DeepSeek's valuation hits $20 billion in first funding round, following the poaching of four core researchers

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1737 words

DeepSeek has opened up to external financing for the first time. Sources familiar with the matter reveal that valuation negotiations have climbed from $10 billion to $20 billion. Meanwhile, four core researchers, including DeepSeek-R1 lead developer Guo Daya, have successively departed for ByteDance, Tencent, Xiaomi, and DeepRoute.ai. This talent drain is seen as the direct catalyst for this shift in financing strategy. (Previous coverage: DeepSeek V4 was released in late April; Liang Wenfeng internally confirmed a million-token context window, running entirely on Huawei Ascend 950 PR chips.) (Background: DeepSeek’s 545% profit margin and "valuation exceeding $10 billion"? Bloomberg remains skeptical, calling it a theoretical figure and noting that users may not necessarily be willing to pay.) DeepSeek has always been an outlier in the Chinese AI scene: independently funded by its parent company High-Flyer Quant, with founder Liang Wenfeng repeatedly rejecting external investment in public. Research independence was considered one of its core competitive advantages, but this stance is now softening. Sources indicate that DeepSeek recently initiated its first external financing negotiations. The initial valuation was approximately $10 billion, but as investor demand heated up, the negotiated price was pushed to $20 billion. Tencent reportedly proposed a 20% stake, which DeepSeek declined, citing a desire to retain control. It is not just external pressure driving this pivot, but also internal talent attrition. Over the past few months, the DeepSeek research team has seen rare talent turnover: - DeepSeek-R1 lead creator Guo Daya jumped to ByteDance’s Seed AI team, with rumors of an annual salary reaching 100 million RMB (approximately $14 million). - V2 core developer Luo Fuli joined Xiaomi’s MiMo spatial intelligence team in November 2025. - Wang Bingxuan moved to Tencent. - Ruan Chong, who was in charge of the Janus-Pro multimodal project, joined the autonomous driving company DeepRoute.ai in January 2026. These four names represent the most core technical paths DeepSeek has pursued over the last two years: reasoning, pre-training, and multimodality. Guo Daya’s departure is particularly symbolic. R1 is the model that ignited global discussion around DeepSeek. The departure of its lead creator for a 100 million RMB salary directly reveals that, in the face of well-funded tech giants, the appeal of "research independence" has a ceiling. High-Flyer Quant achieved a 56.6% return in 2025 and is technically capable of continuing to self-fund. However, the cost of retaining top researchers is clearly not just about money; it involves compensation structures, stock option design, and the long-term appeal of working under a private quantitative fund. Changes in the external competitive landscape have brought more direct benchmarking pressure to DeepSeek. Zhipu AI listed on the HKEX on January 8, 2026, becoming the first large language model company in China to complete an IPO. The following day, MiniMax went public (stock code 00100), with its share price rising 109.1% on the first day. Moonshot’s financing speed has been even more staggering: it completed three rounds in the first three months of 2026. In January, it raised $500 million in a Series C round at a $4.3 billion valuation; in February, the valuation rose to $10 billion with a $700 million follow-on investment from Alibaba and Tencent; in March, it completed another $1 billion financing, reaching a valuation of $18 billion. In three months, its valuation more than quadrupled. There is a common logic behind these figures: the valuations of top-tier Chinese AI companies are being rapidly repriced by capital. If DeepSeek continues to stand outside this pricing game, the difficulty of attracting and retaining talent will only continue to rise. What competitors are gaining is not just money, but a more competitive set of talent incentive tools: stocks, options, and IPO expectations. This is a structure difficult to replicate by relying solely on funding from a quantitative parent company. At the same time the financing news emerged, DeepSeek officially updated its API documentation. According to the official DeepSeek pricing page, the DeepSeek-V4-Pro model API is offering a limited-time 75% discount, valid until May 5, 2026. The discounted pricing is as follows: cache hit input (i.e., overlapping with previous queries, read directly from cache) is 0.25 RMB per million tokens; cache miss input is 3 RMB; output is 6 RMB per million tokens. This pricing move can be

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-22

DeepSeek valuation surpasses $20 billion! Foreign media reports Tencent and Alibaba are competing to invest in the first round of financing.

Similarity 170%關鍵字億美元/deepseek同分類 zh

2026-04-26

OpenAI pursues the Michelin-star route, DeepSeek adopts a pre-made meal strategy; the battle for AI pricing power between China and the US intensifies.

Similarity 120%關鍵字 deepseek同分類 zh

2026-04-26

ComfyUI valuation hits $500 million: AI image generation moves beyond "gacha" mechanics, empowering creators to control every frame of output

Similarity 120%關鍵字億美元同分類 zh

2026-04-24