文章目录AI科技热点日报 | 2026年06月18日 今日摘要一、Transformer 之父 Noam Shazeer 跳槽 OpenAI,大模型架构战再升级事件概要来源 / Sources二、WAIC 2026 将于 7 月 17-20 日上海启幕,超 300 款 AI 产品全球首发事件概要来源 / Sources三、科创板第五套上市标准扩至 AI 大模型企业,资本市场反应积极事件概要来源 / Sources四、穹彻智能完成数亿元新一轮融资,具身智能赛道持续火热事件概要来源 / Sources五、DeepSeek 首轮融资 510 亿元落地,梁文锋个人出资 200 亿事件概要来源 / Sources六、全球人工智能、安全与伦理大会在日内瓦召开,中美欧治理协同成焦点事件概要来源 / Sources七、苹果 WWDC 特别讲座:4 台 Mac Studio 本地运行 Kimi K2.6 万亿参数模型事件概要来源 / Sources八、美联储新主席沃什首秀维持利率不变,释放明显鹰派信号,AI 算力股承压事件概要来源 / Sources行业数据速览媒体人解读参考来源英文来源中文来源 相关阅读AI科技热点日报 | 2026年06月18日综合整理自 The Verge、TechCrunch、Reuters、MIT Technology Review、新华网科技、36氪、虎嗅网、钛媒体、网易科技、雷锋网、财联社、IT之家、界面新闻 等国内外主流科技媒体当日报道,聚焦人工智能领域核心动态,附资深媒体人综合解读。 今日摘要2026年6月18日,AI 行业上演了一场震动硅谷的人才地震——Transformer 论文作者之一、谷歌 DeepMind 工程副总裁兼 Gemini 联合负责人 Noam Shazeer 正式宣布离开谷歌、加入 OpenAI,将担任架构研究负责人,标志着大模型架构探索战再度升级。同日,国务院新闻办公室披露 7 月 17-20 日 WAIC 2026 将在上海举办,超 300 款 AI 产品全球首发;上交所将科创板第五套上市标准扩至 AI 大模型企业,资本市场反应积极;具身智能企业穹彻智能、DeepSeek 首轮融资等重量级资本动作在 18 日密集落地;美联储新任主席沃什首秀议息,维持利率不变但释放明显鹰派信号,AI 算力股承压。整体来看,人才、资本、政策、算力四重奏正在为下半年 AI 产业定调。June 18, 2026, the AI industry witnessed a “talent earthquake” rocking Silicon Valley: Noam Shazeer, one of the authors of the Transformer paper, Google DeepMind’s VP of Engineering and co-lead of Gemini, officially announced his departure from Google to join OpenAI as Head of Architecture Research, signaling an escalation in the battle for next-generation model architecture. On the same day, China’s State Council Information Office revealed that WAIC 2026 will be held in Shanghai from July 17 to 20, with more than 300 AI products making their global debut; the Shanghai Stock Exchange extended the STAR Market’s fifth set of listing standards to AI large-model companies, drawing a positive capital-market response; major capital moves—including a new funding round for embodied-AI startup Noematrix and DeepSeek’s first round of financing—clustered on the 18th; and the U.S. Federal Reserve, under new Chair Kevin Warsh’s debut meeting, held rates steady but delivered a notably hawkish signal, weighing on AI-computing stocks. Taken together, the “quartet” of talent, capital, policy, and compute is now setting the tone for the AI industry in the second half of the year.一、Transformer 之父 Noam Shazeer 跳槽 OpenAI,大模型架构战再升级事件概要当地时间 6 月 17 日,OpenAI 正式向内部员工宣布,谷歌 DeepMind 工程副总裁、Gemini 模型联合负责人、Transformer 论文共同作者 Noam Shazeer 已加入公司,出任架构研究负责人,负责探索新一代 AI 模型架构并推动 Transformer 架构的进一步演进。Shazeer 本人在 X 平台发文确认,称这是一个艰难的决定,对谷歌团队无比自豪;谷歌回应感谢 Noam 多年的重要贡献。这是 Shazeer 第二次离开谷歌——他 2021 年曾离职创办 Character.AI,2024 年以约 27 亿美元的交易重回 DeepMind 主导 Gemini 预训练,如今再度被 OpenAI 挖角,被业内视为大模型架构探索进入后 Transformer 时代的关键信号。OpenAI officially told its staff on June 17 that Noam Shazeer, Google DeepMind’s VP of Engineering, co-lead of Gemini, and co-author of the Transformer paper, has joined the company as Head of Architecture Research, where he will explore next-generation AI model architectures and drive further evolution of the Transformer architecture. Shazeer confirmed the move on X, calling it a “difficult decision” and saying he was “immensely proud” of the Google team; Google responded that it was “grateful for Noam’s many important contributions.” This is Shazeer’s second departure from Google—he left in 2021 to co-found Character.AI, returned to DeepMind in 2024 via an approximately $2.7 billion deal to lead Gemini pre-training, and has now been poached again by OpenAI, an event the industry reads as a key signal that model-architecture exploration is entering a “post-Transformer era.”来源 / Sources腾讯新闻 - Transformer论文作者之一Noam Shazeer宣布离开Google,加入OpenAI智东西 - 突发!谷歌Gemini联席负责人,被OpenAI挖了财联社 - 谷歌DeepMind研究员诺姆沙泽尔将加入OpenAI二、WAIC 2026 将于 7 月 17-20 日上海启幕,超 300 款 AI 产品全球首发事件概要国务院新闻办公室 6 月 17 日发布会披露,2026 世界人工智能大会暨人工智能全球治理高级别会议(WAIC 2026)将于 7 月 17 日至 20 日在上海举办,主题智能伙伴,共创未来,由外交部、科技部、工信部等 10 个部门与上海市政府共同主办,图灵奖得主姚期智担任大会主席,强化学习之父理查德·萨顿任国际联席主席。大会设置世博、张江、西岸三大展区,展览总面积超 10 万平方米,1100 余家企业带来 3000 余项产品,超 300 款 AI 产品全球首发,并首设 OPC 专属展示区,180 家企业携成果入驻。本届共收到 230 个有效申报项目,海外项目占 14.3%,覆盖智能体、算力芯片、具身智能等领域。China’s State Council Information Office announced on June 17 that the 2026 World AI Conference and High-Level Meeting on Global AI Governance (WAIC 2026) will be held in Shanghai from July 17 to 20, with the theme “Intelligent Partners, Co-creating the Future.” It is co-hosted by 10 ministries including the Ministry of Foreign Affairs, the Ministry of Science and Technology, and the Ministry of Industry and Information Technology together with the Shanghai Municipal Government. Turing Award laureate Yao Qizhi serves as conference chair, and “father of reinforcement learning” Richard Sutton as international co-chair. The exhibition spans three zones—Expo, Zhangjiang, and West Bund—covering more than 100,000 square meters, with 1,100 exhibitors, 3,000 products, more than 300 AI products making their global debut, and a first dedicated OPC showcase area hosting 180 companies. The event received 230 valid project applications, with overseas projects accounting for 14.3%, spanning intelligent agents, computing chips, and embodied intelligence.来源 / SourcesIT之家/央视 - 2026 世界人工智能大会将在上海举办,超 300 款 AI 产品全球首发DoNews - 2026 世界人工智能大会今年 7 月在上海举办新华网 - WAIC 2026 倒计时30天发布会在沪举办三、科创板第五套上市标准扩至 AI 大模型企业,资本市场反应积极事件概要上海证券交易所 6 月 17 日发布《上海证券交易所发行上市审核规则适用指引第 10 号——人工智能大模型企业适用科创板第五套上市标准》,进一步规范科技型企业适用第五套上市标准,支持尚未形成一定收入规模的优质人工智能大模型企业适用第五套上市标准,加快推进人工智能创新发展。6 月 18 日,A 股人工智能板块应声上涨,中证人工智能产业指数(931071)早盘强势上涨 1.46%,中际旭创涨 4.29%、光迅科技涨 4.13%、千方科技涨 3.28%;中证人工智能主题指数(930713)上涨 2.11%,AI 人工智能 ETF 平安(512930)上涨 2.24%,冲击 5 连涨;人工智能 ETF 华富(515980)上涨 1.55%,盘中最高涨超 2%。市场普遍认为这是国内 AI 大模型企业上市路径的关键松绑。The Shanghai Stock Exchange (SSE) issued on June 17 the “No. 10 Guideline on the Application of the SSE’s Listing Review Rules—Application of the Fifth Set of Listing Standards of the STAR Market by AI Large-Model Companies,” further regulating the application of the fifth set of listing standards for tech companies and supporting high-quality AI large-model companies that have not yet generated meaningful revenue to apply the fifth set, accelerating AI innovation and development. On June 18, A-share AI-related stocks rallied in response: the CSI AI Industry Index (931071) jumped 1.46% in early trading, with Zhongji Innolight up 4.29%, Accelink up 4.13%, and China Transinfo up 3.28%; the CSI AI Theme Index (930713) rose 2.11%, the Ping An AI ETF (512930) gained 2.24% and is now poised for a fifth straight up day, while the Huafu AI ETF (515980) added 1.55%, touching an intraday gain of more than 2%. The market widely read this as a key easing of the listing path for domestic AI large-model companies.来源 / Sources财联社 - 科创板第五套标准扩至人工智能大模型,AI人工智能ETF平安开盘涨超1.1%新浪科技 - AI人工智能ETF平安上涨2.24%,上交所发布人工智能大模型企业适用科创板第五套上市标准上海证券交易所 - 第十号指引原文四、穹彻智能完成数亿元新一轮融资,具身智能赛道持续火热事件概要6 月 18 日,具身智能企业穹彻智能(Noematrix)宣布完成新一轮数亿元融资,本轮融资由无锡数据集团领投,投资方包括上海交通大学 AI 未来基金(创业基金)、上海创之智科技有限公司(上海创智学院全资子公司)、一村资本等。资金将持续推进具备强泛化性和高自主决策能力的具身智能大模型研发与迭代,加速在酒店洗衣、家庭整理等真实场景的落地。这是穹彻智能近半年内完成的又一轮融资,此前公司已获得 Prosperity7 Ventures、红杉中国、C Capital、阿里巴巴和 Sea 等机构加持。该公司由深耕通用智能机器人领域近十年的非夕科技集团战略孵化,2023 年 11 月成立。On June 18, embodied-AI company Noematrix announced the completion of a new funding round worth several hundred million yuan. The round was led by Wuxi Data Group, with participation from the Shanghai Jiao Tong University AI Future Fund (Entrepreneurship Fund), Shanghai Chuangzhi Technology Co., Ltd. (a wholly owned subsidiary of Shanghai Chuangzhi Institute), and Yicun Capital. The capital will continue to advance RD and iteration of embodied-AI foundation models with strong generalization and high autonomous decision-making, accelerating deployment in real-world scenarios such as hotel laundry and household tidying. This is another funding round Noematrix has closed within the past half year; previously it received backing from Prosperity7 Ventures, Sequoia China, C Capital, Alibaba, and Sea. The company was incubated in November 2023 by Flexiv, a group with nearly a decade of experience in general intelligent robotics.来源 / Sources新浪科技 - 穹彻智能获数亿元战略投资:上海交大、创智学院、无锡数据集团等联合注资投资界 - 穹彻智能完成新一轮数亿元融资界面新闻 - 穹彻智能完成新一轮数亿元融资五、DeepSeek 首轮融资 510 亿元落地,梁文锋个人出资 200 亿事件概要企查查 APP 信息显示,杭州深度求索人工智能基础技术研究有限公司(DeepSeek)完成 A 轮融资,融资金额约 510 亿元人民币(约 74 亿美元),企业估值约 4000 亿元人民币。本轮融资中,创始人梁文锋个人出资约 200 亿元,为最大单一出资方,占比近四成;腾讯出资约 100 亿元,宁德时代体系(含宁德时代及溥泉资本)出资约 50 亿元,网易、京东、Monolith 砺思资本、IDG 资本各出资约 30 亿元,正心谷投资、拾象科技各出资约 15 亿元。投资者锁定 5 年。本轮融资距 DeepSeek 今年 4 月启动融资仅两个多月,刷新国产大模型单轮融资最高纪录。Information from Qichacha shows that Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. (DeepSeek) has closed its Series A funding, raising approximately RMB 51 billion (about USD 7.4 billion) at an enterprise valuation of roughly RMB 400 billion. In this round, founder Liang Wenfeng personally invested about RMB 20 billion, the largest single contribution, accounting for nearly 40% of the round; Tencent invested about RMB 10 billion, the CATL system (CATL and Puquan Capital) about RMB 5 billion; NetEase, JD.com, Monolith Lisi Capital, and IDG Capital each contributed about RMB 3 billion; and Zhengxin Valley Investment and Shixiang Technology each invested about RMB 1.5 billion. Investors face a 5-year lock-up. The round, completed just over two months after DeepSeek kicked off the financing in April, set a new record for the largest single financing round by a domestic large-model company.来源 / Sources腾讯新闻 - DeepSeek获510亿融资,投资方公布:腾讯、宁德时代、京东、网易等速途网 - DeepSeek首轮融资完成!梁文锋出资占四成The Information - DeepSeek Series A 报道六、全球人工智能、安全与伦理大会在日内瓦召开,中美欧治理协同成焦点事件概要2026 年全球人工智能、安全与伦理大会于 6 月 18 日在瑞士日内瓦开幕,会议持续 2 天,由各国外交官、政策制定者、学术界和产业界人士共同参与,围绕人工智能对国际和平与安全的影响展开讨论。议程覆盖 AI 军事化风险、模型开源与闭源治理、AI 伦理国际标准制定、大模型对齐与失控防范等核心议题。大会正值 Anthropic、OpenAI、xAI 等头部企业相继呼吁全球放缓 AI 研发以及欧盟 AI 法案全面生效倒计时的当口,被视为下半年全球 AI 治理议程的关键节点。The 2026 Global Conference on AI, Safety and Ethics opened on June 18 in Geneva, Switzerland, running for two days, with diplomats, policymakers, academics, and industry participants jointly exploring the impact of AI on international peace and security. The agenda covers core topics including the militarization of AI, governance of open- vs. closed-source models, the establishment of international standards for AI ethics, and the prevention of large-model misalignment and loss of control. The conference comes at a sensitive moment—headline companies such as Anthropic, OpenAI, and xAI have successively called for a “global slowdown in AI RD,” and the EU AI Act is in the final countdown to full effect—making it a key milestone on the global AI governance agenda for the second half of the year.来源 / Sources腾讯新闻 - 全球人工智能、安全与伦理大会18日在日内瓦举行央视新闻(CCTV) - 大会专题报道七、苹果 WWDC 特别讲座:4 台 Mac Studio 本地运行 Kimi K2.6 万亿参数模型事件概要苹果官方账号 6 月 18 日在 B 站发布 98 分钟视频,回顾 2026 年全球开发者大会(WWDC)的特别演讲内容,重点演示了用 4 台 Mac Studio 本地运行 1 万亿参数的月之暗面 Kimi-K2.6 模型。该视频名为《Apple 智能与 Xcode 揭秘:特别讲座》,于 2026 年 WWDC 期间在乔布斯剧院录制。苹果软件工程高级副总裁 Craig Federighi 现场展示了 Apple Silicon 统一内存架构对超大参数模型本地推理的支撑能力,以及 Xcode 26 新增的 AI 调试工具链。该演示被视为 Apple Intelligence 国行版落地前夜的关键预热。Apple’s official account released a 98-minute video on Bilibili on June 18, recapping the special-talk content from WWDC 2026, with a focus on a local demo of Moonshot AI’s 1-trillion-parameter Kimi-K2.6 model running on four Mac Studio machines. The video, titled “Apple Intelligence and Xcode Demystified: Special Lecture,” was recorded at the Steve Jobs Theater during WWDC 2026. Apple’s SVP of Software Engineering, Craig Federighi, demonstrated on site how Apple Silicon’s unified-memory architecture supports local inference for very large parameter models, along with the new AI debugging toolchain in Xcode 26. The demo is seen as key pre-launch warm-up for the imminent China launch of Apple Intelligence.来源 / Sources腾讯新闻 - 苹果 AI 讲座重头戏:4 台 Mac Studio 本地运行 Kimi K2.6 模型Apple Developer - WWDC 2026 Special Lecture月之暗面(Moonshot AI) 官方公告八、美联储新主席沃什首秀维持利率不变,释放明显鹰派信号,AI 算力股承压事件概要美东时间 6 月 17 日,美联储 2026 年 6 月议息会议宣布维持联邦基金利率目标区间在 3.5%-3.75% 不变,这是新任主席凯文·沃什就任以来的首次议息会议,也是美联储今年连续第四次维持利率不变,符合市场普遍预期。但会议释放明显鹰派信号:点阵图显示 18 位官员中有 9 位倾向于年内加息,其中 5 位支持加息两次、1 位支持加息三次,利率预测中值从 3 月的 3.4% 升至 3.75%;同时上调 2026 年底核心通胀率预测至 3.3%(此前 2.7%),下调 GDP 增速预测至 2.2%(此前 2.4%)。声明删除有关可能进一步调整利率的措辞。受此影响,美股、黄金等风险资产下跌,英伟达等 AI 算力股承压。The U.S. Federal Reserve’s June 2026 FOMC meeting on June 17 (U.S. Eastern Time) voted to keep the target range for the federal funds rate at 3.50%-3.75% unchanged. It was the first meeting under new Chair Kevin Warsh and the Fed’s fourth consecutive rate hold this year, in line with broad market expectations. However, the meeting delivered a clearly hawkish signal: the dot plot shows 9 of 18 officials leaning toward a rate hike this year, with 5 backing two hikes and 1 backing three; the median rate projection rose from 3.4% in March to 3.75%. The Fed also raised its year-end 2026 core PCE inflation forecast to 3.3% (from 2.7%) and cut its GDP growth forecast to 2.2% (from 2.4%). The statement dropped language about possible further rate adjustments. Risk assets such as U.S. equities and gold declined in response, with AI-computing stocks including NVIDIA coming under pressure.来源 / Sources财联社 - 美联储年内第四次维持利率不变新浪财经 - 沃什首秀!美联储维持利率不变Reuters - Fed holds rates steady at June meeting行业数据速览指标数据备注DeepSeek A 轮融资金额约 510 亿元人民币(74 亿美元)国产大模型单轮融资最高纪录,梁文锋个人出资 200 亿DeepSeek 投后估值约 4000 亿元人民币2026 年 4 月才启动融资,仅用 2 个多月WAIC 2026 参展企业1100 家展览总面积超 10 万平方米,横跨世博、张江、西岸三大片区WAIC 2026 全球首发 AI 产品300 款来自 230 个有效申报项目,海外项目占 14.3%美联储 6 月联邦基金利率3.5%-3.75%沃什首秀,今年连续第四次维持不变,立场偏鹰2026 年底核心 PCE 通胀预测中值3.3%高于 3 月预测的 2.7%18 位 FOMC 官员倾向年内加息人数9 位其中 5 位支持加息 2 次、1 位支持 3 次穹彻智能本轮融资金额数亿元人民币具身智能赛道半年内再次融资苹果 Mac Studio 本地运行 Kimi K2.61 万亿参数4 台 Mac Studio Apple Silicon 统一内存架构中证人工智能产业指数(931071)涨幅1.46%上交所第五套标准扩至 AI 大模型企业刺激上涨AI 人工智能 ETF 平安(512930)涨幅2.24%冲击 5 连涨,近 1 周累计 5.31%谷歌 DeepMind→OpenAI 跳槽人物Noam ShazeerTransformer 论文共同作者、Gemini 联合负责人媒体人解读6 月 18 日这天的 AI 新闻,表面是分散的人、钱、规、算四条线,本质是同一个故事——大模型竞争从参数与算力向架构与人才切换。首先看人才端:Shazeer 离开谷歌加入 OpenAI 不是孤立事件,这是 2024 年他以 27 亿美元从 Character.AI 回到 DeepMind 后第二次出走。考虑到他在 Transformer 架构上的祖师爷地位,这次转会几乎等同于OpenAI 把大模型架构的下一棒握在了手里——结合近期 OpenAI 内部对非 Transformer 架构(如 State-Space、Diffusion-MoE 混合架构)的探索,Shazeer 的加盟,可能预示着 GPT-6 以后的产品形态将出现质变。再看资本端:DeepSeek 首轮 510 亿、梁文锋个人出资 200 亿、锁定 5 年——这种创始人拿大头、产业资本国资跟投的格局,本质是把 DeepSeek 锁定在长期主义赛道,避免被短期估值绑架。穹彻智能这种具身智能企业在不到半年内连融两轮,反映的则是机器人产业从实验室 Demo切换到真场景部署过程中的卡位战。政策端:科创板第五套上市标准向 AI 大模型企业开放,这是国内 AI 行业最重要的资本通道松绑。第五套标准的特点是不强调收入规模,允许尚未盈利的硬科技公司上市,过去主要用于生物医药和半导体,现在 AI 大模型企业获准适用,意味着像 DeepSeek、Kimi、智谱这类重投入、长回收期的公司,A 股 IPO 路径正式打开,直接对标英伟达、Anthropic 走出美股牛市。国际治理端:日内瓦大会与美联储鹰派决议看似无关,实则共同指向一个事实——全球 AI 治理已经从原则宣言进入硬约束阶段。沃什首秀直接抬高年内加息预期,加上特朗普 6 月 8 日签署的AI 行政令要求前沿模型发布前 30 天接受审查,意味着美国 AI 公司在算力贵、监管严、上市路径不确定三重压力下,接下来的半年会非常卷。综合来看,2026 年下半年 AI 行业的三个主旋律:第一,架构创新提速,Transformer 不再是唯一答案,MoE、SSM、Diffusion 混合架构可能并行胜出;第二,中国大模型企业进入融资-上市双通道,资本市场将出现 AI 板块独立行情;第三,具身智能与多模态 Agent 走出 PPT,开始大规模在工业、家庭、物流真实场景中跑数据。A final takeaway: 6/18 is the day the AI industry shifts from “compute and parameters” to “architecture and people.” The Shazeer move, DeepSeek’s RMB 51B round, the SSE’s STAR Market easing, and Warsh’s hawkish debut at the Fed all point in the same direction. Whoever controls the next-generation architecture and the talent to invent it will control the next decade of AI.参考来源英文来源The Information - DeepSeek Series A CoverageReuters Technology - Fed Holds Rates SteadyThe Verge - AI Industry CoverageTechCrunch - OpenAI Talent MovesMIT Technology ReviewVentureBeat - AI Funding NewsIEEE Spectrum中文来源腾讯新闻 - Noam Shazeer 离开 Google 加入 OpenAI智东西 - 突发!谷歌 Gemini 联席负责人,被 OpenAI 挖了财联社 - 美联储年内第四次维持利率不变IT之家 - 2026 世界人工智能大会将在上海举办新浪科技 - 穹彻智能获数亿元战略投资腾讯新闻 - DeepSeek 获 510 亿融资财联社 - 科创板第五套标准扩至人工智能大模型腾讯新闻 - 全球人工智能、安全与伦理大会 18 日在日内瓦举行腾讯新闻 - 苹果 AI 讲座重头戏:4 台 Mac Studio 本地运行 Kimi K2.6钛媒体 - WAIC 2026 倒计时 30 天36氪 - 融资快报虎嗅网网易科技雷锋网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。发布日期:2026年06月18日 相关阅读D60: 给读者的信你的下一步是什么开发中常用素材格式完整清单D47 《数据治理从混乱到可控》防护体系的「治理」篇一人公司OPC典型案例与商业模式研究报告主流 Claw 类产品技术与商业差异全景报告常见误区管理者最容易踩的 5 个坑团队 AI 成熟度自评模型
