DeepSeek valuation jumps as Tencent and Alibaba reportedly in talks
On April 22, reports circulated that Tencent and Alibaba are negotiating investments in DeepSeek, with market sources saying the company’s target valuation has been raised from at least USD 10 billion to more than USD 20 billion (approximately RMB 136.5 billion); Alibaba’s U.S. shares surged in pre‑market trading, at one point rising over 2% before settling at a 0.64% gain by press time, and DeepSeek has not publicly responded. The Information, citing four people familiar with the matter, reported that talks are ongoing and that both the final valuation and financing size remain subject to change; if completed, this would mark DeepSeek’s first external fundraising and a major reversal of founder Liang Wenfeng’s long‑standing stance against outside capital. Previously, DeepSeek had been funded exclusively by hedge fund 幻方量化 and had declined offers from top venture capital firms and tech giants, and earlier market reports of Alibaba interest were denied by DeepSeek.
The company’s latest technical moves have also drawn attention: on April 22 DeepSeek upgraded its official API to support a 1 million‑token context window, a substantial increase from the prior 128k, an update that coincides with expectations that DeepSeek will release its next‑generation flagship model V4 in late April. Public information indicates DeepSeek was founded in July 2023 and gained global attention with its R1 model in January 2025; V4 is described as the first major version update since R1, natively supporting image, video and text generation and adopting a MoE architecture with roughly one trillion parameters while activating about 37 billion parameters per token to keep inference costs comparable with V3. DeepSeek has reportedly completed deep optimization and adaptation with domestic AI chipmakers including Huawei and Cambricon, focusing on compute scheduling and inference efficiency, and recruitment listings show the company is hiring data center operations engineers in Ulanqab, Inner Mongolia, its first direct hires for infrastructure operations—an indication to some observers that V4 has entered large‑scale deployment.
Industry context underscores the stakes: global AI model iteration is accelerating and training and operating costs are rising, and several core DeepSeek researchers have moved to major domestic firms in recent months, including Luo Fuli, who has joined Xiaomi as head of the MiMo team. V4’s original February launch timeline was delayed multiple times due to engineering challenges, and external capital would provide financial resources for procuring compute and stabilizing talent. Market comparisons have been cited in valuation discussions: DeepSeek reportedly referenced the valuation of 月之暗面 as a partial benchmark, and peers MiniMax and Zhipu both listed in Hong Kong in January and have attracted strong investor interest, with Zhipu’s market capitalization reaching HKD 439.6 billion and MiniMax HKD 291.1 billion as of April 22.
DeepSeek’s April 22 API upgrade reportedly expands the knowledge base to May 2025 and enables offline responses accurate to April 2025 news events; the upgraded model remains text‑ and voice‑centric without visual input, and the company had earlier launched an expert mode and file upload support on April 8. According to public reporting, V4 uses an efficiency‑first design philosophy and, if deeply adapted to Huawei’s Ascend chips as well as other domestic accelerators, could alter competitive dynamics in the AI ecosystem. Nvidia CEO Jensen Huang recently warned that China’s large compute resources and top AI talent could offset hardware gaps, and that close adaptation between major domestic models and domestic chips would have significant implications. DeepSeek has not confirmed these external comments.
