AI科技热点日报 | 2026年07月24日

AI科技热点日报 | 2026年07月24日

综合整理自 TechCrunch、新浪财经、IT之家、MSN科技等科技财经媒体当日报道，聚焦 AI、机器人、芯片、前沿技术、行业金融动态。

📌 今日摘要

月之暗面（Moonshot AI）发布全球最大开源模型Kimi K3，以2.8万亿参数刷新开源大模型记录，被华尔街称为继DeepSeek-R1后的第二次"DeepSeek冲击"。欧洲首家人形机器人独角兽SKL Robotics获1.52亿美元融资，投后估值达13.5亿美元。Meta高调入局AI代码大模型赛道，与OpenAI、Anthropic展开正面竞争。中国AI大模型市场规模有望在2026年突破700亿元，印度AI初创融资上半年同比激增超4倍。

Moonshot AI unveiled Kimi K3, the world’s largest open-source model with 2.8 trillion parameters, dubbed the second “DeepSeek Shock” by Wall Street. European humanoid robotics unicorn SKL Robotics raised $152 million at a $1.35 billion valuation. Meta entered the AI coding market to compete with OpenAI and Anthropic. China’s AI large model market is expected to exceed 70 billion yuan in 2026, while Indian AI startup funding surged over 4x year-over-year in H1.

一、Kimi K3发布：全球最大开源大模型震撼登场

事件概要

7月16日，月之暗面（Moonshot AI）正式发布新一代开源基础模型Kimi K3，采用MoE混合专家架构，成为首个突破2万亿参数级别的开源模型，参数规模达2.8万亿。该模型在多项基准测试中表现优异，华尔街分析师认为这是继2025年初DeepSeek-R1之后，中国AI产业制造的第二次"DeepSeek冲击"。Kimi K3的发布进一步巩固了中国在全球开源AI生态中的领先地位，目前国产AI开源大模型全球累计下载量已突破100亿次。

On July 16, Moonshot AI officially released Kimi K3, a new open-source foundation model using Mixture-of-Experts architecture. With 2.8 trillion parameters, it became the first open-source model to exceed 2 trillion parameters. Wall Street analysts consider this the second “DeepSeek Shock” following the release of DeepSeek-R1 in early 2025. The launch further solidifies China’s leading position in the global open-source AI ecosystem, with Chinese open-source large models accumulating over 10 billion global downloads.

来源 / Sources

• 新浪财经 - 2.8万亿参数刷屏全球：Kimi K3之后，中国大模型真的进入"全面赶超"
• 新浪财经 - 国产AI大模型规模有望超700亿元

二、欧洲首家人形机器人独角兽诞生：SKL Robotics获1.52亿美元融资

事件概要

总部位于伦敦的人工智能与机器人公司SKL Robotics Ltd.（品牌名Humanoid）于7月22日宣布完成1.52亿美元融资，投后估值达13.5亿美元，成为欧洲首家人形机器人独角兽企业。本轮融资由专注前沿科技的风险机构Prime Movers Lab领投，舍弗勒、博世、富邦金控创投及Aglaé Ventures等参与投资。该公司专注于开发通用型人形机器人解决方案，旨在将AI技术与机器人硬件深度融合，服务于工业制造、物流仓储等场景。

London-based SKL Robotics Ltd. (branded as Humanoid) announced on July 22 the completion of a $152 million funding round, achieving a post-money valuation of $1.35 billion and becoming Europe’s first humanoid robotics unicorn. The round was led by Prime Movers Lab, with participation from Schaeffler, Bosch, Fubon Financial Holdings Venture Capital, and Aglaé Ventures. The company focuses on developing general-purpose humanoid robot solutions, aiming to deeply integrate AI technology with robotic hardware for industrial manufacturing and logistics applications.

来源 / Sources

• 腾讯新闻 - 欧洲首家人形机器人独角兽诞生
• Yahoo Finance - Booster Robotics Humanoid Robots Claim Championship at RoboCup 2026

三、Meta进军AI代码赛道：与OpenAI、Anthropic正面竞争

事件概要

Meta于7月9日发布重大更新，宣布进军AI代码大模型市场。公司AI负责人亚历山大·王（Alexandr Wang）牵头推出Muse Spark人工智能模型的重大升级，并公布了具有"激进且具有吸引力"的定价策略。SemiAnalysis最新报告指出，经过一年的激进资本投入与架构重组，Meta旗下的Meta Superintelligence（MSL）有望在未来六个月内超越谷歌，与OpenAI、Anthropic形成三足鼎立之势。Anthropic约75%至85%的收入来自API业务，而OpenAI在2026年第一季度仍有超过65%的收入来自订阅模式。

On July 9, Meta announced its entry into the AI coding market with a major update to its Muse Spark AI model. AI chief Alexandr Wang unveiled aggressive pricing to compete with Anthropic and OpenAI. According to SemiAnalysis, Meta Superintelligence (MSL) is expected to surpass Google within six months, forming a tripartite competition with OpenAI and Anthropic. Anthropic derives approximately 75-85% of revenue from API usage, while OpenAI still relies on subscriptions for over 65% of Q1 2026 revenue.

来源 / Sources

• CNBC - Meta jumps into AI coding market to chase Anthropic and OpenAI
• 新浪财经 - Meta入局AI代码大模型赛道
• MIT Technology Review - AI产业正在围着OpenAI和Anthropic两家公司打转

四、中国AI大模型市场规模有望突破700亿元

事件概要

根据艾媒咨询最新数据，2024年中国AI大模型市场规模达294亿元，机构预测2026年市场规模有望突破700亿元，行业增长动能充足。华西证券分析指出，国内开源大模型行业地位显著提升，已完成由全球边缘参与者向核心竞争力的转变。技术端重磅成果持续落地，国产AI大模型综合实力大幅提升，多款模型已跻身全球第一梯队。随着产业生态不断完善、商业化应用场景持续落地，国内大模型行业维持高速增长态势。

According to iiMedia research, China’s AI large model market reached 29.4 billion yuan in 2024 and is projected to exceed 70 billion yuan by 2026. Huaxi Securities notes that domestic open-source large models have significantly improved their industry position, transforming from marginal participants to core competitive forces. With continuous technological breakthroughs and expanding commercial applications, the domestic large model industry maintains rapid growth momentum.

来源 / Sources

• 新浪财经 - 国产AI大模型规模有望超700亿元
• IT之家 - 行业高速扩容！AI大模型2026年规模或将超700亿

五、印度AI初创融资上半年同比激增超4倍

事件概要

印度AI初创企业在2026年上半年迎来投资热潮。据Inc42数据显示，印度AI初创融资同比增长超过4倍，AI成为该国创业生态系统的核心主题。随着全球AI竞赛持续升温，投资者对印度本土AI企业的兴趣急剧上升。尽管融资规模与美国和中国仍有差距，但印度AI初创企业的快速崛起表明该国正试图在全球AI产业版图中占据一席之地，尤其在AI应用层和企业级解决方案领域展现出潜力。

Indian AI startups saw funding surge over 4x year-over-year in H1 2026, according to Inc42 data. AI has become the defining theme of India’s startup ecosystem, with investor interest picking up sharply as global AI competition intensifies. While still behind the U.S. and China in scale, the rapid rise of Indian AI startups indicates the country’s ambition to carve out a position in the global AI landscape, particularly in application-layer and enterprise solutions.

来源 / Sources

• MSN - Indian AI startup funding soars over 4X YoY in H1 2026
• Crunchbase - Q1 2026 Shatters Venture Funding Records

六、英伟达发布Vera CPU：重新定义数据中心AI算力

事件概要

英伟达近期公布新一代数据中心CPU Vera的详细信息，该处理器专门针对AI工作负载优化，直接向AMD和Intel发起挑战。与此同时，英伟达在GTC 2026大会上宣布其智能体AI（Agentic AI）全栈战略正式落地，从硬件架构到生态布局全面升级。黄仁勋在演讲中指出，两年前播下的生成式AI种子已然结出硕果——"有用的人工智能"已经到来。英伟达还推出了革命性的液冷技术，用闭环液冷系统替代传统冷却塔，有望解决AI数据中心40%的能耗瓶颈问题。

Nvidia recently unveiled details of its next-generation data center CPU, Vera, specifically optimized for AI workloads and challenging AMD and Intel. At GTC 2026, Nvidia announced its Agentic AI full-stack strategy, upgrading from hardware architecture to ecosystem layout. CEO Jensen Huang noted that the generative AI seeds planted two years ago have borne fruit—“useful AI” has arrived. Nvidia also introduced revolutionary liquid cooling technology, replacing traditional cooling towers with closed-loop systems to address 40% of AI data center energy consumption bottlenecks.

来源 / Sources

• MSN - Nvidia details its next-generation Vera CPU for AI
• Forbes - Cadence Teams With Nvidia And Google To Redefine AI System Engineering
• 247 Wall St - Nvidia’s Latest AI Breakthrough May Not Be a Chip

行业数据速览

媒体人解读

开源模型竞赛进入新阶段

Kimi K3的发布标志着全球开源大模型竞赛进入一个新的里程碑阶段。2.8万亿参数不仅是一个数字的突破，更代表了中国AI企业在工程实现、训练效率和推理优化方面的综合实力。从DeepSeek-R1到Kimi K3，中国开源模型正在完成从"追赶者"到"并跑者"甚至部分领域"领跑者"的转变。

对于技术管理者而言，这意味着开源生态的选择将更加丰富。企业级AI应用的门槛正在快速降低——不再需要依赖昂贵的闭源API，而是可以在开源模型基础上进行深度定制。建议技术团队密切关注MoE架构在资源受限场景下的应用潜力，以及开源模型与私有数据的结合方案。

同时，Meta入局AI代码赛道、印度AI融资激增、欧洲机器人独角兽诞生，都表明全球AI产业正从"模型能力竞赛"转向"应用场景落地"和"商业模式验证"的新阶段。对于开发者和创业公司，这意味着更多的细分机会正在涌现。

The Open-Source Model Competition Enters a New Phase

The release of Kimi K3 marks a new milestone in the global open-source large model race. With 2.8 trillion parameters, this represents not just a numerical breakthrough but demonstrates the comprehensive capabilities of Chinese AI companies in engineering, training efficiency, and inference optimization. From DeepSeek-R1 to Kimi K3, Chinese open-source models are transitioning from “followers” to “peers” and even “leaders” in certain areas.

For technology leaders, this means richer options in the open-source ecosystem. The barrier to enterprise AI applications is rapidly lowering—companies no longer need to rely on expensive proprietary APIs but can deeply customize based on open-source models. Technical teams should pay close attention to the potential of MoE architecture in resource-constrained scenarios and integration solutions combining open-source models with proprietary data.

Meanwhile, Meta’s entry into the AI coding market, the surge in Indian AI funding, and the birth of Europe’s robotics unicorn all indicate that the global AI industry is transitioning from “model capability competition” to a new phase of “application scenario deployment” and “business model validation.” For developers and startups, this means more niche opportunities are emerging.

参考来源

英文来源

• CNBC - Meta jumps into AI coding market
• Crunchbase - Q1 2026 Shatters Venture Funding Records
• Yahoo Finance - Booster Robotics at RoboCup 2026
• Forbes - Cadence Teams With Nvidia And Google
• MSN - Indian AI startup funding soars
• 247 Wall St - Nvidia’s AI Breakthrough

中文来源

• 新浪财经 - 2.8万亿参数刷屏全球：Kimi K3发布
• 新浪财经 - 国产AI大模型规模有望超700亿元
• 新浪财经 - Meta入局AI代码大模型赛道
• IT之家 - 2026主流端侧大模型芯片
• MIT Technology Review - AI产业正在围着OpenAI和Anthropic两家公司打转
• 搜狐 - 思灵机器人Agile Robots亮相WAIC 2026

本文内容综合整理自上述公开信息来源，媒体人解读仅代表个人观点，仅供参考。

作者署名：刘一说 🎙️

发布日期：2026年07月24日