Snowflake 和 Google Cloud 將 Gemini 3 原生集成到 Snowflake Cortex AI 中，從而徹底改變企業數據策略，無需移動數據即可實現“企業推理”。
NEW YORK, NY – January 6, 2026 – In a move set to redefine corporate data strategies, Snowflake (NYSE: SNOW) and Google Cloud (NASDAQ: GOOGL) have announced a significant expansion of their partnership. This collaboration introduces Google's cutting-edge Gemini 3 model family directly into Snowflake Cortex AI. The integration promises to unlock advanced artificial intelligence capabilities for enterprises within their secure data environments, effectively dissolving the traditional barriers of security concerns and latency often associated with external AI APIs.
紐約州紐約 – 2026 年 1 月 6 日 – 為了重新定義企業數據戰略，Snowflake（紐約證券交易所股票代碼：SNOW）和 Google Cloud（納斯達克股票代碼：GOOGL）宣布大幅擴展雙方的合作夥伴關係。此次合作將 Google 尖端的 Gemini 3 模型系列直接引入 Snowflake Cortex AI。此次集成有望在企業的安全數據環境中解鎖先進的人工智能功能，有效地解決通常與外部人工智能 API 相關的安全問題和延遲的傳統障礙。
Enterprise Reasoning: The Dawn of Zero Data Movement AI
企業推理：零數據移動人工智能的黎明
The core innovation lies in embedding Gemini 3 Pro and Gemini 2.5 Flash into the Snowflake platform, enabling what the companies term "Enterprise Reasoning." This capability allows AI to perform complex, multi-step logical analyses on vast internal datasets without the data ever needing to leave Snowflake's robust security perimeter. This "Zero Data Movement" architecture directly addresses a primary concern for C-suite executives: harnessing the power of generative AI while ensuring absolute control over sensitive corporate intellectual property. The implications are profound, allowing for advanced tasks like complex financial reconciliations and legal audits to be performed with unprecedented reliability.
核心創新在於將 Gemini 3 Pro 和 Gemini 2.5 Flash 嵌入到 Snowflake 平台中，從而實現公司所說的“企業推理”。此功能允許人工智能對大量內部數據集執行複雜的多步驟邏輯分析，而數據無需離開 Snowflake 強大的安全邊界。這種“零數據移動”架構直接解決了高管層的主要擔憂：利用生成式人工智能的力量，同時確保對敏感企業知識產權的絕對控制。其影響是深遠的，允許以前所未有的可靠性執行複雜的財務對賬和法律審計等高級任務。
Technological Underpinnings: Deep Think, Axion Chips, and Massive Context
技術基礎：深度思考、Axion 芯片和海量上下文
Fueling this integration is Gemini 3’s specialized "Deep Think" mode, which prioritizes parallel processing of logical steps for more accurate, complex answers. This has resulted in benchmark-setting performance, including a record Elo score of 1501 on the LMArena leaderboard. Supporting this are infrastructure upgrades, with Snowflake Gen2 Warehouses now leveraging Google Cloud’s custom Arm-based Axion C4A virtual machines, reportedly boosting inference efficiency by 40% to 212%. Furthermore, Gemini 3's 1-million token context window allows for the ingestion of massive datasets, such as entire quarterly reports, drastically reducing the
Gemini 3 專門的“深度思考”模式推動了這種集成，該模式優先考慮邏輯步驟的並行處理，以獲得更準確、更複雜的答案。這帶來了基准設定性能，包括 LMArena 排行榜上創紀錄的 1501 分 Elo 分數。支持這一點的是基礎設施升級，Snowflake Gen2 Warehouses 現在利用 Google Cloud 的基於 Arm 的定制 Axion C4A 虛擬機，據報導將推理效率提高了 40% 至 212%。此外，Gemini 3 的 100 萬個代幣上下文窗口允許攝取大量數據集，例如整個季度報告，從而大大減少了
