind a Trillion Valuation

If capital is the "booster" for DeepSeek's IPO sprint, then technological prowess and commercial progress are the "ballast" supporting it all.

DeepSeek originated from a leading domestic quantitative hedge fund, operating independently since July 2023.

Initially, it relied entirely on stable cash flow from its parent company's substantial assets under management to fund R&D.

The parent fund reportedly delivered strong returns in 2025, generating significant revenue, with the founder holding a majority stake and receiving substantial annual dividends, providing a continuous stream of internal funding for the large model's intensive computing and talent investments.

This unique origin distinguishes DeepSeek from most AI startups dependent on external funding.

Technologically, DeepSeek adopts a Mixture-of-Experts (MoE) architecture, positioning itself for low-cost, high-performance applications.

It has iterated through a full product series covering general dialogue, code generation, mathematical reasoning, multimodal interaction, and lightweight industrial deployment.

Among these, the training cost for its model, positioned as a competitor to leading international models, was reportedly significantly lower than that of overseas rivals.

In April 2026, DeepSeek released a preview of its latest model and made it open-source, featuring a context length in the millions and achieving leading levels in agent capabilities, general world knowledge coverage, and complex reasoning performance within the domestic and open-source model landscape.

According to data from an AI model platform, one of DeepSeek's models has consistently ranked at the top globally for API calls since its launch.

A report on open-source AI also noted that a DeepSeek model topped global monthly usage metrics.

On the commercialization front, DeepSeek has established a three-tier business system: a lightweight AI assistant for consumers (C端), private custom deployment solutions for government and enterprises (B端), and a foundational layer offering paid API calls to global developers via its open-source platform.

Citing sources, reports indicate DeepSeek's recent annualized revenue has reached $4 to $5 billion (approximately RMB 27 to 34 billion), primarily from cloud API call services.

This figure represents a doubling compared to forecasts from early 2025.

More noteworthy is its profitability.

Although DeepSeek's model calling prices are reportedly much lower than those of leading international competitors, the gross margin for the API business of its latest flagship model remains above 50%.

This suggests the company's low-price strategy is not a short-term sacrifice of profits but a sustainable business model enabled by efficient underlying technology.

Internationally, DeepSeek's competitiveness is gaining increasing recognition.

Models recently released by Chinese companies, including DeepSeek, are widely considered highly competitive compared to top-tier American systems.

Instances include a small San Francisco-based AI startup reportedly switching 100% of its traffic from a US provider to DeepSeek; a global ride-hailing giant reassessing its model choices after exhausting its annual AI budget; and a major tech company quietly including DeepSeek in the evaluation list for its AI assistant's underlying engine.

In terms of market share among enterprises for a recent month, DeepSeek reportedly led with 19%.

Based on an annualized revenue of $5 billion, DeepSeek's targeted valuation of around RMB 500 billion represents approximately 148 times its annualized revenue.

This multiple is significantly higher than that of many AI startups but reflects strong market expectations for its future growth potential.

However, DeepSeek's path to the STAR Market is not without challenges.

Once listed, pressure from public market investors for financial returns will force DeepSeek to accelerate its commercialization process.

The transition from "focusing on cutting-edge research and development" to "opening its doors to public investors," and from initially having "no plans to commercialize the model" to facing performance pressure from being a listed entity, presents DeepSeek not only with a test from the capital markets but also a deeper challenge of transforming from a research-driven "lone ranger" into a commercially driven public company.

How to translate top-tier coding capabilities into sustainable enterprise revenue, and how to maintain technological leadership while meeting public market expectations for growth, are critical questions the DeepSeek team must answer.

Nevertheless, when DeepSeek—this Chinese AI frontrunner emerging from a quantitative hedge fund—eventually stands at the bell-ringing ceremony on the STAR Market, it will represent not just the IPO of a single company, but a milestone signal for China's AI industry moving from technological breakthroughs towards a complete capital cycle.