DeepSeek Nears $7.4B Raise at $59B in First-Ever Funding Round

Chinese AI lab DeepSeek is in final talks to raise $7.4 billion in its first-ever funding round, according to reports from Bloomberg, CNBC, and PYMNTS published Wednesday. The round would value the company at between $52 billion and $59 billion and is expected to close in the next two weeks.

The lead investors are Tencent Holdings and CATL, the world's largest battery maker. The state-backed National Artificial Intelligence Industry Investment Fund is also participating, in what analysts are reading as a major endorsement from Beijing. Founder Liang Wenfeng is committing $2.9 billion of his personal capital, an unusual move that signals founder conviction.

Why this matters

DeepSeek has built one of the most capable open-source AI model families in the world without ever taking outside money. Its V4 series matches or beats frontier US models on many benchmarks, runs at a fraction of the cost, and has captured the largest share of open-source model traffic on OpenRouter, the leading multi-model API gateway.

Until now, DeepSeek's funding came from Liang's personal wealth and revenue from High-Flyer, the quantitative hedge fund he founded. That model worked when DeepSeek was a research lab. It doesn't work as the company scales to compete with OpenAI, Anthropic, and Google at the frontier.

A $59 billion valuation also makes DeepSeek the second most valuable private AI company in the world, behind only Anthropic, which is closing its own round at a $965 billion valuation ahead of a confidential IPO filing.

What was announced

Bloomberg reported the round on Wednesday. CNBC and PYMNTS confirmed the structure within hours. DeepSeek is raising $7.4 billion at a post-money valuation between 350 billion yuan and 400 billion yuan, or $52 billion to $59 billion.

Tencent and CATL are the largest external investors. The state-backed National Artificial Intelligence Industry Investment Fund is participating. NetEase, the Chinese gaming developer, and JD.com, the e-commerce giant, are in final talks to join. Founder Liang Wenfeng is putting in $2.9 billion of his own money.

DeepSeek's pitch to investors has been research-first. Senior management has told potential backers that the company will focus on "groundbreaking AI research instead of short-term commercialization," and that Liang is personally committed to keeping the model weights open and pursuing artificial general intelligence on an open-source path.

Industry impact

For the open-source AI ecosystem, this round is a generational moment. The biggest threat to open weights has always been economic: at some point, the cost of frontier training runs exceeds what any open-source lab can sustain. A well-capitalized DeepSeek extends the open-source runway by several years and makes it credible that the frontier could remain at least partly open through 2027 and 2028.

For US AI labs, the round changes the competitive picture in two ways. First, DeepSeek can now match training compute budgets that previously only OpenAI, Anthropic, and Google could afford. Second, the participation of the National AI Industry Investment Fund means Beijing is openly backing DeepSeek as a national champion — making the company harder to sanction and easier to scale inside China.

For enterprise AI buyers, the practical question is whether DeepSeek's open weights remain commercially usable. Existing licenses allow broad commercial use, but a $7.4 billion fundraise creates investor expectations around revenue. Companies with DeepSeek deployments in production should expect more aggressive enterprise sales motions and possibly tiered licensing for the largest customers.

Expert perspectives

ProActive Investors framed the round as "exposing the gulf between China's AI pragmatism and Silicon Valley's valuation machine." DeepSeek is raising at $59 billion. Anthropic is raising at $965 billion. The two companies have comparable model capability and arguably comparable strategic importance, but the valuation gap is roughly 16x. The disparity reflects different access to public-market exit paths more than it reflects technology.

Bloomberg's coverage emphasized founder Liang Wenfeng's personal commitment. Putting $2.9 billion of his own money into the round is the strongest possible signal that he believes the company is undervalued and that he wants to keep voting control. It's also a public commitment that the open-source research mission won't be diluted by investor pressure.

Analysts at Yellow.com and Let's Data Science highlighted CATL as the more strategically interesting investor. A battery and energy company on the cap table of an AI lab is unusual unless the lab is planning to vertically integrate into compute infrastructure. Watch for DeepSeek announcements about energy-aware training, on-site power, or custom hardware in the next 6 to 12 months.

What's next

The round is expected to close within two weeks. After closing, expect DeepSeek to announce major training infrastructure expansion — new clusters, possibly custom silicon partnerships, and more aggressive model release cadence. The company has been training V5 since late Q1; the new capital makes it plausible that V5 ships before the end of 2026 at a meaningful capability jump over V4.

Also watch for international expansion. DeepSeek has been quiet about non-China markets, but a state-backed war chest gives the company the resources to build out enterprise sales in Southeast Asia, the Middle East, and Latin America — markets where US AI labs face procurement and data-residency friction.

For US policy, expect the round to accelerate the conversation about how to handle a well-funded Chinese open-source AI competitor. The Trump executive order signed yesterday gives the federal government a 30-day review window on frontier models from US labs. It has no leverage over DeepSeek, which means the open-source frontier could move outside US regulatory reach.

Bottom line

DeepSeek just became a properly capitalized open-source AI lab with Beijing's backing and the founder's personal $2.9 billion bet. Open weights at the frontier are now economically sustainable through at least 2027. If you're building on open models, this is great news. If you're a US frontier lab, the competitive picture just got harder.