arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.10900cs.NI
收藏

AI 基础设施主权

AI Infrastructure Sovereignty

Ciena(思科系统公司)

Sergio Cruzes

AI总结:

本文探讨了AI基础设施主权的概念,强调在能源、可持续性和网络覆盖等约束下,地区、运营商或国家维持AI系统运营控制的能力,通过数据中心、光传输网络和控制框架的联合设计实现主权。

摘要:

人工智能已从以软件为中心的学科转变为以基础设施驱动的系统。大规模训练和推理依赖于紧密连接的数据中心、高容量光网络和接近物理和环境极限的能源系统。在此背景下,对数据和算法的控制已不够。真正的AI主权取决于在能源可用性、可持续性要求和网络覆盖等约束下部署、运行和适应基础设施的能力。本文介绍了AI基础设施主权的概念,定义为地区、运营商或国家在这些约束下维持AI系统运营控制的能力。核心思想是主权源于三个层次的联合设计:面向AI的数据中心、光传输网络和提供实时可见性和协调的控制框架。本文首先探讨了AI工作负载如何重塑数据中心设计,推动功率密度更高,增加冷却需求,并加强与本地能源系统的关系。在此设定下,碳强度和用水量等因素成为AI部署位置和方式的硬性限制。随后,本文分析了光网络作为分布式AI的骨干,展示延迟、容量、故障域和司法管辖区边界如何直接影响实际成果。在此基础上,本文强调了遥测、代理AI和数字孪生作为实现运营主权的关键使能者。它们共同使在计算、网络和能源领域实现闭环监控、协调和验证系统行为成为可能。

英文摘要:

Artificial intelligence has shifted from a software-centric discipline to an infrastructure-driven system. Training and inference at scale now depend on tightly connected data centers, high-capacity optical networks, and energy systems operating close to their physical and environmental limits. In this context, control over data and algorithms is not enough. Real AI sovereignty depends on the ability to deploy, operate, and adapt infrastructure under constraints such as energy availability, sustainability requirements, and network reach. This tutorial-survey introduces the concept of AI infrastructure sovereignty, defined as the ability of a region, operator, or nation to maintain operational control over AI systems within these constraints. The central idea is that sovereignty emerges from the joint design of three layers: AI-oriented data centers, optical transport networks, and control frameworks that provide real-time visibility and coordination across them. We first examine how AI workloads are reshaping data center design, pushing power densities higher, increasing cooling demands, and tightening the relationship with local energy systems. In this setting, factors such as carbon intensity and water usage become hard limits on where and how AI can be deployed. We then look at optical networks as the backbone of distributed AI, showing how latency, capacity, failure domains, and jurisdictional boundaries directly influence what can be achieved in practice. Building on this foundation, the paper highlights the role of telemetry, agentic AI, and digital twins as key enablers of operational sovereignty. Together, they make it possible to monitor, coordinate, and validate system behavior across compute, network, and energy domains in a closed loop.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑