TL;DR
- First Outside Round: DeepSeek is in talks to raise $300 million or more at a valuation above $10 billion.
- Strategic Reversal: The move marks a sharp shift for founder Liang Wenfeng, who long positioned DeepSeek as a research-first lab.
- Pressure Points: Fundraising talks coincide with a delayed V4 flagship, a talent drain to Xiaomi and ByteDance, and costly Huawei chip compatibility work.
- Valuation Gap: DeepSeek’s $10 billion target sits at a tiny fraction of OpenAI’s roughly $852 billion and Anthropic’s $380 billion valuations.
- Open Questions: Talks remain unclosed, with governance terms, investor identity, and final valuation all pending as V4 nears launch.
China’s DeepSeek AI lab built its reputation on refusing outside money. According to an April 18 report from The Information, the Chinese AI lab is now in talks to raise $300 million or more at a valuation above $10 billion, in its first external capital round.
The proposed round would mark DeepSeek’s first external capital raise since its 2023 founding and a sharp reversal of founder and CEO Liang Wenfeng’s long-stated posture as a champion of research independence. It arrives as DeepSeek confronts a talent drain, a delayed V4 flagship, and U.S. rivals raising at valuations nearly a hundred times larger.
A First Outside Round
DeepSeek is seeking outside funding of $300 million or more at a ten-figure post-money mark, citing The Information’s original report. Until now, High-Flyer Capital Management, the Hangzhou quantitative hedge fund that spawned the AI lab in 2023, funded the operation entirely on its own.
However, earlier approaches from several of China’s largest venture capital firms and technology companies were turned aside, a stance This is consistent with Liang’s view that DeepSeek should prize research independence over outside ownership. The lab previously rejected offers from major Chinese funds and tech conglomerates.
Why the shift matters: Liang has repeatedly argued that hedge-fund parentage gave DeepSeek room to pursue open research without quarterly investor pressure. Recent reporting consistently described the lab as having no immediate fundraising plans. However, current talks, still said to be unclosed, indicate the economics of frontier AI have grown too expensive to sustain on High-Flyer’s balance sheet alone.
Meanwhile, no official confirmation has come. DeepSeek has not commented publicly, and the deal’s structure, including whether Chinese sovereign or strategic capital participates, remains unreported.
Meanwhile, one factor the round does not resolve is governance. Outside investors generally demand seats, information rights, or observer access, and any such arrangement would dilute Liang’s unusually free hand over research direction. Whether he accepts those trade-offs, and in exchange for how much capital, will shape how closely the upcoming V4 and successor models track commercial priorities versus pure benchmark ambition.
Timing also matters for DeepSeek’s open-source posture. DeepSeek R1 and DeepSeek V3 were released with weights and papers that accelerated third-party fine-tuning worldwide.
As a result, incoming investors may press for a more guarded release cadence, paid API tiers, or enterprise-only features that recoup infrastructure costs. Each adjustment, however modest, would narrow the gap between DeepSeek and its more commercially oriented U.S. and Chinese peers.
Pressure from Departures and a Delayed Flagship
Fundraising talks coincide with a string of high-profile exits. Luo Fuli, a co-developer of DeepSeek’s V3 model, left for Xiaomi, and researcher Guo Daya departed to join ByteDance. Both moves point to intensifying competition for Chinese AI talent at a moment when domestic tech giants are accelerating their own model programs.
Building on that pressure, DeepSeek’s next flagship model, V4, has been pushed back several times, in part because engineers are working to make it compatible with Huawei chips, an effort tied to Beijing’s push to reduce reliance on US silicon.
Meanwhile, a gray-scale test interface on Weibo surfaced in early April and hinted at a launch this month, showing three modes beyond Deep Thinking and Smart Search: Fast, Expert, and Vision. The lab is reportedly developing a minimum of three large models on a domestic AI chip-based computing platform.
However, Huawei compatibility work has already cost DeepSeek time. R2 training ran into unstable performance, slower chip-to-chip connectivity, and limitations in Huawei’s CANN software, forcing a reversion to Nvidia hardware earlier this year. CANN (Compute Architecture for Neural Networks) is a heterogeneous computing architecture designed to optimize AI performance across Huawei Ascend hardwaretoolkitV4’s multi-modal expansion, particularly the Vision mode, raises the hardware bar further, along with the capital requirements that come with it. Outside money, on that timeline, looks less like an ideological concession than an operational necessity.
As a result, staff retention pressures compound the math. Xiaomi and ByteDance can match the salaries of frontier-lab peers in a way High-Flyer’s research budget cannot, and each departure multiplies the cost of the next hire. Fresh capital, on DeepSeek’s side, becomes a lever to close that gap before V4 ships.
A Benchmark Against US Peers and the Gap to Chinese Rivals
Valuation context tells its own story. OpenAI said in February 2026 it was raising $110 billion at a valuation of about $840 billion, a figure that climbed to roughly $852 billion by early April.
Meanwhile, Anthropic raised in February at a $380 billion valuation. Venture investors subsequently floated offers valuing the company as high as $800 billion. DeepSeek’s target therefore sits at a tiny fraction of its US peers’ marks.
Inside China, however, the same figure would position DeepSeek among the country’s most valuable AI startups, a peer grouping that now includes MiniMax, which recently went public in Hong Kong, alongside Moonshot, Zhipu, and Baichuan. Gaps with US peers reflect both the harder path to commercial revenue inside China’s regulatory perimeter and the discount investors apply to labs cut off from the premier Nvidia chips.
Fresh capital is expected to fund scaling model training, inference-efficiency work, and commercialization of AI products. DeepSeek has built its brand on high-performance yet cost-efficient large language models. R1’s release was specifically credited with reducing training costs, and the commercialization push suggests the lab plans to press that efficiency advantage into enterprise and consumer channels where revenue, not just research acclaim, is the scoreboard.
For prospective backers, the pitch is asymmetric risk: a lab capable of frontier results on a much smaller capital base, priced at a valuation that leaves room to expand. For DeepSeek, the incoming money buys time, chips, and talent without forcing an immediate commercial repositioning, though the governance trade-offs described above suggest any such reprieve will be temporary rather than structural, and investor pressure will likely intensify as V4 nears launch and later rounds loom into view.
From “No Fundraising” to a Ten-Figure Round
The reversal is sharp. DeepSeek’s R1 drew wide attention across Silicon Valley and Wall Street in 2025, briefly rattling NVIDIA’s market value and prompting a wave of analysis about whether a sanctions-constrained Chinese lab had matched frontier US performance at a fraction of the cost. That episode hardened the lab’s identity as a research-first outfit, and recent reporting continued to describe DeepSeek as fully funded by High-Flyer with no immediate fundraising plans.
As recently as February 2026, prevailing coverage characterized DeepSeek as having no fundraising intentions. However, current talks mark a sharp shift from that earlier posture and the first confirmation that the independence narrative is yielding to the capital-intensity of frontier AI. OpenAI’s approach to Middle East sovereign wealth funds earlier in the year underscored the scale of run-rate every serious AI lab now faces. DeepSeek, founded in 2023, is now confronting a version of the same math.
Meanwhile, talks are ongoing and have not been finalized. Watch points over the coming weeks include an official DeepSeek statement, V4’s launch timing, the specific identity and mix of Chinese sovereign versus strategic capital participating in the round, the governance terms attached to any lead investor, and whether the final valuation holds at its reported mark or drifts upward as the fundraise progresses.
For Liang and the engineers who joined a self-funded lab, the decision pending over the next several weeks is whether accepting outside capital closes the gap to frontier peers or quietly ends the independent research culture that defined DeepSeek’s first three years. Each of those variables will signal how far the lab has traveled from its self-funded origins toward the commercial, capital-dependent shape every frontier AI lab now appears to share.
