[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-nvidia-microsoft-agentic-ai-pc-cloud-local-zh":3,"article-related-nvidia-microsoft-agentic-ai-pc-cloud-local-zh":35,"series-industry-03788c2e-f066-4e34-bd65-3e7f0c70c3b9":88},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":27,"views":31,"created_at":32,"published_at":33,"topic_cluster_id":34},"03788c2e-f066-4e34-bd65-3e7f0c70c3b9","nvidia-microsoft-agentic-ai-pc-cloud-local-zh","NVIDIA 與微軟把代理式 AI 串到雲端與 PC","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Ftag\u002Fnvidia\">NVIDIA\u003C\u002Fa> 與微軟把代理式 AI 串到 Windows PC、雲端與地端，讓開發、推論和部署可以在同一套架構裡完成。\u003C\u002Fp>\u003Cp>這份清單看完，你可以快速判斷 5 種部署路線該選哪一種：是先在個人電腦上做原型、在企業工作站跑重模型，還是直接走 Azure、地端或資料中心級基礎設施。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>運行位置\u003C\u002Fth>\u003Cth>關鍵規格\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>RTX Spark\u003C\u002Ftd>\u003Ctd>Windows 裝置\u003C\u002Ftd>\u003Ctd>1 petaflop AI 效能，最高 128GB 統一記憶體\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>DGX Station for Windows\u003C\u002Ftd>\u003Ctd>Windows 桌機\u003C\u002Ftd>\u003Ctd>最高 20 petaflops FP4，最高 748GB 一致性記憶體\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Microsoft Fabric Data Warehouse\u003C\u002Ftd>\u003Ctd>雲端資料層\u003C\u002Ftd>\u003Ctd>SQL 執行最高比 CPU 基準快 6 倍\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Azure Local 搭配 RTX PRO 6000 Blackwell Server Edition\u003C\u002Ftd>\u003Ctd>地端與主權部署\u003C\u002Ftd>\u003Ctd>支援多節點與 vLLM 執行環境\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Vera Rubin on Azure\u003C\u002Ftd>\u003Ctd>AI 工廠\u003C\u002Ftd>\u003Ctd>每百萬瓦推論吞吐量最高提升 10 倍\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. RTX Spark：把代理式 AI 放進 Windows 個人機\u003C\u002Fh2>\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.nvidia.com\u002F\">NVIDIA\u003C\u002Fa> 與 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fzh-tw\">微軟\u003C\u002Fa>把 RTX Spark 定位成給開發者用的個人 AI 機器，\u003Ca href=\"\u002Fnews\u002Fturbovec-cuts-vector-ram-to-4gb-zh\">重點\u003C\u002Fa>不是取代資料中心，而是讓代理式 AI 可以先在本機完成建置、調校與執行。對想在 Windows 上直接跑本地代理的人來說，這是最接近「隨手可用」的路線。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781275694192-cr3e.png\" alt=\"NVIDIA 與微軟把代理式 AI 串到雲端與 PC\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>它的規格很明確：1 petaflop AI 效能、最高 128GB 統一\u003Ca href=\"\u002Fnews\u002Fnvidia-sk-hynix-memory-ai-factory-partnership-zh\">記憶體\u003C\u002Fa>，還主打未插電也能維持完整 AI 與圖形效能。這讓它適合小型團隊、獨立開發者，或需要常帶著走的原型機。\u003C\u002Fp>\u003Cul>\u003Cli>支援 CUDA、RTX、DLSS、TensorRT\u003C\u002Fli>\u003Cli>將由 Microsoft Surface、ASUS、Dell、HP、Lenovo、MSI 等品牌推出\u003C\u002Fli>\u003Cli>適合本機測試提示詞、工具呼叫與輕量代理流程\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. DGX Station for Windows：把重模型搬到桌邊\u003C\u002Fh2>\u003Cp>如果 RTX Spark 是個人機，DGX Station for Windows 就是面向企業工作流的桌邊級方案。它的定位是長時間運作的代理與模型開發環境，特別適合需要在本機處理更大模型、更多\u003Ca href=\"\u002Fnews\u002Fturboquant-makes-long-context-ai-cheaper-zh\">上下文\u003C\u002Fa>與更重推論負載的團隊。\u003C\u002Fp>\u003Cp>官方釋出的數字也更像工作站而不是一般 PC：GB300 Grace Blackwell Ultra Desktop Superchip、最高 748GB 一致性記憶體、20 petaflops FP4 效能。這代表它能承接更接近前沿模型的本地推論與開發需求。\u003C\u002Fp>\u003Cul>\u003Cli>預計由 ASUS、Dell、GIGABYTE、HP、MSI、Supermicro 在第 4 季推出\u003C\u002Fli>\u003Cli>支援最高 1 兆參數模型\u003C\u002Fli>\u003Cli>搭載 NVIDIA OpenShell 安全執行環境\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. Microsoft Foundry：把多家模型放進同一個代理平台\u003C\u002Fh2>\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fzh-tw\u002F\">Microsoft\u003C\u002Fa> 的 Foundry 正在變成企業組裝代理系統的中樞，而不是只放單一模型的展示櫃。NVIDIA 表示，它的開放模型已經和 \u003Ca href=\"\u002Ftag\u002Fanthropic\">Anthropic\u003C\u002Fa>、\u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa> 模型一起進入 Foundry \u003Ca href=\"\u002Ftag\u002Fagent\">Agent\u003C\u002Fa> Service，企業可以在同一套治理與身分控管下組合不同能力。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781275695776-jdbz.png\" alt=\"NVIDIA 與微軟把代理式 AI 串到雲端與 PC\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>其中最受注意的是 Nemotron 3 Ultra，這是一款面向程式設計、研究與企業工作流的開放前沿推理模型。再加上 Nemotron 3.5 ASR、Nemotron 3.5 Content Safety、Cosmos 3 與 Earth-2，Foundry 的角色就不只是聊天，而是把語音、內容安全、物理 AI、天氣預測都納進代理流程。\u003C\u002Fp>\u003Cul>\u003Cli>支援管理式運算與企業治理\u003C\u002Fli>\u003Cli>可搭配 Agent Toolkit 與 NemoClaw 藍圖做生產部署\u003C\u002Fli>\u003Cli>Anthropic Claude 也可在 Azure 的 NVIDIA GB300 Blackwell Ultra 系統上原生執行\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>4. Fabric 與 Azure Local：資料速度和地端控制一起補齊\u003C\u002Fh2>\u003Cp>代理式 AI 會一直查資料、反覆推理、重複檢索，所以資料層不能慢。NVIDIA 指出，\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fzh-tw\u002Fmicrosoft-fabric\">Microsoft Fabric\u003C\u002Fa> Data Warehouse 已經整合加速運算，內部測試顯示 SQL 執行速度最高比 CPU 基準快 6 倍，在高併發工作負載下也比另外三家雲端資料倉儲供應商快 7 倍。\u003C\u002Fp>\u003Cp>如果資料不能離開現場，\u003Ca href=\"https:\u002F\u002Fazure.microsoft.com\u002Fzh-tw\u002Fproducts\u002Fazure-local\">Azure Local\u003C\u002Fa> 就是另一條路。微軟把 Foundry Local 帶到 Azure Local，再配上 RTX PRO 6000 Blackwell Server Edition，可支援多節點部署與 vLLM 執行環境，適合製造、能源、主權資料中心與低延遲場景。\u003C\u002Fp>\u003Ccode>適合情境：\u003Cbr>- Fabric Data Warehouse：雲端分析與代理查詢\u003Cbr>- Azure Local：地端、混合雲、主權部署\u003Cbr>- vLLM：需要擴展推論且重視延遲的情境\u003C\u002Fcode>\u003Ch2>5. OpenShell 與 Vera Rubin：安全代理和 AI 工廠一起上線\u003C\u002Fh2>\u003Cp>當代理開始真的去動檔案、網路和憑證，安全就不能只靠提示詞。NVIDIA 的 OpenShell 做法是讓每個代理跑在自己的沙箱裡，所有對外呼叫都先經過政策檢查，再決定能不能碰到資源；它也已整合進 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Ffeatures\u002Fcopilot\">GitHub Copilot\u003C\u002Fa>，並以 Apache 2.0 開源。\u003C\u002Fp>\u003Cp>另一端則是資料中心級擴張。微軟表示 Fairwater Wisconsin 已經上線並通過 NVIDIA Vera Rubin 驗證，這套平台可直接進 Azure，不必大改機房。NVIDIA 宣稱它每百萬瓦推論吞吐量最高可提升 10 倍，對正在規劃 AI 工廠的人來說，這是下一代基礎設施的重點訊號。\u003C\u002Fp>\u003Cul>\u003Cli>OpenShell 採模型無關設計\u003C\u002Fli>\u003Cli>政策可寫成程式碼並納入版本控管\u003C\u002Fli>\u003Cli>Vera Rubin 可與 Blackwell 一起部署在 Azure 資料中心\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>怎麼挑：先看你要把 AI 放在哪裡\u003C\u002Fh2>\u003Cp>如果你要的是本機原型與快速迭代，先看 RTX Spark；如果你要在桌邊跑更重的企業模型，DGX Station for Windows 更合適。若你的重點是資料治理、多模型編排與企業代理平台，\u003Ca href=\"\u002Ftag\u002Fmicrosoft\">Microsoft\u003C\u002Fa> Foundry 加上 Fabric 會是核心。\u003C\u002Fp>\u003Cp>需要地端、混合雲或主權部署時，Azure Local 會比純雲端更實用；若你最在意代理安全，OpenShell 值得優先評估；如果你正在規劃下一輪大規模推論基礎設施，Vera Rubin 與 AI 工廠架構就是最該追的方向。\u003C\u002Fp>","5 個重點看懂 NVIDIA 與微軟如何把代理式 AI 從 Windows PC、Azure 到地端整合，並比較效能、記憶體與部署場景。","blogs.nvidia.com","https:\u002F\u002Fblogs.nvidia.com\u002Fblog\u002Fmicrosoft-build-windows-local-cloud-devices\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781275694192-cr3e.png","industry","zh","5c8f62ea-907e-41e9-9a91-7596c457c804",[17,18,19,20,21,22,23,24,25,26],"NVIDIA","微軟","代理式 AI","Windows","Azure","Microsoft Foundry","Microsoft Fabric","Azure Local","OpenShell","Vera Rubin",[28,29,30],"RTX Spark 適合本機開發與個人代理，重點是 Windows 端的低門檻部署。","DGX Station for Windows 面向更重的企業工作流，記憶體與算力都明顯升級。","Fabric、Azure Local、OpenShell 與 Vera Rubin 分別補上資料速度、地端控制、安全與資料中心擴充。",2,"2026-06-12T14:47:45.233052+00:00","2026-06-12T14:47:45.207+00:00","fa1dc5e8-0eec-4179-8dc0-e35a3d82f701",{"tags":36,"relatedLang":47,"relatedPosts":51},[37,38,40,42,45],{"name":18,"slug":18},{"name":20,"slug":39},"windows",{"name":19,"slug":41},"代理式-ai",{"name":43,"slug":44},"Nvidia","nvidia",{"name":21,"slug":46},"azure",{"id":15,"slug":48,"title":49,"language":50},"nvidia-microsoft-agentic-ai-pc-cloud-local-en","NVIDIA and Microsoft unify agentic AI from PC to cloud","en",[52,58,64,70,76,82],{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"3e29acbc-3865-4015-ae0f-ac2d17fbea89","google-gemini-outage-error-1076-june-2026-zh","Google Gemini 出現 error 1076 當機","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781338671050-izsk.png","2026-06-13T08:17:27.225238+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"ebbd7c3b-23a7-4b31-9bae-1a8fb4dc5eef","nvidia-hugging-face-ai-pipelines-zh","NVIDIA 的 Hugging Face 5 類模型最適合誰","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781337771576-6vqm.png","2026-06-13T08:02:19.301779+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"78bb945b-f292-4071-811e-9ac390b68a38","anthropic-public-record-ai-anxiety-policy-zh","Anthropic 把 AI 焦慮變政策","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781327894646-6pyt.png","2026-06-13T05:17:42.429455+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"a69174d1-9768-4144-909a-78ec2517b186","chatgpt-grew-from-chatbot-to-platform-zh","ChatGPT 從聊天機器人變平台","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781325173553-w7ov.png","2026-06-13T04:32:27.586497+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"050bf93c-ddcf-4493-8335-11a67831fcfc","openai-files-confidential-ipo-after-122b-round-zh","OpenAI 密件申請 IPO，估值衝 8520 億","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781323369296-ra5z.png","2026-06-13T04:02:23.888945+00:00",{"id":83,"slug":84,"title":85,"cover_image":86,"image_url":86,"created_at":87,"category":13},"66a93d43-34f4-401b-b8a9-51878e91d60c","government-access-orders-frontier-model-access-zh","政府存取命令就該管住前沿模型存取","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781319763702-t9ak.png","2026-06-13T03:02:19.013704+00:00",[89,94,99,104,109,114,119,124,129,134],{"id":90,"slug":91,"title":92,"created_at":93},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":130,"slug":131,"title":132,"created_at":133},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":135,"slug":136,"title":137,"created_at":138},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]