[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-moonshot-kimi-k3-cloud-revenue-share-talks":3,"topics-all":38,"news-related-5a6c3aea-9376-4a3d-a79f-a0d19406e299":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"5a6c3aea-9376-4a3d-a79f-a0d19406e299","月之暗面 Kimi K3 想从微软、AWS、Google 手里分一杯羹：中美大模型的收益分成时代","月之暗面与微软 Azure、亚马逊 AWS、Google 云就 Kimi K3 托管收入展开分成谈判,要求最高 30%,若达成将成为首个中美大型 AI 模型收益分成协议,反映美企对中国低价的强需求。","中美大模型之间的商业关系正在越过「API 转售」这个粗放阶段,进入真金白银的收益分成谈判。据 Solidot 转载 Reuters 报道,月之暗面(Moonshot AI)正与微软 Azure、亚马逊 AWS、Google 云就其旗舰模型 Kimi K3 的托管服务收入分成展开谈判,寻求从 K3 相关 API 与托管收入中获取最高 30% 的分成(原文:https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85201)。\n\n## 谈判走到哪一步\n\n按 Solidot 转述的说法,这次谈判仍处于早期阶段,是否最终能落地还存在不确定性。但只要任何一家美方巨头点头,这场合作都将创下中美之间第一个大型 AI 模型收益分成协议的记录。一个月之暗面这样的中国头部模型厂商,愿意把核心收入分给美国云厂商,而不是继续走自己直接卖 API 的旧路,这件事本身就值得琢磨。\n\n## 为什么美国云愿意坐下来谈\n\n驱动这场谈判的力量,是中国模型在价格上的代差优势。最近 Ramp 收集的 7 万家美国企业支出数据显示,Anthropic 最强模型 Fable 5 的输入价是每百万 token 10 美元,而 DeepSeek V4-Flash 这一类走「Flash」路线的中文模型已经把输入价压到 0.14 美元——前者是后者 70 倍以上(Reuters 报道经 Solidot 转载)。在这种价格落差下,即便加上 30% 的分成,云厂商把 K3 当作主打商品去卖,毛利模型仍然跑得过自研前沿模型。\n\n对中国模型厂商来说,这是一条用股权换算力的捷径。K3 在国内推理算力本来就吃紧,把一部分流量分流到 AWS \u002F Azure \u002F Google Cloud 的海外集群上,既能避开国内 GPU 紧缺的天花板,又能借助美方云的销售网络直接触达美国企业客户。\n\n## 一个新的「上游 — 下游」关系\n\n过去两年,中美 AI 产业链的关系,大致是「中国出模型,美国出货架」(Hugging Face、英伟达、AWS Bedrock),云厂商只是货架方,模型厂商拿不到分成。这次月之暗面想要的,是把云厂商从货架方变成渠道方,自己按比例抽成。这对所有非美国本土模型都是个风向标——Qwen、DeepSeek、智谱 GLM 的下一步商业路径,大概率也会走类似的「托管 + 分成」打法。\n\n而对 AWS、Azure、Google Cloud 来说,拿下 Kimi K3 也意味着补齐自己模型市场里那块「开源 \u002F 廉价 \u002F 中文友好」的版图,把那些因为价格被前沿模型挡在门外的企业客户重新捞回来。\n\n## 所以呢\n\n这场谈判还没签,变量还很多——出口管制、合规审查、定价权归属,任何一个点都能拖很久。但可以确定的是,中国大模型已经不再满足于卖 API 收 token 费这种苦哈哈的商业模式,它们开始学习游戏厂商的发行分成打法,把云渠道纳入自己的收入结构。对企业 IT 采购来说,接下来 12 个月里出现「Fable 5 + Opus 5 + Kimi K3」这种把中美最强模型装在同一个云控制台里的组合,概率正在变大——价格博弈的下半场,云渠道才是真正的牌桌。","https:\u002F\u002Fwww.reuters.com\u002Fbusiness\u002Fretail-consumer\u002Fchinas-moonshot-talks-with-microsoft-amazon-google-over-k3-revenue-sharing-2026-08-26\u002F","d59894d3-308e-4fd8-8865-86dc1eeac4a2",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"8ddf2b28-0234-41a4-9862-3f0faef96472","market-analysis",{"id":22,"name":23,"slug":23,"description":14,"color":14},"95d7995a-fddb-47ba-b8e6-e976ac65414b","strategy",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"ec16a166-4a0b-4fbd-ab6e-a0857b9e31c4","en","Moonshot's Kimi K3 Wants a Cut from Microsoft, AWS, and Google: The China-U.S. LLM Revenue-Share Era","Moonshot AI is negotiating with Microsoft Azure, Amazon AWS, and Google Cloud over a revenue split on Kimi K3 hosting, seeking up to 30%; if it lands, this would be the first major China-U.S. AI model revenue-sharing deal, reflecting strong U.S. enterprise demand for cheaper Chinese models.","The business relationship between Chinese and American large language models is moving past the rough \"API resale\" era and into a real revenue-sharing negotiation. According to a Reuters report carried by Solidot, Moonshot AI is in talks with Microsoft Azure, Amazon AWS, and Google Cloud over a revenue split on its flagship Kimi K3 hosting services, seeking up to 30% of revenue from K3-related APIs and managed deployments (original: https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85201).\n\n## How far the talks have gone\n\nAs relayed by Solidot, the negotiations are still at an early stage and whether they will land remains uncertain. But if any one of the U.S. cloud giants signs on, this would mark the first major AI-model revenue-sharing agreement between China and the United States. The fact that a Chinese frontier model house like Moonshot is willing to hand over part of its core revenue to an American cloud provider, rather than continuing to sell API access directly, is itself worth examining.\n\n## Why American clouds are willing to sit down\n\nThe force driving these talks is the price differential of Chinese models. Recent Ramp data covering 70,000 U.S. businesses shows Anthropic's flagship Fable 5 priced at $10 per million input tokens, while Flash-class Chinese models such as DeepSeek V4-Flash have pushed input pricing down to $0.14 per million tokens—a more than 70x gap (Reuters report via Solidot). Even after taking a 30% cut, a cloud provider selling K3 as a flagship SKU still runs a healthier margin than its own frontier models.\n\nFor the Chinese model vendor, this is a shortcut to trading equity for compute. K3 is already tight on domestic inference capacity; offloading a slice of traffic to AWS, Azure, and Google Cloud's overseas clusters bypasses China's GPU ceiling while leveraging U.S. cloud sales channels to reach American enterprise customers directly.\n\n## A new upstream-downstream relationship\n\nFor the past two years, the China-U.S. AI supply chain has roughly looked like \"China ships the models, the U.S. ships the shelf\" (Hugging Face, NVIDIA, AWS Bedrock)—the cloud is just the shelf, and the model vendor never gets a cut. What Moonshot wants now is to convert the cloud from shelf to channel, taking a percentage on its own. This is a signal for every non-U.S. model house: Qwen, DeepSeek, Zhipu GLM will likely walk a similar \"hosting-plus-revenue-share\" path next.\n\nFor AWS, Azure, and Google Cloud, winning Kimi K3 means filling the missing \"open-source, cheap, Chinese-friendly\" tile in their model marketplaces—pulling back the enterprise customers who have been priced out of frontier-tier models.\n\n## So what\n\nThe deal isn't signed, and there are many variables left—export controls, compliance review, pricing power. Any one of them can drag this out for a long time. But what's clear is that Chinese frontier models no longer want to make money the old, hard way of charging per-token API fees; they are starting to adopt the game industry's distribution-revenue-share playbook and fold cloud channels into their revenue structure. For enterprise IT procurement, the odds are rising that within the next 12 months we'll see combos like \"Fable 5 + Opus 5 + Kimi K3\" sitting under the same cloud console—in the second half of the pricing war, the cloud channel is the real table.","moonshot-kimi-k3-cloud-revenue-share-talks","2026-08-28T12:30:00Z","2026-08-28T05:04:57.453021Z","2026-08-28T05:04:57.453038Z",true,"agent",169,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"b7668f43-05a7-46d7-845f-27e70fcaceec","Mozilla 报告:中国开放权重距美国前沿模型只差 4.4 个月","mozilla-open-weights-4-month-gap","2026-09-16T13:08:54+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"8d7b30e0-996f-4141-8501-8f464bda6282","中美开放权重参数上限差距拉到 20 倍:Hugging Face 夏季报告里的三条隐藏数据","hugging-face-summer-2026-frontier-ceiling","2026-08-22T14:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"5f4d9df3-673b-4440-9a77-6e8f0e697681","苹果自训中国区专用大模型:放弃「借模型」路线,联合阿里亲自下场","apple-china-specific-llm-alibaba","2026-08-14T19:10:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"03c760d2-6603-4ac8-8773-8236330bc019","白宫豁免中国开放权重模型:开源路线获得 AI 安全审查「白名单」","us-carve-out-chinese-open-weight-ai-2026","2026-08-07T02:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"ad3be632-49a1-44f5-9816-62c168e56467","全球大模型调用量榜前五全是\"中国造\":开源 MoE 正在重写 OpenRouter 的地理坐标","openrouter-top5-china-moe-open-source-2026w31","2026-08-02T03:30:00+00:00"]