美媒The Information報道稱，英偉達和AI芯片創業公司d-Matrix宣佈聯合推出AI芯片系統。d-Matrix主打推理，核心方向是低延遲、低功耗的推理加速。d-Matrix的技術路線是以記憶體爲中心的計算，用3DIMC減少資料搬運。AI推理現在最耗錢的地方，是海量請求、KVcache（上下文快取）、長文本和Agent（能自主拆解並執行任務的AI）連續調用，所以把單次調用成本壓下來，就能吃到更高頻的AI應用場景。 定製芯片市場正在被模型公司集體關注。智譜今天被曝評估自研芯片，DeepSeek同期被曝自研推理芯片，OpenAI已經和Broadcom做Jalapeño（LLM推理芯片），Anthropic也被曝和三星討論自研芯片。模型公司現在都在算同一筆帳：訓練繼續依賴頂級GPU，推理要用ASIC（專用芯片）把成本、延遲和能耗壓下去。英偉達這次合作代表開始把推理ASIC接進自己的系統裏。Etched、Cerebras、Groq、d-Matrix這類公司本來都在搶推理負載，英偉達如果能把它們變成異構計算（GPU、CPU、ASIC混合協同）的一部分，讓GPU負責通用算力和生態控制，

浪潮信息一字漲停，公司預計2026年上半年淨利潤26億元-31億元，同比增長226%-288%。

The Ministry of Human Resources and Social Security, the National Development and Reform Commission, the Ministry of Industry and Information Technology and the National Data Administration jointly issued Implementation Opinions to accelerate development of AI applications in the human-resources and social-security (HRSS) sector. The document sets five‑year targets: by 2026 build AI+HRSS application infrastructure and create about 20 application scenarios anchored on industry large models with corresponding high‑quality datasets; by 2027 roll out a batch of HRSS industry large models and intelligent agents and explore roughly 50 high‑value application scenario enablement paths; by 2030 achieve widespread AI application across the HRSS field.