[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-vector-lakebase-milvus-ai-data-platform-zh":3,"article-related-vector-lakebase-milvus-ai-data-platform-zh":33,"series-industry-ef090e43-8215-497b-9bdd-c43d3d611927":85},{"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":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":32},"ef090e43-8215-497b-9bdd-c43d3d611927","vector-lakebase-milvus-ai-data-platform-zh","Vector Lakebase 把 Milvus 變成 AI 資料平台","\u003Cp data-speakable=\"summary\">\u003Ca href=\"https:\u002F\u002Fzilliz.com\u002F\">Zilliz\u003C\u002Fa> 的 Vector Lakebase 把向量搜尋、探索和批次分析放在同一個資料底座上，讓 AI 團隊少搬資料、多做決策。\u003C\u002Fp>\u003Cp>這次 public preview 的\u003Ca href=\"\u002Fnews\u002Fjensen-huang-lg-ai-cooperation-five-bets-zh\">重點\u003C\u002Fa>，不是再多一個向量資料庫，而是把原本分散的 serving、discovery、analytics 收進同一套架構。對需要在成本、延遲和資料流轉之間取捨的團隊來說，讀完這 5 項後，你大致就能判斷自己該選哪種層級、哪種工作模式，避免把時間花在重複搬運向量資料上。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>延遲\u003C\u002Fth>\u003Cth>QPS\u003C\u002Fth>\u003Cth>Recall\u003C\u002Fth>\u003Cth>儲存／運算模型\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>Performance-Optimized\u003C\u002Ftd>\u003Ctd>單位數毫秒\u003C\u002Ftd>\u003Ctd>1,000 以上\u003C\u002Ftd>\u003Ctd>95-98%，可調到 99%+\u003C\u002Ftd>\u003Ctd>In-memory\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Capacity-Optimized\u003C\u002Ftd>\u003Ctd>100ms 以下\u003C\u002Ftd>\u003Ctd>100-500\u003C\u002Ftd>\u003Ctd>95-98%，可調到 99%+\u003C\u002Ftd>\u003Ctd>Memory + NVMe\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Tiered-Storage\u003C\u002Ftd>\u003Ctd>約 100ms\u003C\u002Ftd>\u003Ctd>10-50\u003C\u002Ftd>\u003Ctd>95-98%，可調到 99%+\u003C\u002Ftd>\u003Ctd>Memory + NVMe + object storage\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>On-Demand Search\u003C\u002Ftd>\u003Ctd>依工作負載而定\u003C\u002Ftd>\u003Ctd>依工作負載而定\u003C\u002Ftd>\u003Ctd>未公開\u003C\u002Ftd>\u003Ctd>Compute 啟用時才付費\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. 三層即時服務，先把成本和速度分開看\u003C\u002Fh2>\u003Cp>Vector Lakebase 仍然是以生產級向量搜尋為核心，但 Zilliz 把它拆成三個 serving 層，讓團隊按延遲與吞吐量選配置。這比單一規格更實際，因為不是每個 AI 應用都需要同樣的熱資料和同樣的速度。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781410666114-09nr.png\" alt=\"Vector Lakebase 把 Milvus 變成 AI 資料平台\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>如果你在做 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> memory、\u003Ca href=\"\u002Ftag\u002Frag\">RAG\u003C\u002Fa> 或面向使用者的語意搜尋，這種分層很有用。熱查詢可以買速度，穩定流量可以換成本，資料團隊不用為了少數高峰時段一直維持最高規格。\u003C\u002Fp>\u003Cul>\u003Cli>Performance-Optimized：1,000+ QPS，單位數毫秒延遲\u003C\u002Fli>\u003Cli>Capacity-Optimized：100-500 QPS，100ms 以下延遲\u003C\u002Fli>\u003Cli>Tiered-Storage：10-50 QPS，約 100ms 延遲\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. 零拷貝語意資料平面，少一次搬運就少一次延遲\u003C\u002Fh2>\u003Cp>Zilliz 的主張很\u003Ca href=\"\u002Fnews\u002Fmastercard-opens-ai-payments-stablecoins-zh\">直接\u003C\u002Fa>：不要把數十億筆向量在不同系統間來回複製。Vector Lakebase 讓 serving、探索和分析都對同一份邏輯資料操作，底層則共享 lake-native storage，減少多套管線之間的同步成本。\u003C\u002Fp>\u003Cp>這種設計最有感的地方，是縮短從資料進入到模型改善的迴圈。同一份資料可以同時支援 production query、探索式分析和訓練資料準備，不必先 ETL 到另一個平台再重做一次。\u003C\u002Fp>\u003Cul>\u003Cli>一份邏輯資料，三種工作模式共用\u003C\u002Fli>\u003Cli>共享 lake-native storage\u003C\u002Fli>\u003Cli>適合 gigabytes 到 petabytes 規模\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. On-Demand Search，讓間歇性工作不用常駐開機\u003C\u002Fh2>\u003Cp>On-Demand Search 是這次最偏成本控制的功能。Zilliz 說，團隊可以直接在外部 data lake 上做搜尋，只在 object storage 和實際啟用的 compute 上付費，不必為了偶爾的查詢把基礎設施長時間維持在待命狀態。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781410664520-8fea.png\" alt=\"Vector Lakebase 把 Milvus 變成 AI 資料平台\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>對流量不均的工作很合適，例如離線 enrichment、一次性調查、夜間批次任務，或是需要查但不想搬資料的情境。這也讓資料可以留在原本的位置，平台只負責把搜尋能力接上去。\u003C\u002Fp>\u003Ccode>適用情境：ad hoc semantic search、overnight deduplication、periodic embedding refresh、lake query without data copies\u003C\u002Fcode>\u003Ch2>4. Interactive Discovery，把探索留在同一套底座上\u003C\u002Fh2>\u003Cp>Interactive Discovery 介於 production serving 和 offline analytics 之間。它的價值不是再做一個分析工具，而是讓資料科學家和 ML \u003Ca href=\"\u002Fnews\u002Fllm-research-engineers-post-training-services-zh\">工程師\u003C\u002Fa>可以直接在同一份向量資料上做探索，不必先搬到另一個 stack。\u003C\u002Fp>\u003Cp>如果團隊常要看 cluster 分布、比較 retrieval 行為，或在 retrain 前先找弱標籤，這個模式會比傳統切資料、匯出、再分析的流程順很多。探索和服務共用底座，也比較不容易出現資料版本不一致。\u003C\u002Fp>\u003Cul>\u003Cli>適合互動式分析\u003C\u002Fli>\u003Cli>與 production search 使用同一份資料\u003C\u002Fli>\u003Cli>減少分析團隊與 serving 團隊之間的交接成本\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. Batch Analytics，補上 AI 迴圈裡的資料整理段\u003C\u002Fh2>\u003Cp>Zilliz 把 AI 系統描述成一個循環：服務、學習、改善資料，再回到服務。Batch Analytics 就是負責「改善資料」的那一段，尤其適合處理大量語料、準備訓練集，或在規模上做離線加工。\u003C\u002Fp>\u003Cp>真正的意義在於，批次作業不必再依賴另一套資料管線。若同一批向量同時支援即時搜尋與離線處理，團隊就能把 feedback loop 收得更緊，少做重複同步，也少一層故障點。\u003C\u002Fp>\u003Cul>\u003Cli>Semantic deduplication\u003C\u002Fli>\u003Cli>Multi-petabyte training-data prep\u003C\u002Fli>\u003Cli>大型離線處理可直接跑在同一底座上\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>怎麼挑，才不會買錯層級\u003C\u002Fh2>\u003Cp>如果你的應用吃的是即時回應和高查詢量，先看 Performance-Optimized。若你想在成本和反應速度之間找平衡，Capacity-Optimized 通常比較好落地；流量穩定、比較在意儲存成本時，Tiered-Storage 會更合理。\u003C\u002Fp>\u003Cp>On-Demand Search 適合尖峰不固定的工作，Interactive Discovery 適合資料科學和分析團隊，Batch Analytics 則適合想把 serve-learn-improve 迴圈留在同一平台的 AI 團隊。多數情況下，真正該問的不是哪一項最強，而是哪一種組合能少掉最多重複基礎設施。\u003C\u002Fp>","5 種能力看 Zilliz Vector Lakebase 如何把即時服務、探索與批次分析整合到同一個 AI 資料底座。","www.prnewswire.com","https:\u002F\u002Fwww.prnewswire.com\u002Fnews-releases\u002Fzilliz-launches-vector-lakebase-extending-the-worlds-most-adopted-vector-database-into-a-unified-data-platform-for-ai-302796419.html",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781410666114-09nr.png","industry","zh","23b08a3d-b593-4447-82a7-1ed322336d19",[17,18,19,20,21,22,23,24],"Zilliz","Vector Lakebase","Milvus","向量資料庫","AI 資料平台","vector search","batch analytics","interactive discovery",[26,27,28],"三層即時服務讓團隊按延遲、吞吐與成本選規格。","零拷貝資料平面把 serving、探索和批次分析放在同一份資料上。","On-Demand Search、Discovery、Batch Analytics 分別對應間歇查詢、互動分析與資料整理。",0,"2026-06-14T04:17:19.97915+00:00","2026-06-14T04:17:19.973+00:00","2ee38a94-afaa-4784-992e-f1a5a362782e",{"tags":34,"relatedLang":44,"relatedPosts":48},[35,37,39,41,43],{"name":17,"slug":36},"zilliz",{"name":21,"slug":38},"ai-資料平台",{"name":18,"slug":40},"vector-lakebase",{"name":19,"slug":42},"milvus",{"name":20,"slug":20},{"id":15,"slug":45,"title":46,"language":47},"vector-lakebase-milvus-ai-data-platform-en","Vector Lakebase makes Milvus a full AI data platform","en",[49,55,61,67,73,79],{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"category":13},"20b743fc-22d8-4d4b-a063-81e5d792bc07","white-house-ai-action-plan-bitcoin-stablecoins-zh","$2.7 兆：白宮把 AI、比特幣、穩定幣綁在一起","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781424165506-3whs.png","2026-06-14T08:02:21.217359+00:00",{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"f0725dcd-5415-4280-b680-32ca1a54277f","cloudflare-q1-2026-revenue-growth-analyst-upgrades-zh","Cloudflare Q1 營收成長 34%","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781421470388-edry.png","2026-06-14T07:17:27.263939+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":13},"53035011-4d58-40e3-b273-552c33ef6c00","zilliz-vector-lakebase-unified-ai-data-stack-zh","Zilliz Vector Lakebase 不是加功能，而是在壓縮 AI 資…","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781412461503-bnm0.png","2026-06-14T04:47:17.905903+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":13},"1479dc4b-5684-4a0c-83f4-ee09595092bf","mastercard-opens-ai-payments-stablecoins-zh","Mastercard 讓 AI 直接付穩定幣","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781408867181-ad8z.png","2026-06-14T03:47:21.631777+00:00",{"id":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"category":13},"d88f40d6-a3d3-4bf5-b149-bf17bcaf009b","5-ji-jie-kan-chuan-zhe-zhi-jia-ren-xing-ji-qi-ren-ying-pian-zh","5 個細節看穿這支假人形機器人影片","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781406168128-28g0.png","2026-06-14T03:02:20.268133+00:00",{"id":80,"slug":81,"title":82,"cover_image":83,"image_url":83,"created_at":84,"category":13},"1e72405b-51cf-4651-9957-a87f067fd43b","jensen-huang-lg-ai-cooperation-five-bets-zh","黃仁勳與 LG 的 5 個 AI 合作重點","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781399872213-o4jc.png","2026-06-14T01:17:22.011305+00:00",[86,91,96,101,106,111,116,121,126,131],{"id":87,"slug":88,"title":89,"created_at":90},"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":92,"slug":93,"title":94,"created_at":95},"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":97,"slug":98,"title":99,"created_at":100},"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":102,"slug":103,"title":104,"created_at":105},"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":107,"slug":108,"title":109,"created_at":110},"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":112,"slug":113,"title":114,"created_at":115},"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":117,"slug":118,"title":119,"created_at":120},"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":122,"slug":123,"title":124,"created_at":125},"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":127,"slug":128,"title":129,"created_at":130},"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":132,"slug":133,"title":134,"created_at":135},"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"]