The story of DeepSeek's first funding round is a tale of immense demand, elusive access, and a staggering amount of capital. The company, which once vowed to avoid external financing and an IPO, has just completed a landmark first round exceeding RMB 50 billion, a process rife with brokers, false promises, and intense competition for a piece of the most coveted AI deal in China.
A source involved in the transaction has revealed that DeepSeek, a foundational AI large model company, concluded its first-round financing negotiations, with agreements signed late last month. This round raised over RMB 50 billion, approximately $7.4 billion. Liang Wenfeng, the founder, contributed RMB 20 billion personally, the single largest investment. Tencent invested RMB 10 billion, Contemporary Amperex Technology Co. Limited (CATL) invested RMB 5 billion, while NetEase, JD.com, and IDG Capital each invested RMB 3 billion. The National AI Industry Investment Fund contributed RMB 1 billion.
The source added that following this round, DeepSeek's pre-money valuation stands at around RMB 350 billion. Including an additional 5% ESOP (employee stock ownership plan) pool, the effective pre-money valuation is approximately RMB 367.5 billion, or about $54.3 billion.
The final list of investors shows significant changes from earlier market rumors, with several top-tier strategic investors and market-oriented institutions joining. Notably, and to the surprise of the source, the allocation secured by state-owned capital was ultimately the smallest.
Since April, DeepSeek's soaring valuation has captivated the private market. Its valuation jumped from $20 billion in April to around $54 billion currently. Upon completion, this round makes DeepSeek the Chinese AI large model company with the largest-ever initial fundraising round.
This marks the company's first external fundraising in its three-year history, having previously adhered to a "no financing, no IPO, no commercialization" stance. The market's focus on the final investor list was unprecedented. Investors scrambled for allocations, with one reportedly paying a RMB 5 million "meeting fee," traveling to Hangzhou three times, and waiting outside DeepSeek's offices, all without securing a meeting with Liang Wenfeng. Another state fund investor met with eight self-proclaimed intermediaries claiming to have allocations, with little success.
Even now, with the round closed, individuals are still peddling purported DeepSeek investment shares. The fervor in China's AI sector continues unabated, with companies like Zhipu and MiniMax seeing their market caps rise post-Hong Kong listings, and Moonshot AI and StepFun securing multi-billion dollar rounds. Top startup valuations are doubling within months. DeepSeek's funding process epitomizes the capital market's "fear of missing out" (FOMO).
Brokers and Phantom Allocations
Numerous investment firms have been scrambling for access, leading to a proliferation of brokers and intermediaries of dubious origin. One investor, Li Jing, was tasked with finding an allocation. After news of the fundraising broke, he traveled to Hangzhou three times hoping to meet Liang Wenfeng, returning empty-handed each time. In late May, Li Jing was introduced to someone claiming to have connections to Liang, who demanded a RMB 5 million fee to arrange a meeting. Li Jing arrived in Hangzhou early, staking out the headquarters, but upon hearing the fundraising situation had changed, he canceled the meeting and retrieved his money.
"The ability to pay the money is a form of capability in itself," Li Jing remarked. For most LPs, the biggest challenge isn't the amount to invest but finding the right person to give it to. With opaque channels, people are profiting at every intermediary step.
DeepSeek is undeniably one of China's hottest AI investment targets. Despite the investor list being finalized, social media is still flooded with offers to sell DeepSeek shares.
Globally, the AI investment boom persists, driving valuations for hot targets sky-high. Following its latest $6.5 billion round, U.S. model startup Anthropic's valuation reached $96.5 billion, surpassing OpenAI. For such sought-after targets, investing through Special Purpose Vehicles (SPVs) was once common practice, allowing GPs to bundle assets for new LPs who often lacked the clout for direct investment.
Such layered investments incur extra management fees, a method rarely used by top-tier funds. However, many investors told us they were willing to pay these extra fees, betting that DeepSeek's valuation would continue to rise and they would profit upon a future IPO.
Posing as an LP, we contacted one channel contact. He claimed to have secured a RMB 500 million allocation from the "National Integrated Circuit Fund," available only to individuals. Investors would enter as direct LPs in the state-owned shareholder structure, provided they submitted proof of sufficient assets within two days, with subscription closing as early as this week.
He also mentioned that the FA and GP would charge upfront management fees of 8% and 9% respectively, totaling a staggering 17%. This means on a RMB 100 million investment, RMB 17 million would be deducted upfront—a rarity in standard private equity, where GPs typically charge around 2% annually. We learned that intermediaries selling DeepSeek shares often demand a lump sum covering 3 to 5 years of management fees.
These FAs also create an atmosphere of extreme scarcity. Another FA, claiming to have a RMB 2 billion DeepSeek allocation from a "local state-owned entity," said their channel had several funds of RMB 2 billion each, "basically snapped up in two or three days." However, when asked for the GP's name, they became evasive, only mentioning a "long-term cooperative relationship." Yet, two weeks later, the same intermediary said that channel still had available quota.
Despite the dubious sources, some are willing to gamble. Investor Chen Duo, who works for a state fund managing tens of billions, spoke with eight different FAs. Most claimed allocations from "a local state-owned entity," but after Chen Duo provided proof of funds (POF), communication typically ceased. He considers any offer without the ability to verify the underlying investment agreement or confirmation letter on-site as mere "hype."
Through these interactions, Chen Duo discerned the fundraising logic: many institutions first raise a fund, then use the committed capital to seek an allocation from DeepSeek. FAs then craft a seemingly credible channel using claims of high-level connections, pressuring investors for quick proof of funds. "These institutions want Liang Wenfeng to think they already have the money to secure the quota," he explained. This creates a chicken-and-egg cycle, as raising funds doesn't guarantee a DeepSeek allocation.
An investment manager at a local state fund expressed frustration that FAs still approach him to introduce LPs. "DeepSeek isn't short of LPs and doesn't need FAs," he said, adding that only very small institutions or individual LPs might be persuaded by such unclear channels.
A source close to DeepSeek revealed that all capital in this round came from main funds, with no SPV allocations. Reports indicate Liang Wenfeng and his team required verification of all participating LP identities to prevent shares from going to unknown investors. The source stated that FAs still trying to raise funds for this round won't succeed, questioning their motives.
Who Made the Cut?
Tang Yu, a partner at a private equity fund, has followed DeepSeek since its viral fame in February 2025. After news of the fundraising broke, Tang Yu was privately commissioned by a state-backed fund with RMB 3 billion to find a channel. Using personal connections, he contacted a senior executive at幻方量化, a quantitative hedge fund controlled by Liang Wenfeng that has long funded DeepSeek's R&D.
Despite their long-standing relationship and the substantial sum, the executive merely provided a company email, instructing him to send institutional introductions, letters of intent, and proposals, stating they would be "reviewed uniformly by the responsible person." No further response was received.
Tang Yu believes all serious inquiries to this executive received similar replies. On May 28th, he inquired again about the fundraising progress and was told it was too late for this round, with the timing of the next round unknown.
Multiple investors involved told us that Liang Wenfeng personally selected the investors for this round, favoring those who wished for his success or had good relationships with him. Contact was primarily through his invitation; unsolicited attempts to establish effective connections, even with multiple meeting attempts, largely failed. Another investor, Hu Jie, concurred that funds lacking a personal connection or reputation with Liang had little chance.
Hu Jie, from a market-oriented investment firm where AI is a key focus, had little initial interest. Early information suggested state capital would dominate the round, and the deal size was already large, making the potential return on investment unclear. He wasn't alone in this view; many initially believed it was a state-led deal.
However, after the May holiday, Hu Jie noticed several investment firms, including Monolith Capital and Hillhouse Capital, engaging intensively with DeepSeek. His firm subsequently joined the negotiations.
To the surprise of many participants, the final state capital allocation was lower than expected. Reports indicate that only the National AI Industry Investment Fund invested directly in DeepSeek itself, free from lock-up periods and with voting rights. Other institutions invested in a limited partnership managed by Liang Wenfeng, not directly in DeepSeek, with their shares locked up for five years.
Public information shows the National AI Industry Investment Fund was established in early 2025. Its recent investments focus on AI, chips, and robotics, including companies like Moonshot AI.
Several state-backed funds actively sought allocations in this round. Chen Duo, serving a state fund, contacted eight FAs before getting close. Another local state fund investment manager focused on AI said he started following DeepSeek during the 2025 Spring Festival, also targeting明星模型公司 like Zhipu and Moonshot AI, but failed to invest in any, often unable to even make contact.
"Objectively speaking, state capital has its limitations," he said. "Large companies require rapid capital deployment in single, large tranches. For state funds, that implies significant risk."
The investor list also includes several tech giants. Among them, Tencent secured the largest allocation, followed by CATL. While earlier reports suggested Alibaba was in talks, the company has publicly denied this, expressing limited interest.
Several AI investors analyzed that between Tencent and Alibaba, Tencent had stronger motivation to join. Its foundational model capabilities have long lagged behind the top tier, creating a stronger desire for strategic cooperation with DeepSeek. Alibaba, having built its own ecosystem from models to chips, has less strategic urgency to invest.
A transaction participant believed the foundation for cooperation between DeepSeek and Alibaba was weaker than with Tencent.
CATL's participation as a new energy giant highlights AI model makers' strategic focus on the energy variable behind computing infrastructure. A source revealed DeepSeek is building its own data centers to operate its computing power. Job postings in late April sought roles like Data Center Senior Operations Engineer and Senior Delivery Manager in Ulanqab, responsible for the full lifecycle management of data centers.
DeepSeek's Valuation Rationale
According to multiple AI investors, DeepSeek opened its doors to funding for at least two reasons. First, continuous, large-scale investment in training compute is essential for model iteration, and external capital is crucial for this long-cycle endeavor, as self-generated assets from its quantitative hedge fund affiliate may no longer suffice.
An analyst noted that for top model companies like Anthropic, the business model is both "top-tier" and under "immense pressure" because Scaling Law remains in effect. Each generation's training cost exceeds the last, requiring continuous investment in compute and data. Falling behind means being forgotten. For AI startups, the private market alone can no longer fully fund model iteration, pushing them toward public markets.
Second, it allows for market validation of employee stock options to retain talent. This round created an additional 5% ESOP pool, worth roughly RMB 17.5 billion. However, prior analysis indicated that as of the V4 model release, DeepSeek's core research team had not experienced significant talent attrition.
Hu Jie mentioned that DeepSeek employees also hope for an IPO. An AI professional acquainted with DeepSeek R&D staff said it's unsurprising employees expect a listing: "People are not saints; those staying at DeepSeek cannot rely on idealism alone."
Since 2025, China's leading AI model companies have been fundraising and listing at a rapid pace. Beyond Zhipu and MiniMax's Hong Kong listings earlier this year, StepFun recently raised about $2.5 billion with an expected cornerstone valuation around $10 billion. Public records show Moonshot AI has raised approximately $4.4 billion with a latest pre-money valuation of $20 billion. Moonshot is dismantling its VIE structure for a listing, while StepFun aims to file as early as the first half of this year.
In contrast, DeepSeek started fundraising last but secured about $7.4 billion in its first round with a $54.3 billion valuation, far surpassing its competitors.
Internal and external factors drive DeepSeek's high valuation. Several transaction participants believe the current range is reasonable, as its inference efficiency is widely recognized as industry-leading.
Since its breakthrough R1 model last year, DeepSeek has maintained low cost and high performance. For example, the V4-Pro released in April requires only 27% of the FLOPs per token compared to its predecessor V3.2 for a 1 million token context. This efficiency advantage translates directly into API pricing, with V4-Pro at $0.025 per million tokens, among the lowest globally. This extreme affordability has made DeepSeek immensely popular with global developers, topping OpenRouter's token consumption chart this month for its V4 Flash model.
A dollar fund professional familiar with AI told us that on OpenRouter, Chinese model usage has surged from 1% three years ago to 60%. Chinese companies have successfully extended the supply chain cost-reduction advantage to AI, turning large models into a standardized commodity, with DeepSeek executing this most effectively.
Furthermore, investors and industry experts analyzed that DeepSeek's high valuation also stems from its rising status as a national-level AI strategic platform.
External market conditions also play a role. The secondary market performance of Zhipu and MiniMax has influenced pricing for later entrants. Zhipu, for instance, listed with a market cap of HK$57.9 billion, which has since risen to over HK$740 billion. A state fund professional familiar with AI venture capital noted the sharp post-Spring Festival rally in对标模型公司 like Zhipu and MiniMax created a powerful wealth effect in the capital markets.
This professional also believes the current moment is critical for discussing large model commercialization. After three to four years of development, technology can now support mature product deployment, and the market is eager to validate AI's commercial value. Consequently, more private capital is flowing into model companies, which state-backed funds previously viewed as lacking certainty.
Contrary to the prevailing FOMO sentiment, some investors who never sought involvement consider DeepSeek's current valuation too high and "unfathomable." They view participation as consuming excessive time and money with a low probability of success, making the potential return on investment unattractive. Hu Jie describes DeepSeek as a "white horse" stock—a high-market-cap, relatively stable "top performer."
Facing the capital markets, DeepSeek cannot avoid the challenge common to all tech startups: balancing technological ideals with commercial returns long-term. How to commercialize remains a crucial question hanging over its high valuation. A report in March indicated DeepSeek executives acknowledged external focus on business model implementation and technological progress, stating the company was making efforts, trying various approaches, and had preliminarily validated some paths.
Compared to other model companies and internet giants, DeepSeek's progress on the product front has been slower, with its focus historically on model training. Regarding whether DeepSeek should develop products, Hu Jie said Liang Wenfeng likely realized the importance of productization around late last year, but the team hadn't found a suitable lead, struggling with the product roadmap. "The product direction is something Liang Wenfeng has to think through himself," he noted. Recent job postings show DeepSeek is now hiring for various product-related roles.
This doesn't necessarily signal a strategic shift. According to a May 22nd media report, Liang Wenfeng told investors the company would continue advancing open-source AI models with the goal of achieving Artificial General Intelligence (AGI). In his view, DeepSeek's core mission is to push technological boundaries, not pursue profitability.
Hu Jie stated that fundraising is merely a capital operation; DeepSeek's daily operations and technical roadmap are not necessarily bound by capital market pressures.
An investor on the final list told us that for DeepSeek, short-term commercialization is less important; achieving AGI is paramount, and China must have its own AGI.
(Names of interviewees have been changed at their request.)
Comments