DeepSeek Is Raising $7.4 Billion at a $74 Billion Valuation Ahead of a 2027 IPO

DeepSeek is finalizing a new funding round of roughly 50 billion yuan (about $7.4 billion) at a pre-money valuation near 500 billion yuan, or about $74 billion, according to reporting from Bloomberg and several outlets tracking Chinese AI financing. The round is expected to close around the end of August 2026, and people familiar with the plans say the Hangzhou company could file for an initial public offering as soon as the end of this year, with a debut on Shanghai’s STAR Market in 2027. Existing backers including Monolith, Shixiang Capital, and battery maker CATL are participating, with several state-linked and domestic funds in discussions to join.

I have been watching DeepSeek since the R1 moment in early 2025, when a lab most Western investors had never heard of briefly rearranged the entire conversation about how much money a frontier model actually costs. What is striking about this round is how conventional it now looks. The company that once symbolized doing more with less is raising one of the largest AI checks in China to buy the one thing it said it did not need as much of: compute.

What is actually in the round

The numbers being reported are consistent across sources, even if the framing shifts. DeepSeek is raising about 50 billion yuan. The pre-money valuation sits near 500 billion yuan, which converts to roughly $74 billion at current rates. That is a large step up from its first outside round, which closed in June 2026 at a post-money valuation in the range of $50 billion after raising about $7 billion.

DeepSeek has not published a press release confirming the terms, and Bloomberg has noted it could not independently verify some of the details circulating in Chinese media, as Tech Startups reported. Treat the exact figures as close approximations rather than audited numbers until the company or a prospectus says otherwise.

The July pause and the leaked transcript

This round did not move in a straight line. In late July 2026, DeepSeek told prospective investors it was suspending the raise after a transcript of comments attributed to founder Liang Wenfeng went viral on Chinese social platforms. The posts described Liang talking candidly about China’s dependence on Nvidia chips and about the country still trailing the United States on the most advanced AI work. Bloomberg reported it had not confirmed the authenticity of the transcript, and DeepSeek did not publicly quote or deny specific lines.

What reportedly bothered Liang was not only the content but the leak itself, since the remarks appeared to come from a closed meeting tied to the first financing deal. The pause lasted weeks rather than months. By late August the round was back on, at a higher valuation than the version floated in July, which had targeted a pre-money figure of at least 480 billion yuan.

It is a small episode with a larger lesson. DeepSeek spent 2025 as a symbol other people got to define. Now that it is raising money at a bank-sized valuation, every offhand comment from its founder is a market-moving document, whether or not he meant it to be.

Why a lean lab suddenly needs $7 billion

The reason the money matters is compute. Reporting on the round says proceeds will fund about a gigawatt of additional computing capacity, on top of spending on larger models and on hiring against Alibaba’s Qwen team, Tencent, and Zhipu. A gigawatt is the unit people now use to describe serious AI infrastructure, and it is the same scale being thrown around for Western projects like the Nvidia-backed OpenAI data center build in Ohio.

DeepSeek’s early reputation was built on efficiency. The V3 and R1 training runs were reported to cost a fraction of what comparable Western models spent, which is part of why the January 2025 disclosures rattled markets. That efficiency story was always partly a constraint dressed as a choice. Export controls limit what advanced silicon DeepSeek can legally buy, so getting more output per chip was survival, not just philosophy.

The current flagship, the DeepSeek V4 family, arrived on April 24, 2026, with V4-Pro aimed at reasoning quality and V4-Flash aimed at low-latency production and agent workloads. A next-generation reasoning model in the R2 mold has been rumored for over a year and has not shipped, with reporting attributing the delay to Liang not being satisfied with its performance. Training the next tier, and serving it to a user base that now spans consumer apps and API customers, is a capital problem before it is a research problem.

The IPO angle

The part that would have sounded far-fetched a year ago is the listing plan. DeepSeek is reportedly preparing to file for an IPO by the end of 2026 and to debut on Shanghai’s STAR Market in 2027. The STAR Market is China’s board for domestic technology companies, and a DeepSeek listing there would be a milestone for Beijing’s push to keep strategic AI firms funded and traded onshore rather than in New York or Hong Kong.

It also changes the company’s incentives. A pre-IPO DeepSeek has reasons to raise API prices toward sustainable margins, to formalize governance, and to control its public messaging in a way the scrappier 2025 version did not. Some of that is healthy. Some of it means the lab that made frontier AI feel briefly cheap is now on the same treadmill as everyone else.

The bigger picture

Two things can be true at once. DeepSeek genuinely pushed the field toward more efficient training and more open weights, and that pressure has not gone away. And DeepSeek is now a heavily capitalized national champion that needs gigawatts of power, billions in funding, and a stock listing to keep competing. The gap between those two identities is the story of Chinese AI in 2026, the same tension visible when regulators forced consumer AI products to change overnight under China’s AI companion law.

For anyone outside China, the practical takeaway is that the cost floor for frontier AI is rising everywhere. The lab that argued otherwise is raising $7.4 billion to stay in the race.

Frequently asked questions

How much is DeepSeek raising and at what valuation?

Reports point to about 50 billion yuan, roughly $7.4 billion, at a pre-money valuation near 500 billion yuan, or about $74 billion. The round is expected to close around the end of August 2026. DeepSeek has not issued an official confirmation of the terms.

Why did DeepSeek pause the round in July?

The company told prospective investors it was suspending the raise after a transcript of comments attributed to founder Liang Wenfeng, discussing China’s reliance on Nvidia chips and its lag behind the US, spread on social media. Bloomberg said it could not verify the transcript. The round resumed weeks later at a higher valuation.

Is DeepSeek going public?

People familiar with the plans say DeepSeek could file for an IPO as early as the end of 2026, targeting a 2027 listing on Shanghai’s STAR Market. Nothing has been formally filed yet, so the timeline could shift.

What will the money be used for?

Reporting on the round says the capital will fund roughly a gigawatt of additional computing capacity, development of larger models, and recruiting against rivals such as Alibaba’s Qwen team and Zhipu. Compute is the largest single line item.

Does this mean DeepSeek’s low-cost training claims were wrong?

Not exactly. The efficiency gains DeepSeek demonstrated with V3 and R1 were real and still influence how models are trained. But serving a large user base and training each new generation is expensive regardless of how efficient the method is, and export controls make hardware harder to get, which pushes total spending up.