[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-49a1ed2e-9f37-478b-9f8d-f349d3cba5ff":3},{"id":4,"title":5,"summary":6,"original_url":7,"source_id":8,"tags":9,"published_at":23,"created_at":24,"modified_at":25,"is_published":26,"publish_type":27,"image_url":13,"view_count":28},"49a1ed2e-9f37-478b-9f8d-f349d3cba5ff","Google 对 Meta 的 Gemini 配额按下暂停键:算力配给时代来临","《金融时报》上周披露,Google 已对 Meta 使用 Gemini 模型设限——原因是 Meta 想要的算力超出了 Google 能交付的能力,与商业竞争无关。\n\n据 FT 援引知情人士,Google 约在今年 3 月通知 Meta,无法满足其希望采购的 Gemini 算力规模,导致 Meta 部分内部 AI 项目被迫延期。Meta 受冲击最大,其他 Google 大客户也受影响,但程度较轻。Meta 内部随后开始鼓励员工更高效使用 token。\n\n信号其实早埋下:Google Cloud 一季度财报里,CEO Sundar Pichai 已承认算力约束限制了 Google Cloud 的更高增长,并导致 backlog 环比近乎翻倍。换言之,Google 自己也在被算力卡脖子,没有余力把最稀缺资源全分给对手。\n\nAI 行业正从「芯片紧缺」过渡到「推理算力紧缺」。即便头部厂商每年向数据中心投入数百亿美元,前沿模型推理所需的 HBM、加速卡、互连带宽依然供不应求,容量调度正在变成隐形的权力杠杆。自研模型未跑顺、又大量调用外部 API 的厂商,首当其冲。\n\n对开发者,两个现实正在浮现:头部模型 API 不再是「开箱即用、永不饱和」的公共资源,容量规划需要前置到产品设计阶段;token 效率、推理压缩、小模型蒸馏等技术从「加分项」变成「必修课」——这正是 InfoKV、KV 缓存压缩、长上下文稀疏注意力等一批工作今年活跃的根因。\n\nGoogle 给 Meta 的不是禁令,是一份算力紧缺的行业通知书。下半年竞争焦点,正从 benchmark 跑分转向谁能更稳地交付算力。","https:\u002F\u002Fwww.cnbc.com\u002F2026\u002F06\u002F28\u002Fgoogle-limits-metas-use-of-its-gemini-ai-models-ft-reports.html","b506c01d-ef58-49cc-8ba1-351a47e7d6d1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a9524a82-a7c5-4daa-bb4b-a7ee77bb0b94","gemini",{"id":18,"name":19,"slug":19,"description":13,"color":13},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm","2026-06-28T22:00:00Z","2026-06-28T22:06:57.254711Z","2026-06-28T22:06:57.254721Z",true,"agent",3]