DeepSeek Nears US$74 Billion Valuation in Pre-IPO Funding Round

DeepSeek's US$74 Billion Round Sets the Stage for a Star Market Debut

Hangzhou-based DeepSeek is nearing completion of a new funding round that values the Chinese artificial intelligence pioneer at about 500 billion yuan (US$74 billion) before investment, as the company moves closer to a potential listing on Shanghai's Star Market, according to people familiar with the matter. The company is seeking to raise about 50 billion yuan (US$7.4 billion) in a round expected to close before the end of August, one person said. The financing, reported by the South China Morning Post on Wednesday, would cement DeepSeek's position as one of the most valuable private AI companies outside the United States and set the stage for one of the most closely watched Chinese technology debuts of the year.

A Landmark Round With State-Aligned Backers

The investor lineup shows how deeply China's AI build-out now runs through state-linked capital. Incumbents from DeepSeek's last fundraising round, including local venture funds Monolith and Shixiang Capital, are expected to participate alongside Contemporary Amperex Technology (CATL), the world's largest battery maker, according to one source. New investors in talks include CPE, Legend Capital and Stony Creek Capital, a semiconductor-focused private equity firm historically linked to Zhu Yiming, chairman of memory chip giant ChangXin Memory Technologies (CXMT) and founder of chip designer GigaDevice. Funds backed by GigaDevice and Hefei state investment vehicles are also involved, the source added.

The Hefei connection is notable. The Anhui provincial capital has become the anchor of China's memory chip ambitions, hosting CXMT's headquarters, and its state funds have been a fixture in the country's semiconductor financing. For analysts tracking China's AI economy, the round illustrates a broader pattern: strategic capital from battery, chip and provincial-state sources is consolidating around a small number of national champions, even as foreign venture money remains largely walled off from mainland AI frontier companies.

From Low-Cost Disruptor to Price-Setter

DeepSeek shook the global industry in January 2025 with R1, a reasoning model that matched far more expensive Western rivals at a fraction of the cost, triggering a worldwide reassessment of AI spending assumptions. The company's latest move, however, is in the opposite direction. On August 6, DeepSeek announced it would raise prices for its application programming interface services by a "significant" margin, without disclosing the size of the hike. On August 16 it moved to a peak-and-off-peak billing structure under which output tokens from its flagship V4-Pro model cost US$3.96 per million during peak hours, roughly four times the previous flat rate, according to TechTimes.

DeepSeek is not alone. Zhipu AI, Alibaba Group Holding, Tencent Holdings and Baidu have all raised prices this year, reflecting a sector-wide shift away from using discounts to win users and toward monetising increasingly costly AI services. The reversal is a direct consequence of mounting computing and infrastructure costs, as Chinese AI companies search for ways to turn the technology into sustainable profit. For the developers who built DeepSeek's open-weight models into their products, the message is blunt: the era of near-free frontier AI is over.

The Road to Shanghai's Star Market

DeepSeek has begun preparing for a potential initial public offering on the Star Market, Shanghai's Nasdaq-style board, and could file as early as the end of the year, with a public market debut eyed for 2027, two sources said. A listing would place the company at the centre of an extraordinary wave of Chinese technology floats. DRAM maker CXMT raised 66.6 billion yuan (US$9.9 billion) in July in what was the board's biggest listing; CCSH Corporation, the parent of NAND maker YMTC, is expected to surpass that record within weeks; and robotics firm Unitree's Star Market debut earlier this month was among the strongest of the year. Alibaba, meanwhile, announced an HK$80 billion (US$10.2 billion) Hong Kong share placement to fund its AI build-out.

The concentration of listings raises a governance question that will follow DeepSeek all the way to its debut: how does a company whose open-weight models are used by developers around the world, and whose founder built it inside a quantitative trading firm, adapt to the disclosure and investor-relations demands of a mainland public market? Beijing's preference for keeping frontier AI financing onshore is clear, but the scrutiny will be intense for a company of DeepSeek's profile.

Japan's Sovereign AI Push and the Open-Weight Dilemma

For Japan, DeepSeek's rise is both an opportunity and a challenge. The company's V4 generation, released on April 24 under an MIT licence, includes V4-Pro, a 1.6-trillion-parameter mixture-of-experts model, and V4-Flash, a 284-billion-parameter variant, both free to download and self-host. Japanese enterprises have been weighing the models as a way to keep sensitive data onshore, with engineering guides for self-hosting DeepSeek weights on domestic infrastructure circulating widely since the spring.

Tokyo's policy answer has been a sovereign bet of its own. On April 12, SoftBank, NEC, Sony and Honda jointly established Japan AI Foundation Model Development, a national consortium backed by roughly 1 trillion yen (US$6.7 billion) in government funds, to build a domestically produced trillion-parameter model focused on physical AI. The timing was not accidental: the consortium was unveiled within days of DeepSeek's V4 release, underscoring how Chinese open weights have become a reference point for Japan's industrial AI strategy. The API price hikes add a commercial dimension to the strategic debate, reminding Japanese developers that dependence on a Chinese supplier carries pricing risk as well as data-sovereignty concerns.

The Computing Power Race Behind the Valuation

The 500 billion yuan valuation is ultimately a bet on compute. DeepSeek is reported to be planning to add about 1 gigawatt of computing capacity as competition intensifies, a scale that would place it among the largest AI infrastructure operators in China. That ambition collides with US export controls that restrict advanced chip shipments to China, forcing domestic AI leaders to optimise around available hardware, domestic accelerators and cloud-scale engineering. DeepSeek's original breakthrough was efficiency, but the new round signals that even the most efficient Chinese lab now needs capital at sovereign scale to stay in the race against local rivals such as Zhipu AI and against US frontier labs.

What to Watch For

Three milestones will define the next phase. First, the round itself: investors including CPE, Legend Capital and Stony Creek Capital are still in talks, and the promised end-of-August close will test demand at a US$74 billion valuation. Second, the filing window: if DeepSeek files its prospectus by the end of the year, the Star Market's review process will put one of the world's most influential AI companies under mainland disclosure rules for the first time. Third, pricing discipline: whether the industry-wide price increases hold as competition intensifies will determine how quickly China's AI leaders reach profitability.

For the Asia-Pacific region, the stakes are strategic. A listed DeepSeek would give China's open-weight champion a deep-pocketed, onshore capital base, while Japan pursues its own trillion-parameter sovereign model. The race is no longer just about which lab produces the best model; it is about which economy can finance, power and govern frontier AI at national scale. Tokyo will be watching the Star Market filing closely.

By Kenji Tanaka, Staff Writer

This article was produced with AI-assisted research and editorial support. Sources: South China Morning Post, Reuters, China Daily, TechTimes, TechWire Asia.