[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-推論成本":3},{"tag":4,"articles":10},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":9},"4f2b3013-a727-4196-b25f-a2f32866dfcd","推論成本",6,"推論成本指的是模型在實際服務時，每次生成、回應或代理執行所消耗的算力、記憶體、延遲與雲端費用。從 GPU\u002FCPU 架構、模型大小到批次與快取策略，這些取捨直接影響 AI 產品能否規模化。","Inference cost is the ongoing compute, memory, latency, and cloud spend required when a model serves real requests. It shapes choices around GPU and CPU architecture, model size, batching, and caching, and it often determines whether AI products can scale economically.",[11,20,28,35,42,49],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"838cb5fd-5651-49fb-9b4c-c2dbde25ca02","claude-opus-45-gpt-parameters-estimate-zh","Claude Opus 4.5 和 GPT 到底多大","GPT-4 常被估到 1.6 兆參數，但 GPT-4o 可能只有 200B 到 300B。Claude Opus 4.5 的真實大小沒公開，重點其實是成本、延遲和效能比。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775207388141-adee.png","zh","2026-04-03T09:09:28.833454+00:00",{"id":21,"slug":22,"title":23,"summary":24,"category":25,"image_url":26,"cover_image":26,"language":18,"created_at":27},"53332d20-ac19-4066-bbfd-4164923130f7","cloud-infrastructure-spend-jumps-ai-demand-zh","AI 需求把雲端支出推高 29%","Q4 2025 全球雲端基礎設施支出達 1109 億美元，年增 29%。AI 訓練、推論與雲端工作負載一起拉高 hyperscaler 投資。","industry","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775200180978-1ufe.png","2026-04-03T07:09:17.443915+00:00",{"id":29,"slug":30,"title":31,"summary":32,"category":25,"image_url":33,"cover_image":33,"language":18,"created_at":34},"343a707c-98b0-45e6-8c1b-5e64948c0705","alibaba-risc-v-ai-cpu-server-chips-zh","阿里巴巴 RISC-V AI CPU 進軍伺服器","阿里巴巴 DAMO 推出 64 位元 RISC-V CPU，最高 3.2 GHz、採 TSMC 5nm，瞄準 agentic AI 與伺服器推論，直接碰 Arm、x86 和 Nvidia 的地盤。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775197501811-3h4f.png","2026-04-03T06:24:43.084283+00:00",{"id":36,"slug":37,"title":38,"summary":39,"category":25,"image_url":40,"cover_image":40,"language":18,"created_at":41},"b6c9a490-84a6-483c-b763-73ff60ca5a91","nvidia-rubin-ai-infrastructure-2026-zh","NVIDIA Rubin 把 AI 基礎設施拉到新尺度","NVIDIA Rubin 以六顆晶片組成平台，主打推論成本最高降 10 倍，並把 Vera Rubin NVL72 推向雲端與企業 AI。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1774497418478-ye2x.png","2026-03-30T06:18:52.965441+00:00",{"id":43,"slug":44,"title":45,"summary":46,"category":25,"image_url":47,"cover_image":47,"language":18,"created_at":48},"32608593-3793-4893-9ddc-cd601910e56b","openai-compute-bet-reshaping-ai-economics-zh","OpenAI 豪賭算力，AI 經濟學變了","OpenAI 年化營收傳出已超過 200 億美元，但晶片、電力與資料中心支出也一路暴衝。需求很真實，問題是帳怎麼算得過去，這正是 AI 產業現在最難迴避的現實。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1774597964781-1feq.png","2026-03-27T01:41:43.897881+00:00",{"id":50,"slug":51,"title":52,"summary":53,"category":25,"image_url":54,"cover_image":54,"language":18,"created_at":55},"9036b6b8-8f30-4e2e-a373-c1d185f59309","openai-2026-cash-crunch-and-growth-bet-zh","OpenAI 2026 燒錢壓力變難忽視","OpenAI 可能在 2026 年衝出高營收，但算力、推論成本、廣告計畫與競爭壓力也一起放大。問題不在成長夠不夠快，而是收入能不能追上 GPU、資料中心與企業銷售的帳單。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1774597806922-1yiw.png","2026-03-27T01:36:49.476412+00:00"]