China Chip Fund Leads DeepSeek $45B Funding Round

DeepSeek Built Its Brand Proving Compute Doesnt Matter. Now the Chip Fund Is Its Biggest Backer.

China state chip fund leading DeepSeek B at $45B valuation — more than four times the $10B being discussed three weeks prior, according to the Financial Times and Reuters. The lead investor is the China Integrated Circuit Industry Investment Fund, the Big Fund that has deployed more than $50 billion across Chinas chip ecosystem since 2014. The fund was founded in January 2025 with 60 billion yuan, roughly $8.8 billion. A single investment of $3-4 billion would represent a substantial portion of that capital.

The number is the story everyone is reporting. The irony is the story worth telling.

DeepSeek became a global name in January 2025 when its R1 model sent US technology stocks into a brief panic. The claim: frontier-class reasoning trained for roughly $6 million. Compute was not the prerequisite the industry had assumed. The efficiency gospel.

Eighteen months later, Deepseeks V4 model launched on April 24 — the same week these funding negotiations became public — and explicitly ran on Huawei Ascend chips, not Nvidias. The company that spent two years arguing it didnt need American hardware is now backed by the fund whose entire purpose is to make China self-sufficient in that hardware. Jensen Huang, asked about this scenario: The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.

The deal hasnt closed. DeepSeek, the Big Fund, Tencent, and Alibaba all declined to comment. But the direction is clear, and the structural tension is the story.

DeepSeek is losing ground commercially. Moonshot AIs ARR crossed $100 million in March. MiniMax reported $150 million in annualized recurring revenue as of February. DeepSeek has disclosed no revenue and took no external funding until this year. One investor who has studied the round told 36kr that the valuation has less to do with fundamentals and more to do with establishing a pricing anchor for employee stock options — a structural necessity for a company that has been private since 2023 with no external valuation mark. Researchers are also leaving: Luo Fuli, a notable DeepSeek researcher, departed last year to lead Xiaomis AI model team. The V4 release did not move US markets — unlike R1, which briefly erased hundreds of billions in semiconductor market capitalization. The market noticed the gap between the claim and the current state.

Independent evaluation confirms it. NISTs CAISI evaluation this week found DeepSeek V4 is the most capable Chinese model tested to date, but lags the US frontier by roughly eight months. DeepSeeks own benchmarks had placed V4 alongside GPT-5.4 and Opus 4.6. CAISIS independent assessment, using held-out benchmarks, puts V4s capability level closer to GPT-5, which was released eight months ago. On cost efficiency, V4 scored competitively against GPT-5.4 mini — but that comparison is itself revealing: V4 is being priced against a mid-tier American model, not the frontier.

And yet the valuation is $45 billion. For context: MiniMax, with $150M ARR and a proven business, raised at a fraction of that. The gap is not explained by fundamentals. Its explained by strategic significance.

The Big Funds mandate, historically, has been silicon — fabs, equipment, materials. It has not previously led investments in frontier model labs. The structural shift is the point. As one analysis noted, if Huawei chips can effectively support a frontier-adjacent model, it gives China a clearer path around export restrictions — potentially creating new domestic chip demand at scale rather than merely reallocating existing demand. Beijing appears to have concluded that the answer to American chip export controls is not more chips, but more capital directed at the labs that have shown they can produce results without them. DeepSeek is the proof of concept. The fund is betting the concept scales.

What that means for DeepSeeks unusual open-source commitment is the open question. Chinese state-backed investors typically bring expectations around alignment with national strategy — a signal that the company is now considered strategically important. Whether the Big Funds stake comes with governance rights that constrain DeepSeeks release practices is not yet known. The aislop analysis notes that China is reportedly tightening rules around how AI startups accept foreign capital without government approval — a broader context in which DeepSeeks state-backed round is less exception than pattern.

Founder Liang Wenfeng had been preparing the cap table. In April, he injected personal capital that increased his direct stake from roughly 1% to 34%, with total ownership at around 84%. The injection appears designed to consolidate founder control ahead of outside investors arriving — and, as one investor noted, to provide a market-based pricing anchor for employee stock options that had been unpriced since the companys founding.

Three things matter when this round closes. First, whether Tencent and Alibaba stay in at the reported valuation. Second, what DeepSeek buys with the money — if that means Huawei Ascend chips at scale, the irony is complete. Third, whether the open-weights release strategy survives the state capital transition intact.

The efficiency breakthrough that once seemed to threaten the chip industry has found its largest investor among the chip industrys deepest pockets. That is either the most consequential validation of DeepSeeks model, or the beginning of a different kind of dependency. The round is not closed. But the direction is the story.