Core Viewpoint - The A-share computing power sector has seen significant gains, driven by the launch of DeepSeek's "DeepSeek V3.1" model, which is optimized for domestic chip architectures, indicating a potential for large-scale adoption of domestic chips in AI applications [1][3][10]. Group 1: Market Performance - The stock of Cambricon (寒武纪) surged by 20%, reaching a market capitalization of 520 billion RMB [1]. - Semiconductor stocks, including SMIC, saw substantial increases, with SMIC's A-shares rising by 14.19% and H-shares by 10.06%, marking the highest single-day gain since October of the previous year [1]. - The Wind semiconductor index rose by 7.31%, reaching its highest level since April 2022, while the Wind technology index increased by 3.07%, setting a new historical high [4][6]. Group 2: DeepSeek V3.1 Launch - DeepSeek's "DeepSeek V3.1" was released on August 21, featuring parameters designed for the next generation of domestic chips, which has sparked expectations for mass production of domestic AI chips [3][10]. - The model aims to enhance reasoning performance while maximizing the theoretical computing power and energy efficiency of domestic chips, despite their current inferiority to NVIDIA GPUs [8][10]. Group 3: Industry Implications - The release of DeepSeek V3.1 is seen as a catalyst for the domestic AI chip ecosystem, with analysts suggesting that it could accelerate the adoption of domestic computing power solutions [10]. - Companies like Huawei and Cambricon are making strides in the domestic chip market, with Cambricon's "Siyuan 590" chip reportedly achieving performance levels close to NVIDIA's A100 in certain tasks [11]. - The Ascend 910D chip from Huawei is also highlighted for its potential to surpass NVIDIA's H100 in theoretical computing power, particularly in localized applications [11].

Core Insights - DeepSeek is leveraging NVIDIA's GPU capabilities while adapting to domestic chips, potentially reducing memory usage by up to 75% and decreasing reliance on imported advanced GPU chips [1][34][39] - The release of DeepSeek V3.1 marks a significant step towards the Agent era, showcasing a hybrid reasoning architecture that supports both thinking and non-thinking modes [3][10] - The upgrade enhances DeepSeek's efficiency, allowing it to answer questions using fewer tokens and less time, improving user experience and economic considerations [4][6] Technical Developments - DeepSeek V3.1 utilizes UE8M0 FP8 scale parameter precision, which significantly reduces memory and bandwidth requirements, thus improving training and inference efficiency [11][15][30] - The model has undergone extensive retraining with an additional 840 billion tokens, achieving a context length of 128k [7][10] - The API Beta interface now supports strict function calling, enhancing reliability and usability in enterprise applications, aligning with trends seen in other major AI companies [8][9] Market Implications - The advancements in DeepSeek's technology are expected to invigorate the Chinese capital market, reflecting a growing focus on technological self-sufficiency [2] - As domestic GPU manufacturers adopt FP8 precision, the demand for NVIDIA's H20/B30 chips may decline, especially if the next generation of domestic GPUs can efficiently run large models [36][38] - The shift towards UE8M0 and ultra-low precision training could lead to a gradual reduction in reliance on NVIDIA's ecosystem, fostering a more independent Chinese AI chip and model landscape [42] Competitive Landscape - Despite the innovations from DeepSeek, NVIDIA maintains its competitive edge with superior bandwidth, interconnect capabilities, and a robust software ecosystem [40][41] - The ongoing evolution of low-precision digital representation technology, as exemplified by DeepSeek's UE8M0, may accelerate the development of next-generation domestic chips [39][42] - The industry is witnessing a potential shift where companies may prioritize domestic solutions over NVIDIA's offerings, particularly in cost-sensitive scenarios [38][42]

Group 1 - The A-share computing power sector has become a leader, with notable stocks like Cambricon Technologies reaching a market value of 520 billion RMB after a 20% surge [1] - The release of DeepSeek V3.1, which is optimized for domestic chip structures, signals potential large-scale adoption of domestic chips in AI applications [1][2] - The semiconductor market is experiencing a speculative surge, with the Wind semiconductor index rising by 7.31%, the highest since April 2022 [2] Group 2 - DeepSeek V3.1 integrates deep thinking and fast thinking, aligning with recent trends in AI model development [3][4] - The new model's parameters are designed for next-generation domestic chips, aiming to maximize theoretical computing power and efficiency despite being slightly inferior to Nvidia GPUs [4][5] - The performance of domestic chips like the Cambrian's Siyuan 590 and Huawei's Ascend 910D is approaching that of Nvidia's A100, indicating a competitive edge in specific applications [6] Group 3 - The release of DeepSeek V3.1 is seen as a precursor to the domestic chip market's growth in computing power [7] - Analysts caution that uncertainties regarding model compatibility with domestic chip manufacturers and development progress may pose risks [7]

8月21日晚间，杭州深度求索公司（DeepSeek）悄然于大模型资源网站HuggingFace上线 "DeepSeekV3.1" 版本。同时，该公司宣称其模型版本使用适配国产芯片结构的参数精度，暗示国产芯 片未来有望得到大规模使用。 分析人士认为，在国际贸易保护主义背景下，国际芯片产业链供应链的不确定性犹存。DeepSeek-V3.1 发布，以及腾讯业绩会透露GPU数量足够等信息，均利好AI算力国产化和本土AI芯片发展，国产算力 板块再度迎来催化。 国产芯片+模型的"弯道超车" 21世纪经济报道记者赵云帆 8月22日，A股算力板块再度成为"执牛耳者"。截至收盘，本轮行情的"股 王"寒武纪以20%涨停报收，总市值来到惊人的5200亿人民币；晶圆龙头中芯国际A/H股分别上涨 14.19%和10.06%，其中A股创下去年10月以来单日涨幅之最。光模块"双雄"新易盛、中际旭创续创新 高，并分别上涨6.19%和5.16%；国产IDC龙头中科曙光亦以涨停报收，市值接近去年11月高点。 8月21日晚间，深度求索在官方订阅号发布文章《DeepSeek-V3.1 发布，迈向 Agent 时代的第一步》， 明确指出"DeepSe ...

阿里发布编程平台Qoder；可灵AI推出基于2.1模型的全新首尾帧功能；Meta与谷歌云签署6年超100亿美 元合作协议。 ①特斯拉牵手豆包大模型与DeepSeek，均通过火山引擎接入； ④蚂蚁集团与北医三院达成战略合作，共建"AI医疗联合实验室"； ⑤可灵AI推出基于2.1模型的全新首尾帧功能； ⑥kimi-k2-turbo-preview模型输出速度已提升至每秒60 Tokens，最高可达每秒100 Tokens； ⑦Meta与谷歌云签署为期6年、价值超100亿美元合作协议，聚焦AI基础设施领域； ⑧Anthropic据悉正在洽谈筹集最高达100亿美元新一轮融资。 ②DeepSeek-V3.1上线火山方舟； ③阿里发布编程平台Qoder，可实现AI自主研发； ...

A股市场延续上攻态势，沪指突破3800点，续创十年新高。 以芯片为代表的国产算力，成为托举这波科技牛市的主力。 寒武纪直接20cm涨停，市值突破5000亿，股价超过1200元。 科创板芯片股全数上涨，科创芯片指数居然涨了10.05%，领涨市场主要芯片指数；跟踪该指数的科创芯片ETF（588200）飙涨11.92%。 | | 代码 | 名称 | 涨跌幅 ▼ | | --- | --- | --- | --- | | | 000685 | 科创芯片 | 10.05% | | 2 | 980017 | 国证芯片(CNI) | 9.82% | | 3 | 931865 | 中证未含 | 9.52% | | 4 | 000682 | 科创信息 | 9.44% | | 5 | 932066 | 未含体行业精选 | 9.08% | | 6 | H30007 | 芯片产业 | 8.87% | | 7 | 990001 | 中华未曾体验片 | 8.87% | | 8 | 932087 | 集成电路 | 8.69% | | 9 | 000688 | 科创50 | 8.59% | | 10 | H30184 | 中证全指半导体 | ...

今日（8.22），算力板块大涨，同类规模领先的云计算ETF汇添富（159273）大涨2.59%。资金面上，云计算ETF汇添富（159273）溢价高企达0.34%，盘中 再度吸金超1600万元，截至昨日，上市12天已有10天获净流入，累计"吸金"超3亿元！同类规模、流动性持续领先！ | 分时 多日 1分 5分 15分 30分 60分 ▼ | | | | | | 综合屏 F9 前复权 超级叠加 画线 工具 谷 ② | >> | | --- | --- | --- | --- | --- | --- | --- | --- | | 159273.SZ[云计算ETF汇添富] 2025/08/22 收 1.147 幅２.59％(0.029) 开１.118 高１.148 低１.112 坪 1.135 四 | | | | | | | | | MA5 1.123↑ MA10 1.082↑ | 2025/08/06-2025/08/22(13日)▼ ■ | | | | | | | | 1.148 | | | | | | | 1.140 | | | | | | | | | 1.110 | | | | | | | | | 1.08 ...

注：智通财经VIP为内容资讯产品，并非投资建议。以下内容仅为资讯价值展示非对相关公 司的推荐建议，非未来走势预测。投资有风险，入市需谨慎。 8月1日-31日，智通财经会员月震撼来袭，海量VIP资讯体验+年度最惠价，多福多财，狂欢一整月~ 转自：智通财经 活动期间签到有礼，价值最高800元代金券/128元等单篇解锁券签到即送，每周一签到福利 翻倍！签到不缺席，最高可领1850券包~ 「VIP资讯年度最惠价来了」智通财经VIP订阅享超值特惠，选订1个月、3个月限时85折； 选订6个月、12个月限时8折，最高可省2778元。； 若您是智通财经App新注册用户，立即领268新手券，新用户完成首次订阅包时段服务，再 加赠VIP资讯时长服务！ 话不多说，一起来看看吧~→直达活动现场 直结有数 4 / 7 | 流停 /9 VS 跌停2 [3 SC 2025/08 LL 上证指数 3825.76 +1.45% | 深圳成指 | 12166.06 | | --- | --- | | +2.07% | | 创业板指 2682.55 +3.36% 7 今日行情 今日市场震荡走高, 沪指站上 3800点, 科创 50 涨超 8% 创 ...

本文来自微信公众号：凤凰网科技 （ID：ifeng_tech），作者：姜凡，编辑：董雨晴，题图来自：AI生成 DeepSeek在8月21日正式发布V3.1，一则官方补充留言激起千层浪。 是的，因为官方在V3.1的发布文章里把有关过程算力支持的描述写得太隐晦，又自己在评论区补充了一句。 一夜过去，短短的一句话，A股市场集体"开香槟"。据智通财经报道，全市场超2800只个股上涨。从板块来看，算力股全线爆发，云天励飞等多股涨停。 芯片股集体大涨，寒武纪涨停创历史新高，中芯国际大涨14%，海光信息涨停。 资本市场的反应并不是对一句口号的追捧，而是看到了国产芯片在算力赛道上真正切入国际前沿的可能。DeepSeek的这句话，不仅被视作一次"技术换 挡"的信号，更是国内算力产业链罕见的共振时刻。 DeepSeek这枚炸弹，到底是什么？ 要理解"UE8M0 FP8"为何能引发如此大的关注，必须将其拆解为两个部分：作为本体的"FP8"和作为灵魂的"UE8M0"。FP8本身是一种将数字压缩到仅用8 位（bit）来存储的浮点格式，旨在用更少的数据位宽降低AI大模型训练和推理中对显存带宽的空前压力。 而DeepSeek的模型一直使用 ...

2025年8月19日，DeepSeek在Hugging Face平台正式开源新一代大模型DeepSeek-V3.1，并于8月21日面向全球开发者发布完整版本。官方将其定位为"迈向 Agent（智能体）时代的第一步"，通过Post-Training优化显著提升了工具使用与智能体任务能力，同时宣布API价格调整及国产芯片深度适配方案，引发行 业震动。 01 技术升级：更强记忆、更优架构、更稳工具调用 更强大的记忆能力：相比V3版本64K的上下文窗口，新一代模型将处理能力提升至128K，可处理相当于30万汉字的长文本，为长文档分析、复杂代码生 成和深度多轮对话提供了巨大支持。这一升级使得模型在工具调用、函数执行及复杂推理任务中的表现提升约40%，特别是在数学证明、编程协作与多模 态任务规划方面展现出色能力。 架构设计升级，有助于智能体调试和优化：DeepSeek-V3.1实现了从单一推理模式向双模式架构的升级。原来的架构相对单一，更多是基于传统的 Transformer架构进行优化，主要侧重于文本生成和基础问答等功能，对于智能体所需的复杂任务处理、多步骤推理以及工具协同等方面的支持较为有限。 而此次升级后的API体 ...