DeepSeek is reportedly preparing for a fundraising round that would put an extraordinary price on the ability to develop models and secure the computing needed to run them. The Wall Street Journal reports that the Hangzhou-based company is seeking US$7.4 billion, in a transaction that could value it at about US$74 billion before the new investment. The report cited people familiar with the matter. It did not say that the round had closed, and its figures should be read as reported terms rather than a completed deal.
The proposed raise is notable not simply because of its size. The Journal said DeepSeek aims to use the money for research and development and for expanding computing infrastructure. That combination shows how capital needs in frontier AI are changing. A successful model may create demand, but maintaining progress requires expensive work on training, inference, engineering talent, systems software, and access to clusters that can deliver AI services at scale.
The report said existing shareholders, including Monolith Management, Shixiang, and battery maker CATL, were expected to put more capital into the company. It also said some funds backed by local Chinese government vehicles planned to invest. Those details remain subject to the same anonymous-source qualification. What is already clear is that DeepSeek is at the center of investor interest because its models helped force a reassessment of how much capability can be built with an open-weight approach.
A Reported Valuation Reflects a Compute-Intensive Strategy
The US$74 billion figure would be a sharp step up from the more than US$50 billion valuation that the Journal said investors assigned DeepSeek in June. Valuations of this kind often reflect expectations rather than current financial results. Investors are pricing a company’s ability to convert technical momentum into enterprise use, developer adoption, and eventually sustainable revenue. In AI, they are also pricing the cost of keeping that momentum alive.
DeepSeek has become a high-profile example of a Chinese lab whose work affects decisions across the industry. The relevance of its models has reached hardware providers as well as application builders. Nvidia’s new support for Chinese open systems, including DeepSeek V4 Flash, underlines that those models now matter to the global infrastructure market. The resulting feedback loop is powerful: a more widely used model attracts engineering support, which can make it easier for more developers to adopt.
The proposed funding also points to a challenge that affects all Chinese AI companies. A model can be available under an open-weight structure, but the company behind it still needs substantial capital to improve the system, run services for customers, and build the tools that turn research into a reliable product. That is where computing infrastructure becomes central. Training and inference are separate costs, and the latter can rise rapidly if a model gains broad use.
EastFrontier’s report on Mucang’s Series B showed why investors increasingly look beyond processors alone. High-speed networking and cluster architecture shape whether large numbers of accelerators can work as an effective system. A DeepSeek raise aimed at infrastructure would fit into the same wider effort to make AI capacity more available and usable inside China.
The Structure of the Deal Matters as Much as Its Size
The Journal said DeepSeek’s earlier funding process required many investors to put money into a limited partnership managed by founder Liang Wenfeng. In the reported new round, the company would allow more direct investment. If accurate, that change matters because direct ownership can broaden the list of funds able to participate and make the company’s capitalization easier for outside investors to understand.
The report also said DeepSeek is targeting an initial public offering next year. That is an aspiration, not a confirmed timetable. A fundraising round can prepare a company for a listing by expanding its balance sheet, setting a market valuation, and bringing in investors who can support later financing. It can also raise expectations for governance, revenue disclosure, and a convincing account of how costly AI development becomes a durable business.
The company’s location in Hangzhou matters in this context. The city has a large technology base, close links to Alibaba’s ecosystem, and a growing population of AI startups and cloud users. Yet the model market is increasingly national and global. DeepSeek’s influence reaches beyond the city because enterprises, researchers, and developers can download, adapt, or host open models in many different environments.
That global reach is part of the appeal and the risk. U.S. officials and lawmakers have raised concerns about Chinese open models, while Western companies must decide whether they are comfortable incorporating them into their own products. At the same time, China’s leading firms are pushing model families into workplace tools, cloud services, and industry-specific applications. The launch of QwenWork outside China is one recent example of the commercial paths available to a Chinese model ecosystem.
China’s AI Capital Cycle Is Becoming More Demanding
A US$7.4 billion raise would stand out even in a sector accustomed to large numbers. The reason is straightforward: building, tuning, and operating advanced systems requires ongoing investment rather than one-time product development. Companies need chips and data-center capacity, but they also need people capable of improving algorithms, adapting models to customers, managing safety and reliability, and running services efficiently.
China’s domestic chip and infrastructure companies are raising capital for a related reason. Enflame’s planned Shanghai listing demonstrates the appetite for financing businesses that can supply more of the hardware stack locally. Model developers such as DeepSeek need that ecosystem to mature, even if their own funding round is aimed at a different layer of the market.
The result is a more intertwined capital cycle. Infrastructure providers need model customers. Model companies need reliable infrastructure. Investors must judge whether demand for AI services will grow quickly enough to support both. That is why a prospective DeepSeek financing round is more than a valuation story. It is a measure of investor confidence that Chinese open models can become platforms for a long-lived AI economy.
For now, the basic facts remain provisional. DeepSeek has not announced a completed US$7.4 billion transaction or a US$74 billion valuation. The Journal’s reporting is still significant because it shows the scale of resources investors may be prepared to commit. Whether those reported terms become final will say much about the market’s confidence in DeepSeek’s next phase.