距离DeepSeek首轮创下行业纪录的510亿元融资落地，仅仅过去不到四个月，这家大模型领军企业的第二轮超级融资又行至交割尾声。据彭博社10月22日最新报道，DeepSeek本轮融资规模已锁定在约1000亿元人民币，投前估值维持在约5000亿元。知情人士透露，这笔交易在推进过程中经历剧烈的动态调整，最初计划募集金额为500亿元，随后在各方产业资本与老股东的角逐下两次调高目标，最终融资金额直接翻倍至千亿量级。
Less than four months after DeepSeek's first round set an industry record at 51 billion yuan, the large model leader's second super funding round is nearing closing. According to Bloomberg's latest report on October 22, DeepSeek's current round has locked in about 100 billion yuan, with pre-money valuation maintained at about 500 billion yuan. People familiar with the matter revealed the deal underwent dramatic dynamic adjustments during advancement—initially planned at 50 billion yuan, then raised twice amid competition from industrial capital and existing shareholders, ultimately doubling to the hundred-billion level.
在资金盘切分上，新老股东的出资呈现出对半均分的态势，老股东腾讯与宁德时代各自追加约100亿元，成为本轮基石投资方；而新进方则全面倒向硬科技阵营，包括石溪资本、中芯聚源、CPE源峰、君联资本以及汽车巨头吉利等。若将本轮千亿规模与6月底由梁文锋及核心机构敲定的510亿元首轮融资叠加，DeepSeek在短短百余天的时间跨度内累计吸纳资本已超1500亿元，创下全球人工智能领域资金吞吐速度的全新纪录。
On capital allocation, new and existing shareholders split roughly evenly. Existing shareholders Tencent and CATL each added about 10 billion yuan, becoming cornerstone investors this round; new entrants tilted fully toward the hard-tech camp, including Shixi Capital, SMIC Juyuan, CPE Yuanfeng, Legend Capital, and automotive giant Geely. Combining this hundred-billion round with the 51 billion yuan first round finalized by Liang Wenfeng and core institutions at end-June, DeepSeek has absorbed over 150 billion yuan in just over a hundred days, setting a new global AI record for capital throughput speed.
与外媒消息遥相呼应的，是A股市场在资金通道端的密集动向。近日，完美世界密集披露投资进展公告，宣布两只关键外部专项基金已全额募集完毕，其中天津砺思星深基金规模达75.20亿元，拾象尚势基金规模达36.50亿元，两只基金合计吸纳外部资金111.70亿元。在这两只大额专项基金背后，浮现出多家A股上市公司的身影。完美世界分别认缴5000万元和1亿元，以1.5亿元总出资完成双通道押注；世纪华通则通过全资子公司认缴拾象尚势6.5亿元；算力服务商利通电子此前亦公告出资9500万元认购砺思星深份额。
Echoing foreign media reports, A-share markets show dense capital channel activity. Recently, Perfect World密集 disclosed investment progress announcements, declaring two key external special funds fully raised—Tianjin Lisi Xingshen Fund at 7.52 billion yuan and Shixiang Shangshi Fund at 3.65 billion yuan, together absorbing 11.17 billion yuan in external capital. Behind these two large special funds appear multiple A-share listed companies. Perfect World subscribed 50 million and 100 million yuan respectively, completing dual-channel bets with 150 million total; Century Huatong subscribed 650 million yuan in Shixiang Shangshi through a wholly-owned subsidiary; compute service provider Litong Electronics previously announced 95 million yuan to subscribe Lisi Xingshen shares.
与此同时，中国证券投资基金业协会公示信息显示，砺思星深已于2026年9月15日完成私募投资基金备案，备案编码为SEX397。据知情人士透露，砺思星深与拾象尚势两只基金均是DeepSeek本轮融资的外部资管通道，随着相关基金相继完成实缴封账，外部社会资本向DeepSeek核心资产池注入的渠道已基本落锁。砺思资本之所以值得重点关注，在于它在DeepSeek的上一轮融资中，已经成为外部SPV重仓DeepSeek的核心通道。这家由前红杉中国合伙人曹曦与前博裕资本合伙人Tim Wang于2021年创办的硬科技投资机构，投资覆盖AI大模型、国产GPU、机器人及AI基础设施等领域，月之暗面、DeepSeek、宇树科技等均出现在其投资版图中。
Meanwhile, Asset Management Association of China public information shows Lisi Xingshen completed private fund filing on September 15, 2026, with filing code SEX397. People familiar with the matter revealed both Lisi Xingshen and Shixiang Shangshi are external asset management channels for DeepSeek's current round; as related funds complete paid-in closing, channels for external social capital injecting into DeepSeek's core asset pool are basically locked. Lisi Capital deserves attention because in DeepSeek's previous round it already became the core channel for external SPVs heavily investing in DeepSeek. This hard-tech investment institution, founded in 2021 by former Sequoia China partner Cao Xi and former Boyu Capital partner Tim Wang, covers AI large models, domestic GPUs, robotics, and AI infrastructure, with Moonshot AI, DeepSeek, and Unitree all appearing in its investment map.
更为关键的资本化信号也在同步浮现。彭博社援引知情人士消息指出，在推进本轮千亿级融资的同时，DeepSeek正同步筹备科创板IPO事宜，计划最快于今年底递交上市申请，争取在2027年正式挂牌。前高瓴创投合伙人严文韬履新出任CFO，亦为这场涉及千亿资金配置与多层穿透监管的资本运作补齐了专业操盘手。
A more critical capitalization signal is also emerging. Bloomberg citing people familiar with the matter noted that while advancing this hundred-billion funding round, DeepSeek is simultaneously preparing for STAR Market IPO, planning to submit listing application as early as end of this year, striving for official listing in 2027. Former Hillhouse Ventures partner Yan Wentao has taken up the CFO position, filling in a professional operator for this capital operation involving hundred-billion capital allocation and multi-layer penetration regulation.
完美世界、世纪华通和利通电子的入局，让DeepSeek本轮融资的产业协同色彩更加具象。完美世界与世纪华通两家游戏龙头合计向相关专项基金认缴出资8亿元，算力服务商利通电子则认缴9500万元。三家上市公司业务各异，却清晰切中了下一代游戏内容生成与算力基础设施等核心环节。今年8月，DeepSeek曾对外发起一轮游戏开发和3D渲染领域的高薪招聘，引发游戏行业广泛关注，其岗位包括Code Agent数据工程师、Web 3D与渲染方向，要求候选人不必精通模型训练，但要真正做过游戏，重点看重游戏开发者在工程方法、复杂交互以及多年积累的行业研发经验。
Perfect World, Century Huatong, and Litong Electronics' entry makes DeepSeek's industrial collaboration more concrete. The two gaming leaders Perfect World and Century Huatong together subscribed 800 million yuan to related special funds; compute service provider Litong Electronics subscribed 95 million yuan. The three listed companies have different businesses but clearly hit core links including next-generation game content generation and compute infrastructure. In August this year, DeepSeek launched a high-salary recruitment round in game development and 3D rendering, drawing widespread gaming industry attention; positions included Code Agent data engineer and Web 3D/rendering directions, requiring candidates to "not necessarily be proficient in model training but to have truly made games," valuing game developers' engineering methods, complex interaction, and years of accumulated industry R&D experience.
从更广泛的投资人名单看，石溪资本、中芯聚源等半导体产业资本的加入，则为本轮融资注入了深层的硬科技底色。此前，DeepSeek在有限算力条件下的架构优化广受行业认可，尤其是围绕华为昇腾等国产算力的紧密合作与适配优化。而伴随中芯系等半导体核心资本的入场，其未来在国产芯片底层适配与定制化算力集群搭建，甚至是自研芯片的动作值得持续追踪。整体来看，DeepSeek的资本来源正在从原有的互联网大厂与纯财务投资机构，进一步横向扩展至半导体制造、汽车工业以及能源硬件等广泛的实体产业阵营。
From a broader investor list, semiconductor industrial capital including Shixi Capital and SMIC Juyuan injects deep hard-tech底色 into this round. Previously, DeepSeek's architecture optimization under limited compute was widely recognized, especially close cooperation and adaptation optimization around domestic compute including Huawei Ascend. With core semiconductor capital like SMIC entering, its future moves in domestic chip底层 adaptation, customized compute cluster building, and even in-house chip development deserve continuous tracking. Overall, DeepSeek's capital sources are expanding horizontally from original internet giants and pure financial investors to broad实体 industry camps including semiconductor manufacturing, automotive industry, and energy hardware.
除了投资人结构的变化，创始人梁文锋近期密集布局具身智能的动向，也为这轮超级融资揭示了新的战略走向。此前，多家媒体报道称，梁文锋近期与多位具身智能创业者交流，探讨机器人、大模型与现实世界交互等方向。知情人士透露，DeepSeek正在考虑是否进一步进入这一领域，可能的路径包括自建团队研发，或通过投资相关企业切入。而在投资方面，DeepSeek已经押注三大具身企业。宇树科技已披露，DeepSeek此前已斥资约1.41亿元人民币入股，切入机器人硬件、运动控制与通用开发测试资源。Winverse AI拟领投，由腾讯前AI科学家胡瀚创立，DeepSeek拟领投其种子轮融资，投后估值约4亿美元。俞涛的创业公司方向为大模型与具身智能结合，两轮估值分别约为7000万美元和2亿美元，近日有媒体曝出已获得DeepSeek及一线基金参与投资。
Beyond investor structure changes, founder Liang Wenfeng's recent dense布局 in embodied intelligence reveals new strategic direction for this super funding. Multiple media reports said Liang Wenfeng recently exchanged with multiple embodied intelligence entrepreneurs on robots, large models, and real-world interaction. People familiar with the matter revealed DeepSeek is considering further entering this field, possible paths including building teams for in-house R&D or entering through investing in related companies. On investment, DeepSeek has already bet on three embodied enterprises. Unitree Technology disclosed DeepSeek previously invested about 141 million yuan, entering robot hardware, motion control, and general development testing resources. Winverse AI planned lead investment, founded by former Tencent AI scientist Hu Han, with DeepSeek planning to lead its seed round at about $400 million post-money valuation. Yu Tao's startup direction combines large models with embodied intelligence, with two rounds valued at about $70 million and $200 million; recent media reports say it has received DeepSeek and first-tier fund participation.
具身智能涉及真实世界数据交互、多物理场仿真模拟以及高精度的运动控制，对算力资源、工程测试环境以及硬件本体有极高的持续消耗。这在某种程度解释了DeepSeek本轮融资规模为何在短时间内大幅上调至千亿级别，也揭示了其战略野心，从云端虚拟代码的模型推演，正式拓界迈向具备物理交互能力的现实智能实体。
Embodied intelligence involves real-world data interaction, multiphysics simulation, and high-precision motion control, with extremely high continuous consumption of compute resources, engineering test environments, and hardware bodies. This partly explains why DeepSeek's current round was significantly raised to the hundred-billion level in a short time, and reveals its strategic ambition—expanding from cloud virtual code model inference toward real intelligent entities with physical interaction capabilities.
