DeepSeek Raises $7.4B in Its First Outside Round

DeepSeek has raised more than 50 billion yuan — about $7.4 billion — in the first round of outside investment in its history, valuing the Chinese lab at roughly $50 billion. Tencent and battery giant CATL are among the largest external backers, and China's state-backed AI investment fund took a direct stake that, unusually, keeps its voting rights. Until now DeepSeek had been bankrolled entirely by High-Flyer, the quantitative hedge fund founded by CEO Liang Wenfeng, who personally put in around 20 billion yuan of this round himself.

The structure is as telling as the size. Rather than buying into DeepSeek directly, outside investors funded a limited partnership managed by Liang, they receive no voting rights, and their money is locked up for five years. That is a long way from a standard venture term sheet, and it spells out exactly what Liang wanted: capital without ceding control. As recently as April the lab was reportedly seeking money at a $10 billion valuation — the figure has quintupled in two months.

Why a self-funded lab finally took the money

DeepSeek built its name by refusing the usual playbook. Its V3 and R1 models stunned the industry in early 2025 by matching frontier performance at a fraction of the training cost, and it shipped them as open weights while OpenAI and Anthropic kept theirs closed. In April 2026 it released V4, its largest open-weights model yet, trained to run on Huawei chips under US export controls, and cut prices to the bone — a permanent 75% discount it says makes V4 around 11x cheaper on input and 35x cheaper on output than OpenAI's GPT-5.5. Giving the model away and gutting prices is not a business that funds its own GPU bills. Frontier training runs cost billions, and even a hedge fund's balance sheet has a floor.

Why it matters

If you use a cheap or free AI through some app or API, there is a real chance DeepSeek's weights are underneath it — and this round shows who pays for that. Chinese tech money and state capital are betting $50 billion that an open, low-cost alternative to the American labs is worth backing. Liang told investors he prioritizes "foundational AI research and AGI development over short-term profits" and will keep open-sourcing models. That is good news if you depend on open weights, but it is worth remembering the cap table now includes a state fund with an actual seat at the table.

The number that lingers is the gap. DeepSeek is doing competitive frontier work at a $50 billion valuation while OpenAI and Anthropic each circle a trillion — it is matching the leaders for what is, to them, a rounding error. That pressure, more than any single model launch, is the reason the price of using AI keeps falling for everyone downstream.