Mistral CEO Arthur Mensch 讨论公司 40 亿欧元投资欧洲数据中心、为企客户拥有完整 AI 堆栈的重要性，以及如何构建有韧性的主权 AI 基础设施。
Mistral CEO Arthur Mensch discusses the company's €4 billion investment in European data centers, the importance of owning the full AI stack for enterprise customers, and how they are building resilient, sovereign AI infrastructure.
要点 · TL;DR
欧洲有两年窗口期建设独立 AI 基础设施，否则将面临美国主导风险。 Europe has a two-year window to build independent AI infrastructure or risk US dominance.
企业采用 AI 需至少 2 倍回报，否则支出将崩溃。 Enterprise AI adoption needs 2x ROI to avoid spending collapse.
AGI 是方向而非终点，实际整合才是关键。 AGI is a direction, not a destination; practical integration is key.
核心观点 · Key points
欧洲有两年窗口期来建设独立 AI 基础设施，否则将面临被美国科技巨头控制的风险。 Europe has a two-year window to build independent AI infrastructure or risk losing control to US tech giants.
企业采用 AI 需要强劲的投资回报率；支出必须产生至少两倍回报，否则整个体系可能崩溃。 Enterprise AI adoption requires strong ROI; spending must yield at least 2x return to avoid collapse.
智能体平台在采购等关键流程中需要人类参与验证。 Agentic platforms need human-in-the-loop validation for critical processes like procurement.
开源模型是实现定制化、成本优化和避免供应商锁定的关键。 Open-source models are key to customization, cost optimization, and avoiding vendor lock-in.
AGI 是进步的方向而非最终目标；真正的挑战在于实际整合。 AGI is a direction of progress, not a final destination; real challenge is practical integration.
反共识 · Contrarian takes
尽管当前需求超过供给，美国超大规模云服务商可能在 AI 基础设施上过度支出。 US hyperscalers may be overspending on AI infrastructure despite current demand exceeding supply.
欧洲强大的电网和能源可用性是 AI 数据中心的战略资产。 Europe's strong grid and energy availability are strategic assets for AI data centers.
AI 主权不仅是欧洲的问题；每个国家都需要 AI 战略自主。 AI sovereignty is not just European; every state needs strategic autonomy in AI.
企业采用 AI 的瓶颈在于组织变革，而非模型智能。 Enterprise AI adoption is bottlenecked by organizational change, not model intelligence.
AI 网络安全需要开源模型和动态访问控制，而不仅仅是更好的模型。 Cybersecurity in AI requires open-source models and dynamic access control, not just better models.
AGI 是一个混乱而具体的方向，需要理解物理和人类行为。 AGI is a messy, concrete direction requiring understanding of physics and human behavior.
本期章节 · Chapters（共 12）
引言与计算策略Introduction and Compute Strategy
GPU 即服务与 Token 生成GPU as a Service and Token Generation
欧美 AI 基础设施支出对比Europe vs US AI Infrastructure Spending
欧洲 AI 基础设施窗口期Europe's AI infrastructure window
定制芯片与基础设施Custom Chips and Infrastructure
智能体平台与氛围Agentic Platform and Vibe
自主性与基础设施Autonomy and Infrastructure
企业影响与组织变革Enterprise Impact and Organizational Change