DeepSeek spent its first three years refusing outside money entirely. It was founded and funded by Liang Wenfeng through his quantitative hedge fund, and that self-funded posture became part of its identity. That changed in 2026. The company raised $7.4 billion in its first-ever external round in late May. Now, weeks later, it wants more.
The new round targets a pre-money valuation of $71 billion. That is up from $52 billion after the first round, which was itself up from around $20 billion in early April talks. In roughly three months, the company's estimated value has more than tripled. The pace reflects real investor appetite, not just hype: DeepSeek's models are actually being used at scale.
The reason DeepSeek needs this money is tied directly to its pricing. Its V4-Pro model currently costs businesses around 97% less per use than OpenAI's GPT-5.5. On output tokens, the gap is even wider: DeepSeek's cheaper variant costs roughly 100 times less than GPT-5.5 at equivalent context lengths. That kind of pricing wins customers fast. Spending data from Ramp, which tracks real purchases across more than 50,000 companies, shows DeepSeek was among the fastest-growing software vendors among US businesses in June.
But selling at near-cost prices while simultaneously building data centers and buying AI chips is a cash-intensive operation. The money from both rounds is going toward exactly that: physical computing infrastructure and recruitment. DeepSeek has also launched a push to build its own AI chip, reducing its dependence on Nvidia and Huawei hardware.
On top of the new funding round, Bloomberg reports that DeepSeek is preparing for a stock market listing in mainland China. The company is working with accounting firms to complete its financial reports by year-end, a required step before filing. The target is a filing in late 2026 or early 2027, with a public debut in 2027. This means DeepSeek is doing three things simultaneously: raising private capital, building infrastructure, and preparing to go public.
For business operators, the relevant question is what DeepSeek's growth actually means in practice. The short answer: it is the primary force pushing AI costs down right now, and that trend is likely to continue as long as it has capital to sustain it. Tasks that seemed too expensive to automate on Western AI platforms are becoming economically viable on DeepSeek's pricing. That changes the math on dozens of back-office, document processing, and customer communication use cases.
The catch is the data question. DeepSeek's own privacy policy states that all user data is stored on servers in China. Under Chinese law, the government can require access to that data without the company being able to refuse or challenge the request. Multiple governments, including those in Italy, Australia, South Korea, and Taiwan, have restricted or banned it in public-sector settings. Several US states and federal agencies have done the same. South Korea's data protection authority found that DeepSeek transferred user prompts to other Chinese companies without user consent.
This creates a clear split. The open-source version of DeepSeek's models, which businesses can run on their own servers, carries none of those risks: the data never leaves your infrastructure. Using DeepSeek's own app or API directly is a different matter entirely, especially for anything involving client data, contracts, or internal business information.
DeepSeek's IPO, if it proceeds, would arrive as OpenAI and Anthropic are also preparing for public listings targeting valuations around $1 trillion each. The difference in scale is significant: $71 billion versus $1 trillion. But DeepSeek's lower valuation reflects a different business model, one built around open-source distribution and low-cost access rather than locked-in subscriptions. Its long-term commercial plan involves selling private, on-site deployments to enterprises that cannot send data to any public cloud, whether Chinese or American.