文章目录AI科技热点日报 | 2026年6月30日 今日摘要一、国务院加力推进AI创新突破A股AI产业链放量大涨事件概要来源 / Sources二、DeepSeek V4正式版官宣7月中旬上线API首推峰谷定价事件概要来源 / Sources三、韩国抛出4755万亿韩元三大超级项目国家下场打造AI算力底座事件概要来源 / Sources四、OpenAI GPT-5.6三档模型亮相360/Sakana推Mythos平替抢窗口事件概要来源 / Sources五、智元第15000台具身机器人量产下线距万节点仅用3个月再刷新行业纪录事件概要来源 / Sources六、谷歌云×SandboxAQ开放科学AI模型 限制Meta调用Gemini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6月30日综合整理自The Verge、TechCrunch、VentureBeat、Reuters Technology、CNN Business、钛媒体、36氪、虎嗅、新华网科技、IT之家、网易科技、同花顺财经等国内外主流科技媒体6月29日-30日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30日中国AI产业链进入政策催化模型迭代终端爆发三浪叠加阶段。国务院常务会议6月29日研究人工智能发展、强调加力推进AI创新突破、超大规模智算集群建设次日A股CS人工智指数(930713)放量大涨4.13%紫光股份涨停、景嘉微涨14.25%DeepSeek团队6月30日官宣V4正式版将于7月中旬上线并首次引入峰谷定价机制韩国政府6月29日联手三星、SK海力士抛出合计4755万亿韩元的三大超级项目打造国家AI算力底座OpenAI在6月27日率先放出GPT-5.6三档模型后全球大模型竞速进入分场景定价新阶段具身智能侧智元机器人6月28日第15000台精灵G2量产下线距万台节点仅3个月再刷新全球行业纪录。On June 30, 2026, China’s AI industrial chain entered a three-wave convergence of “policy catalysis plus model iteration plus terminal explosion.” A State Council executive meeting on June 29 reviewed AI development and urged “stepping up AI innovation breakthroughs and the construction of ultra-large intelligent computing clusters.” The very next day, the CSI AI Index (930713) surged 4.13 percent, with Unisplendour hitting the daily limit up and Jingjia Micro climbing 14.25 percent. The DeepSeek team officially announced on June 30 that the V4 formal release would arrive in mid-July, debuting an unprecedented peak-valley pricing mechanism. On June 29, the South Korean government, partnering with Samsung and SK Hynix, unveiled “three mega projects” totaling KRW 4,755 trillion to build the nation’s AI computing backbone. After OpenAI released the three-tier GPT-5.6 model on June 27, the global large-model race entered a new phase of “scenario-based pricing.” On the embodied intelligence side, Zhiyuan Robotics rolled the 15,000th Lingxi G2 robot off the production line on June 28, setting a fresh global industry record just three months after crossing the 10,000-unit mark.一、国务院加力推进AI创新突破A股AI产业链放量大涨事件概要国务院总理李强6月29日主持召开国务院常务会议听取人工智能发展情况汇报。会议指出要深刻把握人工智能演进趋势完善支持政策和治理体系牢牢掌握发展主动权。要加力推进人工智能创新突破加快关键技术攻关和超大规模智算集群建设强化高质量数据供给加强人才、资金等要素保障支持企业开展基础研究和前沿探索。要深入实施人工智能行动发挥我国产业体系完备、应用场景丰富等优势促进智能产品和服务加快规模化商业应用。要守牢人工智能安全底线完善科技伦理、测试认证等制度规则构建动态适应、分级分类的安全监管体系加强国际人工智能治理合作。会议精神次日迅速传导至资本市场6月30日早盘截至10点41分CS人工智指数(930713)强势上涨4.13%紫光股份涨停景嘉微上涨14.25%奥比中光上涨8.50%中科曙光、和而泰等个股跟涨科创人工智能ETF华宝(589520)场内涨幅盘中上探2.25%、11天8阳。资金面上AI产业链景气持续上行全球近20家模拟及功率半导体企业将于7月1日启动新一轮涨价三、SK海力士亦同步发布大规模投资计划AI上游景气度继续向上游卡点环节传导。China’s State Council Premier Li Qiang chaired an executive meeting on June 29 to review AI development progress. The meeting called for deeply grasping AI’s evolution trend, improving supporting policies and governance systems, and seizing the initiative in development. It demanded intensified efforts to advance AI innovation breakthroughs, accelerate key-technology research and the construction of ultra-large intelligent computing clusters, strengthen high-quality data supply, and bolster talent and funding guarantees. The meeting also called for deeply implementing the “AI Plus” initiative, leveraging China’s complete industrial system and rich application scenarios to accelerate large-scale commercial use of intelligent products and services, while guarding the safety bottom line by improving ethics, testing and certification regimes and strengthening international AI governance cooperation. Market reaction was immediate: as of 10:41 a.m. on June 30, the CSI AI Index (930713) jumped 4.13 percent, with Unisplendour hitting the daily limit and Jingjia Micro up 14.25 percent, Orbbec up 8.50 percent, and Sugon and Hetai following suit; the Huabao Sci-Tech Innovation AI ETF (589520) advanced as much as 2.25 percent, with eight up days in the last eleven. Some 20 global analog and power-semiconductor vendors are set to launch another round of price hikes on July 1, while Samsung and SK Hynix announced major investment plans, indicating that AI’s upstream prosperity is continuing to transmit toward choke-point segments.来源 / Sources国务院李强主持召开国务院常务会议听取人工智能发展情况汇报 - 同花顺财经央视财经政策加码AI创新突破6月30日澎湃新闻AI产业链景气持续上行6月30日证券时报6月30日操盘必读影响股市利好或利空消息二、DeepSeek V4正式版官宣7月中旬上线API首推峰谷定价事件概要DeepSeek团队6月30日正式宣布DeepSeek V4正式版计划于2026年7月中旬上线并首次同步引入峰谷定价机制。根据官方公布的价格表高峰时段API价格为平时价格的两倍平时价格与现行V4 API定价持平。这意味着DeepSeek V4在保持平价门槛的同时将通过价格杠杆缓解推理算力峰谷差、引导企业级用户错峰使用。V4预览版自2026年4月24日发布以来已在长上下文百万字、Agent能力与世界知识推理性能上实现国内与开源领域的领先。本次正式版在保留开源、普惠定位的基础上加入分时定价被业内视为国产大模型从卷性能转向卷商业化运营的标志性事件。截至6月30日11点13分AI产业链整体走强、科创人工智能ETF汇添富(589560)涨超2.5%市场对国产模型迭代加速的预期正在快速反映。The DeepSeek team officially announced on June 30 that the DeepSeek V4 formal release will launch in mid-July 2026, simultaneously debuting an unprecedented peak-valley pricing mechanism. According to the published price schedule, peak-hour API pricing will be twice the off-peak price, while off-peak pricing will remain flat relative to current V4 API pricing. The structure preserves an accessible entry threshold while using price levers to smooth inference load across the day and steer enterprise users toward off-peak usage. Since the V4 preview was released on April 24, 2026, DeepSeek has led the domestic and open-source fields in long-context (one-million-character) handling, agent capabilities and world-knowledge reasoning. The launch of a formal version keeps that open, inclusive positioning while adding time-based pricing, an event industry watchers view as a marker of domestic large models shifting from pure performance competition to commercialization operations. By 11:13 a.m. on June 30, the AI industrial chain broadly strengthened, with the Huatai-PB Sci-Tech Innovation AI ETF (589560) up over 2.5 percent as market expectations of accelerating domestic-model iteration were quickly priced in.来源 / Sources腾讯财经DeepSeek-V4正式版官宣7月上线API引入峰谷定价6月30日同花顺财经DeepSeek V4正式版官宣7月上线6月29日-30日财联社AI大模型迭代加速DeepSeek V4正式版计划7月中旬上线6月30日量子位/CSDNDeepSeek-V4 详细解读三、韩国抛出4755万亿韩元三大超级项目国家下场打造AI算力底座事件概要6月29日韩国政府联手三星电子、SK海力士两大半导体巨头公布以半导体等为核心的三大超级项目两家巨头未来将在AI相关领域总计投入4755万亿韩元约合20.92万亿元人民币规模达韩国史上最大量级。同日三星电子单独宣布规模达2655万亿韩元的投资计划其中2030万亿韩元投向包括平泽、龙仁在内的首尔都市圈半导体产业集群625万亿韩元投向湖南、忠清、岭南等地区旨在将韩国打造成为全球高科技产业集群。三星电子会长李在镕表示当前产能已不足以满足市场需求公司正考虑在韩国光州建设新的投资基地SK海力士方面亦同步加速HBM4E 12层下一代产品的样品出货并已与NVIDIA达成多年期技术合作伙伴关系。资本市场层面6月30日亚洲早盘受消息提振韩国救市预期叠加利好带动整个AI科技板块反弹。On June 29, the South Korean government, together with Samsung Electronics and SK Hynix, announced “three mega projects” centered on semiconductors, with the two giants pledging a combined KRW 4,755 trillion (about RMB 20.92 trillion) in AI-related investment, the largest such plan in Korea’s history. Samsung separately disclosed a KRW 2,655 trillion investment plan the same day, with KRW 2,030 trillion going to the Seoul metropolitan semiconductor cluster including Pyeongtaek and Yongin, and KRW 625 trillion going to the Honam, Chungcheong and Youngnam regions, aiming to position Korea as a global high-tech industrial cluster. Samsung chairman Lee Jae-yong said current capacity is insufficient to meet demand and that the company is considering a new investment base in Gwangju. SK Hynix is accelerating sample shipments of its next-generation 12-layer HBM4E, and has signed a multi-year technology partnership with NVIDIA. In capital markets, the early Asian session on June 30 was buoyed by these announcements, with Korean “rescue” expectations combined with positive news lifting the broader AI tech complex.来源 / Sources腾讯财经AI重磅政策利好深入实施人工智能行动6月30日腾讯财经AI早报三星计划在韩国投资2655万亿韩元6月30日SK hynix NewsroomSK hynix Ships Samples of 12-Layer Next-Gen ‘HBM4E’21世纪经济报道Asia Market Daily Samsung and SK Hynix to Unveil 650B Chip Investment in South Korea6月28日-29日四、OpenAI GPT-5.6三档模型亮相360/Sakana推Mythos平替抢窗口事件概要6月27日OpenAI正式推出GPT-5.6有限预览版一口气发了三个型号Sol太阳主打复杂推理与代码、Terra地球面向日常工作、Luna月亮走轻量高速路线。据TechCrunch报道旗舰Sol在Terminal-Bench 2.1编程测试上比Claude Mythos有显著优势。中信证券测算2026年5月Anthropic ARR年度经常性收入达470亿美元在2B市场实现对OpenAI的反超OpenAI此次通过三档模型场景化定价反击意在C端保住ChatGPT的同时在B端补齐Coding及Agent的能力缺口。与此同时6月29日TechCrunch披露Anthropic Mythos 5和Fable 5因出口禁令被全球下架360推出Tulongfeng屠龙风直接对标Mythos东京初创Sakana AI同期发布Fugu河豚前沿模型亚洲其他AI初创也纷纷推出Mythos-like模型抢占市场空白。OpenAI与Anthropic的IPO竞速也在加速多方消息显示OpenAI正为2026年第四季度公开上市做准备。On June 27, OpenAI officially launched the limited preview of GPT-5.6 in three tiers: Sol (the flagship targeting complex reasoning and coding), Terra (for everyday work) and Luna (a lightweight, high-speed model). According to TechCrunch, the flagship Sol delivered a meaningful edge over Claude Mythos on the Terminal-Bench 2.1 coding benchmark. CITIC Securities estimates put Anthropic’s May 2026 ARR at USD 47 billion, surpassing OpenAI in the enterprise market; OpenAI’s “three-tier model plus scenario-based pricing” counterpunch is designed to defend ChatGPT on the consumer side while closing coding-and-agent capability gaps on the enterprise side. Meanwhile, on June 29 TechCrunch reported that Anthropic’s Mythos 5 and Fable 5 models had been pulled from global markets under export-control restrictions, with 360 launching “Tulongfeng” directly to compete with Mythos and Tokyo startup Sakana AI releasing its Fugu frontier model on the same day. Other Asian AI startups also rolled out “Mythos-like” models to fill the void, as OpenAI accelerates toward a Q4 2026 IPO in parallel with Anthropic.来源 / Sources腾讯财经AI行业晨报 2026-06-29TechCrunchAsian AI startups launch Mythos-like models as Anthropic’s export ban drags on搜狐Anthropic与OpenAI竞速GPT-5.6升级对决Claude Mythos6月7日中华网/CSDNAnthropic与OpenAI 2026营收对决6月23日五、智元第15000台具身机器人量产下线距万节点仅用3个月再刷新行业纪录事件概要6月28日智元机器人宣布第15000台通用具身机器人精灵G2正式量产下线距离公司2026年3月突破万台量产节点仅用时不足3个月再次刷新全球具身智能行业量产速度纪录。本次下线的第15000台产品即日交付给龙旗科技工厂开展智能制造作业。智元合伙人、高级副总裁、具身业务总裁姚卯青在仪式现场指出量产是具身智能走出实验室、落地产业的核心根基规模化生产持续摊薄硬件成本打开了商用普及空间。从2025年年底5000台到2026年3月10000台再到6月15000台智元量产曲线呈指数式爬坡亦使具身智能正式进入工厂能用、上岗能干的工程化落地阶段。机构预计以智元为代表的中国具身机器人军团将在2026下半年集中进入汽车零部件、3C组装、新能源工厂等真实场景具身智能赛道的投资逻辑正从概念验证过渡到回本周期测算。On June 28, Zhiyuan Robotics announced the 15,000th Lingxi G2 general embodied robot formally rolled off the production line, just under three months after the company crossed the 10,000-unit mass-production mark in March 2026, setting a fresh global industry mass-production speed record. The 15,000th Lingxi G2 was delivered the same day to Longcheer Technology’s factory for intelligent-manufacturing operations. Yao Maoqing, Zhiyuan Partner, Senior Vice President and President of the Embodied Business, said on site that mass production is the core foundation for embodied intelligence to leave the lab and land in industry, with scale production continuously amortizing hardware costs and opening up commercial adoption space. From 5,000 units by end-2025 to 10,000 in March 2026 and 15,000 in June, Zhiyuan’s mass-production curve is climbing exponentially, taking embodied intelligence into a true engineering deployment phase. Institutional forecasts expect the Chinese embodied-robot camp, led by Zhiyuan, to concentrate on entering real automotive-parts, 3C assembly and new-energy factory scenarios in H2 2026, as the embodied-AI investment thesis shifts from concept validation to payback-period modeling.来源 / Sources钛媒体智元第15000台通用具身机器人正式量产下线6月28日澎湃新闻智元第15000台具身智能机器人下线6月29日新华网/红星新闻智元第15000台具身机器人下线6月28-29日智元机器人官网六、谷歌云×SandboxAQ开放科学AI模型 限制Meta调用Gemini事件概要6月29日Alphabet旗下谷歌与软件公司SandboxAQ达成合作将通过其云服务平台向企业及研究机构开放SandboxAQ专精于加速药物研发、材料科学、半导体制造的大型量化模型Large Quantitative Models。研究人员可以在Google Cloud Marketplace无缝采购和部署SandboxAQ方案并将谷歌的Gemini语言模型与基于科学方程和实验室数据训练的AI系统结合使用。SandboxAQ近期还发布了依托NVIDIA BioNeMo提供算力支撑的G蛋白偶联受体GPCR虚拟筛选解决方案帮助研发人员低成本锁定潜在候选药物。另一边6月29日AI早报披露谷歌对Meta调用Gemini大模型作出限制主因Gemini需求过大、算力供给跟不上要求Meta控制调用规模与节奏——这从一个侧面印证算力即定价权正成为新一轮大厂博弈焦点。On June 29, Alphabet’s Google reached a cooperation with software firm SandboxAQ to expose SandboxAQ’s “Large Quantitative Models” (LQMs) through its cloud platform, enabling enterprises and research institutions to use AI specialized in accelerating drug discovery, materials science and semiconductor manufacturing. Researchers can seamlessly procure and deploy SandboxAQ solutions on Google Cloud Marketplace and combine Gemini language models with AI systems trained on scientific equations and laboratory data. SandboxAQ had recently released a G-protein-coupled receptor (GPCR) virtual-screening solution powered by NVIDIA BioNeMo, helping researchers cost-effectively shortlist candidate molecules. On a separate front, the AI morning brief on June 29 disclosed that Google is restricting Meta’s use of the Gemini large model because demand has outstripped supply, requiring Meta to control call volume and pacing, a signal that “compute is pricing power” is becoming the new focal point of big-tech tussles.来源 / Sources腾讯财经谷歌云将提供用于科学研究的专业AI模型6月29日腾讯财经谷歌云牵手软件公司SandboxAQ6月29日腾讯财经AI早报算力告急谷歌限制Meta对Gemini AI大模型的使用6月30日SandboxAQ × Google Cloud合作背景行业数据速览指标数据备注CS人工智指数9307136月30日涨幅4.13%国务院常务会议利好兑现紫光股份 6月30日表现涨停智算AI芯片主线共振景嘉微 6月30日涨幅14.25%GPU国产替代主线奥比中光 6月30日涨幅8.50%机器人视觉3D感知韩国三大超级项目投资总额4,755万亿韩元≈20.92万亿人民币三星SK海力士联动三星电子单独投资计划2,655万亿韩元韩国史上最大单笔投资DeepSeek V4定价策略高峰时段平价×2业内首创峰谷定价DeepSeek V4正式版上线时间2026年7月中旬同步引入时分定价智元机器人量产规模15,000台6月28日距万节点仅3个月智元精灵G2批量交付客户龙旗科技ODM产线量产即落地Anthropic 2026年5月ARR470亿美元2B市场反超OpenAIOpenAI GPT-5.6子型号Sol/Terra/Luna三档6月27日预览版WPS OpenAI GPT-5.6定价分档高峰为平价2倍与DeepSeek同步方向联想万全异构智算平台V5.06月24日发布MWC26上海展出联想AI可信一体机MWC上海首发“硬件级身份证”MWC26上海参会者37,300人/143国国际参会33%ChinaJoy 2026主题“与AI同游”7月31日-8月3日全球科技巨头2026资本开支指引6,500亿-7,250亿美元同比60%~77%媒体人解读6月30日这一天的AI产业图谱有三条主线值得关注。第一条是政策与资本共振中国国务院将AI创新突破超大规模智算集群列入6月29日常务会议重点次日A股AI板块就把预期兑现成了涨停板和ETF的11天8阳资本市场用真金白银给出了态度——这不是主题炒作而是政策与企业现金流、行业景气在时间窗口上的强共振。第二条是模型层的分场景定价成形OpenAI GPT-5.6三档与DeepSeek V4的峰谷定价几乎同步出现标志着头部大模型厂商已集体从卷能力、卷价格进入分场景定价算力调度的新阶段谁能更精细化地把TCO总拥有成本讲圆、谁就能拿到B端真正的大单。第三条是基础设施侧的国家级下注韩国4755万亿韩元、三星单笔2655万亿、SK海力士HBM4E与NVIDIA的多年期合作本质上是用主权级资本开支对冲AI算力即定价权的风险预示着2026下半年到2027年AI算力链将从公司间博弈升级为国家间博弈。综合判断下一阶段值得关注的三件事① DeepSeek V4正式版发布时的实际推理价格与生态接入② 国务院AI行动方案的细则落地时间表③ 韩国半导体投资计划对HBM全球定价的二次传导。短期看AI产业链仍处于政策模型终端三浪叠加的红利期但下半年需警惕峰谷定价在B端传导不及预期、出口管制在更多模型上复制的双重风险。Three main narratives define the AI industry map on June 30. First, policy and capital resonance: China’s State Council elevated “AI innovation breakthroughs plus ultra-large intelligent compute clusters” to the top of its June 29 executive-meeting agenda, and the A-share AI complex immediately converted that expectation into daily-limit moves and 8-up-in-11-days ETF runs. Capital voted with real money, signalling that this is not thematic speculation but a hard convergence of policy, corporate cash flows and industry prosperity in a tight time window. Second, “scenario-based pricing” is crystallizing at the model layer: OpenAI’s three-tier GPT-5.6 and DeepSeek V4’s peak-valley pricing arrived almost simultaneously, marking the moment top-tier large-model vendors collectively moved from competing on capability and headline price to competing on scenario-based pricing and compute scheduling. Whoever can articulate TCO most precisely wins the real enterprise deals. Third, sovereign-level bets on infrastructure: Korea’s KRW 4,755 trillion package, Samsung’s solo KRW 2,655 trillion plan, and the SK hynix-NVIDIA multi-year HBM pact are essentially sovereign-grade capex hedging against “compute as pricing power,” telegraphing that from H2 2026 into 2027 the AI compute chain escalates from corporate tussles to state-on-state competition. Three things to watch next: (1) DeepSeek V4’s actual inference pricing and ecosystem integrations at formal release, (2) the implementation timetable for the State Council’s “AI Plus” action plan, and (3) how Korea’s semiconductor investment plan transmits secondarily into global HBM pricing. In the near term, AI’s industrial chain stays in the dividend phase of “policy plus model plus terminal” triple waves, but the back half of the year needs to guard against two risks: peak-valley pricing failing to transmit into enterprise contracts, and export-control measures replicating across more model families.参考来源英文来源TechCrunch: Asian AI startups launch Mythos-like models as Anthropic’s export ban drags onTechCrunch: Why Wall Street thinks US memory maker Micron is the next NvidiaSK hynix Newsroom: HBM4E and NVIDIA partnership announcementsThe Verge: AI labels and online platform coverage (June 28)Reuters Technology: AI infrastructure coverageAsia Market Daily: Samsung and SK Hynix to Unveil $650B Chip Investment中文来源同花顺财经李强主持召开国务院常务会议听取人工智能发展情况汇报央视财经政策加码AI创新突破6月30日腾讯财经DeepSeek-V4正式版官宣7月上线6月30日腾讯财经AI早报 三星计划在韩国投资2655万亿韩元6月30日钛媒体智元第15000台通用具身机器人正式量产下线澎湃新闻智元第15000台具身智能机器人下线腾讯财经谷歌云将提供用于科学研究的专业AI模型6月29日腾讯财经MWC26上海闭幕AI与6G驱动通信迈向智能协同36氪ChinaJoy 2026定档7月31日主题与AI同游新华网2026年ChinaJoy将于上海启幕,人形机器人成年度核心看点证券时报操盘必读 影响股市利好或利空消息2026年6月30日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6月30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