IMPLICATOR
In at least one investor meeting this month, Liang Wenfeng told prospective backers that DeepSeek would keep developing open-source AI models while pursuing artificial general intelligence, according to Bloomberg. The Hangzhou lab has operated as a research shop funded from Liang's quantitative hedge fund, High-Flyer, since 2023. Bloomberg said DeepSeek is now discussing a 70 billion yuan round, about $10 billion, at a pre-money valuation near $45 billion. The lab's first external financing.
The likely investor list includes the National Artificial Intelligence Industry Investment Fund, Beijing's strategic AI vehicle, alongside Tencent, IDG Capital, and Monolith Capital. The state fund is in talks to invest about 10 billion yuan. Liang himself could inject about 20 billion yuan. The round would replace High-Flyer's solo balance sheet with state-linked capital while keeping Liang's research-first AGI pitch in place on paper.
Key Takeaways
- DeepSeek is discussing a 70 billion yuan round while keeping Liang Wenfeng's AGI pledge in place.
- A state-backed AI fund may invest about 10 billion yuan, with Liang potentially adding 20 billion yuan.
- The round would link DeepSeek's open-source strategy more tightly to China's domestic AI stack.
- Security concerns remain sharpest around hosted services, while local open weights carry different risks.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
The investor list is not finalized
Bloomberg said discussions remain in flux and investment amounts could still change. Representatives for the state-backed fund, Tencent, IDG, and Monolith declined to comment or did not respond to requests. The finance ministry, which administers state-backed investment vehicles, did not respond to a faxed request.
Reuters reporting two weeks earlier had put the round at $3 billion to $4 billion at as much as a $50 billion valuation. The Wall Street Journal cited the same upper bound. The South China Morning Post said AI-focused affiliates under the third phase of the China Integrated Circuit Industry Investment Fund, known as the "Big Fund III," were in the financing.
The headline number has moved fast. Bloomberg's $10 billion target sits more than 30 times above the $300 million The Information first reported DeepSeek pursuing in April. The state-fund participation, at about 10 billion yuan, would be the largest single state check disclosed for a Chinese frontier-model maker so far.
What Liang has and has not changed
Liang's older interview with the Chinese outlet Waves, translated by ChinaTalk in late 2024, gave the lab its founding line: "Our destination is AGI." He told the interviewer DeepSeek had no short-term fundraising plans. "Money has never been the problem for us; bans on shipments of advanced chips are the problem," he said. He also resisted the easy capital argument: "More investments do not equal more innovation. Otherwise, big firms would've monopolized all innovation already."
Two of Liang's 2024 positions hold in the present round. AGI is still the stated destination, and chip access is still the binding constraint, with V4's training reportedly migrated toward Huawei Ascend hardware. On fundraising he has reversed himself, with Bloomberg reporting him personally injecting about 20 billion yuan into the round.
The Implicator reported in early May that talent attrition triggered the reversal. Reuters cited the 2025 departure of Luo Fuli, who left to lead Xiaomi's MiMo model team. ChinaTalk's April write-up added that DeepSeek had also lost contributors to Tencent, ByteDance, and DeepRoute.ai. Fortune reported this week that government-linked investors in China went from fewer than 10 AI deals annually before 2018 to more than 140 in 2025, and that semiconductor round sizes reached a median $30.48 million this year. The pool of capital DeepSeek is now drawing from did not exist at this scale when Liang said money was not the problem.
Track China's AI power shift
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Where the new money has to land
V4 set the public terms for that landing. DeepSeek released V4-Pro and V4-Flash on April 24, with V4-Pro priced at $1.74 per million input tokens and $3.48 per million output tokens, a fraction of comparable U.S. systems. ChinaTalk's April assessment said the company's own technical report effectively placed V4 about "3 to 6 months behind" state-of-the-art frontier systems. Huawei confirmed same-day Ascend 950 supernode support; Cambricon followed.
Agents are the second product front. The South China Morning Post reported in mid-May that DeepSeek hired Cui Tianyi, a former Jane Street software engineer in Hong Kong, in March to lead a new AI "harness" team, the infrastructure that turns a model into an agent.
Liang has not abandoned the open-source argument. He told ChinaTalk's 2024 interviewer: "We will not change to closed source. We believe having a strong technical ecosystem first is more important." A Qwen employee, quoted in the ChinaTalk V4 write-up, gave the counter-position: "the golden age of nonprofit AI development is over." Bloomberg's account of the new round ends on the commercial pivot: "The startup is now expanding into agentic AI in the wake of OpenClaw's emergence, tapping a wave of enthusiasm for software that can carry out tasks without human intervention."
The trust line stays at the hosted edge
U.S. regulators have focused on DeepSeek's hosted services rather than the open weights themselves. WIRED reported in January 2025 that DeepSeek's own English-language privacy policy said it stored user data "in secure servers located in the People's Republic of China." The U.S. House Select Committee on the CCP called DeepSeek "a profound threat to our nation's security" in an April 2025 report. A bipartisan finding. The same report alleged the app "siphons data back to the People's Republic of China," "covertly censors and manipulates information pursuant to Chinese law," and was built using "stolen U.S. technology."
DeepSeek's response has been to keep the weights open and the API price low. Locally run V4 does not transmit user prompts to a server in Hangzhou. Regulators in Washington can still object to the app, the API, and Chinese state-linked financing appearing together in a single company.
The final investor list is the next document on the calendar. Bloomberg said the round was in its "final leg." That filing will name how much of DeepSeek's research lab now sits on a state-linked balance sheet, and how much remains with the founder who once said he had no plans to fundraise.
Frequently Asked Questions
How much is DeepSeek trying to raise?
Bloomberg reported DeepSeek is discussing a 70 billion yuan round, about $10 billion, while seeking a pre-money valuation near $45 billion. Earlier reports had put the first external round at $3 billion to $4 billion and up to $50 billion.
Who may invest in the round?
Likely investors include China's National Artificial Intelligence Industry Investment Fund, Tencent, IDG Capital and Monolith Capital, according to Bloomberg. Details remain in flux and representatives for several parties declined or did not respond to requests for comment.
Is DeepSeek abandoning open source?
No. Bloomberg reported Liang Wenfeng pledged to keep developing open-source models while pursuing AGI. That keeps DeepSeek's older research identity in place even as state-backed capital enters the funding talks.
Why does state-backed capital matter here?
State-backed capital would give Beijing a financial role in one of China's most visible open-model companies. It could also connect DeepSeek more tightly to domestic chip and AI infrastructure priorities.
What is the main risk for Western users?
The strongest data-transfer concerns involve DeepSeek's app, API and hosted services, where user data may touch servers in China. Locally run open weights do not carry the same data-transfer profile, though provenance and censorship concerns remain.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: [email protected]
