Disclaimer: deepseek.ai is an independent community and fan site, not officially affiliated with DeepSeek AI. Figures below are drawn from Bloomberg, TechNode, Pandaily and CnEVPost reporting and remain subject to confirmation.
Round Timeline: $10B Rumor → ~$7.4B Close
The story evolved substantially between May and July 2026:
- 22 May 2026 — Bloomberg: reports the round is "advancing" toward roughly $10B, with founder Liang Wenfeng framing the raise around Artificial General Intelligence rather than enterprise revenue.
- 4 June 2026 — TechNode: negotiations reported at around $7B, with Tencent and CATL named as backers.
- Mid-June 2026: the round closed at approximately 50 billion yuan (~$7.4B), valuing DeepSeek near $50B. The "50 billion yuan round" and the "$7.4B round" are the same financing in two currencies. See our deep-dives on the 50 billion yuan structure and the founder-centric deal geometry.
- 25 July 2026 — Bloomberg: a second round, reportedly targeting a $74B valuation, was paused.
The original $10B number reflected the fundraising target, not the final print. A lab that opens with a $10B ask and closes at ~$7.4B is still doing an extraordinary round by any historical measure, but the pitch of "AGI at any price" was moderated by real due diligence.
Who Put In What
The most unusual feature of the round is the founder's own participation. Reported allocations:
- Liang Wenfeng (founder): 20 billion yuan — roughly 40% of the entire round.
- Tencent: 10 billion yuan.
- CATL: 5 billion yuan (~$740M).
- NetEase: 3 billion yuan.
- JD.com: 3 billion yuan.
- Remaining balance: spread across additional domestic strategic and financial investors.
External investors are reported to hold no voting rights and to be bound by a five-year lock-up. Liang retains control with a reported ~39% stake. In other words: the capital arrived, the governance did not change.
AGI as a Fundraising Thesis
Most frontier labs avoid the AGI word in investor decks. Western counterparts prefer "enterprise AI" or "productivity tooling" — language that is easier to underwrite. DeepSeek did the opposite, and that framing matters:
- Long-horizon capital. An AGI pitch filters out investors chasing 24-month ARR curves and attracts patient capital for multi-year research bets.
- Open-weight credibility. Continuing open releases — the same strategy behind R2 and the V4 compressed-attention work — only makes sense if monetisation is explicitly deprioritised.
- Governance without dilution of control. No voting rights plus a five-year lock-up means the research agenda stays where it was.
Why Investors Believed
Through most of 2023 and 2024, foreign capital treated DeepSeek as a curiosity. Low-cost open models forced the industry to reconsider training economics. The question stopped being "can a Chinese lab catch up?" and became "do you actually need OpenAI-scale spending to reach the frontier?"
- V4's 1.6T-parameter MoE architecture demonstrating efficient sparse compute.
- Aggressive cost-per-token undercutting on the DeepSeek API.
- Sustained open-weight releases turning the community into a free distribution and validation network.
How DeepSeek Differs from Western Labs
| Theme | DeepSeek Positioning |
|---|---|
| Monetisation | Explicitly secondary to research |
| Model access | Open-source / open-weight emphasis |
| Capital strategy | Self-funded historically, now scaling externally |
| Stated ambition | AGI |
| Competitive angle | Efficient training, lower cost per FLOP of capability |
What to Watch Next
- The paused second round. Bloomberg reported on 25 July 2026 that a follow-on raise at a ~$74B valuation was put on hold. Whether it restarts, and at what price, is the clearest signal on how investors currently value the lab.
- Next model release: a raise of this size is typically followed within ~6 months by a flagship model — see our V4 tracker.
- Pricing behaviour: the permanent V4-Pro price cut and the retirement of the legacy model IDs show how the lab is now managing serving cost against capital.
Bottom Line
DeepSeek didn't raise ~$7.4B to ship a better chatbot. Even at the lower-than-rumoured close, the pitch — pursue AGI, stay open-weight, prioritise research over revenue — reframes what the next phase of the frontier-AI race looks like. Whether the AGI framing holds, the round confirms that the frontier is no longer a one-continent story.