DeepSeek valuation hits $71 billion as it launches second funding round

DeepSeek has begun talks for a second funding round at a roughly $71 billion pre-money valuation, just weeks after closing its first external raise, as the Chinese AI lab races to fund data centers, proprietary chips, and a hiring push that could double its workforce.

"The speed of this raise reflects the reality that even the most efficient AI lab cannot escape the capital intensity of infrastructure at scale," a person familiar with the company's strategy said.

The new round targets at least 10 billion yuan ($1.5 billion), according to Bloomberg, and values the company about 37% higher than the roughly $52 billion post-money valuation from its first round in June. That initial raise of about $7 billion included investments from Tencent at roughly $1.5 billion, battery maker CATL, and founder Liang Wenfeng, who committed about $3 billion of his own money. Gaming firm NetEase and retailer JD.com were also in talks to participate.

The rapid-fire fundraising marks DeepSeek's transition from a research lab known for cost-efficient model training into a capital-intensive infrastructure builder. The company is planning gigawatt-scale computing parks — projects that can cost tens of billions of yuan each — and has begun developing proprietary AI inference chips to reduce reliance on Nvidia and Huawei. It is also preparing for an initial public offering on mainland China's A-share market, with a potential filing this year and a listing targeted for 2027.

From cost efficiency to capital intensity

DeepSeek was founded in 2023 by Liang through High-Flyer, the quantitative hedge fund he also built, and initially funded entirely from internal capital. The company upended Silicon Valley in early 2025 with its open-source R1 reasoning model, which matched frontier Western systems on benchmarks like MMLU while costing a fraction to train. That reputation for efficiency made its decision to seek outside funding in June a turning point for the industry.

Now the calculus has shifted. The company's push into agentic AI products — which require continuous multi-turn reasoning and tool invocation — consumes exponentially more computing power per task than traditional chatbot interactions. DeepSeek's Harness team, responsible for building these Agent products, has publicly described being "severely understaffed" due to business expansion, conducting high-frequency interviews daily to fill roles.

The infrastructure demands extend beyond software. DeepSeek posted recruitment plans in June for an IDC data center team to build gigawatt-scale computing parks covering the full chain from planning to operations. For reference, Chinese clean energy company Jinko Power Technology signed an agreement in April for a 1-gigawatt computing center project in Ningxia with a planned total investment of about 24.5 billion yuan ($3.6 billion).

On the chip front, Reuters reported this month that DeepSeek began work on proprietary AI inference chips about a year ago and is now engaging with chip design firms, wafer foundries, and memory companies. The effort aims to reduce dependence on external suppliers including Nvidia and Huawei, whose advanced chips face US export restrictions that make procurement more difficult and expensive for Chinese firms.

IPO plans and the valuation gap

DeepSeek is simultaneously pushing toward public markets. The company is working with auditors to finalize its financial statements by the end of this year, a prerequisite for filing an IPO prospectus on the Shanghai Stock Exchange's STAR Market, according to people familiar with the matter. A listing could come as early as 2027, though the timeline remains subject to market conditions and the company's own performance.

At $71 billion, DeepSeek's valuation remains modest compared with US rivals. Anthropic is valued near $965 billion, and OpenAI is chasing a $1 trillion valuation. The gap underscores China's position in the global AI race: strong model performance, but far less access to capital and a ceiling on advanced chip imports.

Liang structured the first funding round to preserve control — most investors put money into a limited partnership he controls, with no voting rights and a five-year lock-up. The one exception was China's state-backed National AI Industry Investment Fund, which invested directly with voting rights and no lock-up, signaling deepening government interest in the company.

DeepSeek's official version of the V4 model is scheduled for release around mid-July, the first major product launch since completing its initial funding round. The model's performance will offer investors their first look at whether the company's rapid scaling can sustain the technical edge that made it a household name.

For investors, the key question is whether DeepSeek can convert its massive capital raise into a durable competitive moat. Nvidia shares, which fell sharply when DeepSeek's R1 model first demonstrated that frontier AI could be built with fewer chips, have since recovered as the market recognized that cheaper inference drives more total compute demand — a dynamic that could repeat as DeepSeek scales. The company's ability to execute on its data center and chip ambitions while maintaining its research edge will determine whether the $71 billion valuation proves a floor or a ceiling.

This article is for informational purposes only and does not constitute investment advice.