Kubernetes 是管理应用程序如何在数千个 AI 工厂中运行的开源系统，已成为 AI 的基础设施，且对其的需求仍在不断增长。
为了帮助满足这些需求，NVIDIA 正在深化其对开源的承诺，这体现在 KubeCon + CloudNativeCon 上，这是 Kubernetes 和云原生社区的旗舰会议。
NVIDIA 正在向云原生计算基金会 (CNCF) 贡献 NVIDIA OpenShell，这是一个用于保护和治理自主代理的开源运行时。CNCF 是 Linux 基金会旗下的非营利组织，也是 KubeCon 的主办方。作为 CNCF Sandbox 项目，OpenShell 将由将其集成到产品中的组织进行治理，并得到全球安全专家的专业知识的支持。
NVIDIA 还正在贡献 Agent Substrate，这是一个由 Google 主导的开源框架，旨在 Kubernetes 上大规模运行长期运行的代理工作负载，该项目已被纳入 CNCF Sandbox。
此外，NVIDIA 宣布了 NVIDIA DSX AI Factory Platform 的新组件：NVIDIA Kubernetes Engine、NVIDIA Kubernetes Recipes 和 NVIDIA Snapshot。这些工具帮助团队在 Kubernetes 上部署、管理和扩展 AI 工厂的 GPU 基础设施。
总之，这些进步标志着使 Kubernetes 成为大规模 AI 工作负载的主要基础架构迈出了重要一步——赋能开发人员构建更安全、更高效的未来 AI 基础设施。
在开放环境中保护 AI 代理
NVIDIA 将 OpenShell 贡献给 CNCF，将自主代理的安全基础设施交到社区手中，以便构建、检查和改进——加速开发并加强其安全措施。
“云原生社区面临一个新的基础设施挑战：为代理提供安全的操作环境，”Linux 基金会首席执行官 Jim Zemlin 表示。“NVIDIA 将 OpenShell 贡献给 CNCF，为开发人员提供了共同的基础来构建，并有机会塑造代理安全的发展方式。将这项工作引入社区， enables 对基础设施的更广泛协作，这对于下一代云原生应用至关重要。”
OpenShell 使组织能够控制 AI 代理可以执行的操作。它对代理活动应用确定性控制，因此团队可以在代理使用工具、访问推理服务和执行工作负载时强制执行策略。其可扩展的架构允许团队连接自己的计算资源、推理提供商、凭据存储和监控工具。这使得组织能够在不同的部署环境中应用治理，同时将 OpenShell 与现有的企业系统集成。
数百家组织已经使用了 OpenShell，自发布以来不到一年时间，该项目已收到数万份贡献。
在 Kubernetes 上扩展代理执行
治理代理只是挑战的一部分。组织还需要大规模部署它们。
NVIDIA 正在贡献给 Agent Substrate，这是一个源自 Google 的 CNCF Sandbox 项目，并确保其与 NVIDIA Dynamo（用于模型推理）和 KAI Scheduler（用于共享集群资源分配）协同工作。这一努力将代理执行、推理服务和资源调度整合在 Kubernetes 上。
Agent Substrate 在任务之间保留代理环境，并在代理空闲时释放计算资源。通过在共享的代理池上挂起和恢复这些环境，它使团队能够扩展代理工作负载，而无需为每个会话保持专用的计算资源运行。
NVIDIA 和 Google 还在合作集成 OpenShell 和 Agent Substrate，以便企业能够在云端和本地系统上保护和治理代理。
为 AI 工厂构建开放基础
AI 工厂需要为客户提供托管 Kubernetes 服务，同时保留对基础设施、软件和客户体验的控制权。三个新的开源项目扩展了 DSX AI Factory Platform，使 Kubernetes 更易于运行，并使推理扩展更快。
NVIDIA DSX AI Factory Platform 软件堆栈，从 GPU 基础设施和 Kubernetes 管理到工作负载编排和模型服务。
NVIDIA Kubernetes Engine (NVKE) 和 NVIDIA Kubernetes Recipes (NVKR) —— 现已公开预览 —— 为在 GPU 基础设施上提供托管 Kubernetes 服务提供了开源构建块。它们自动化集群生命周期管理并提供签名、版本固定的镜像。
NVKE 和 NVKR 启动时得到了生态系统的支持，扩大了其在 AI 基础设施平台和云服务中的影响力。Mirantis、Netris、OpenNebula Systems、Rafay、Spectro Cloud 和 vCluster，以及 Firmus 和 SK Telecom，已与 NVIDIA 就集成计划达成一致，将 NVKE 和 NVKR 的功能带入他们的产品中。
NVIDIA Snapshot 解决了快速增加推理容量的挑战。加载模型和预热工作负载可能会延迟扩展和恢复。这个开源、模块化组件可以在几秒钟内恢复检查点副本，无需重新加载模型或预热，支持 Kubernetes 上推理和代理工作负载更快的启动、恢复和弹性。
设计合作伙伴 Crusoe 和 Saturn Cloud 已在他们自己的工作负载上使用 NVIDIA HGX B300 系统 评估了 Snapshot。Crusoe 是 Snapshot 的上游贡献者，在 Crusoe Cloud 上测试了 GLM-5.3，将工作负载准备就绪的时间从大约八分钟减少到 31 秒。Saturn Cloud 是一个令牌工厂平台，对 Snapshot 进行了从零扩展推理的基准测试，将 vLLM 启动时间从两到 8.5 分钟减少到六到 34 秒。
通过参加 NVIDIA at KubeCon 了解更多。
查看原文
Kubernetes, the open source system that manages how applications run across thousands of AI factories, has become the foundation for AI, and the demands on it are only growing.
To help meet them, NVIDIA is deepening its commitment to open source at KubeCon + CloudNativeCon, the flagship conference for the Kubernetes and cloud-native communities.
NVIDIA is contributing NVIDIA OpenShell, an open source runtime for securing and governing autonomous agents, to the Cloud Native Computing Foundation (CNCF), a nonprofit that is part of the Linux Foundation and the host of KubeCon. As a CNCF Sandbox project, OpenShell will be governed by the organizations building it into their products and strengthened by the expertise of security professionals worldwide.
NVIDIA is also contributing to Agent Substrate, an open source Google-led framework for operating long-running agent workloads at scale on Kubernetes that has been accepted into the CNCF Sandbox.
Additionally, NVIDIA announced new components of the NVIDIA DSX AI Factory Platform: NVIDIA Kubernetes Engine, NVIDIA Kubernetes Recipes and NVIDIA Snapshot. These tools help teams deploy, manage and scale GPU infrastructure on Kubernetes for AI factories.
Together, these advancements mark a step forward in making Kubernetes the leading foundation for AI workloads at scale — empowering developers to build safer, more efficient AI infrastructure for the future.
Securing AI Agents in the Open
NVIDIA’s contribution of OpenShell to CNCF puts security infrastructure for autonomous agents in the community’s hands to build, inspect and improve — accelerating development and hardening its safeguards.
“The cloud-native community has a new infrastructure challenge: giving agents secure environments in which to operate,” said Jim Zemlin, CEO of the Linux Foundation. “NVIDIA’s contribution of OpenShell to CNCF gives developers a shared foundation to build on and an opportunity to shape how agent security evolves. Bringing this work into the community enables broader collaboration on infrastructure that will be critical to the next generation of cloud-native applications.”
OpenShell gives organizations control over what AI agents can do. It applies deterministic controls to agent activity, so teams can enforce policies as agents use tools, access inference services and execute workloads. Its extensible architecture lets teams connect their own compute, inference providers, credential storage and monitoring tools. This allows organizations to apply governance across different deployment environments while integrating OpenShell with their existing enterprise systems.
Hundreds of organizations have put OpenShell to use, and the project has received tens of thousands of contributions in less than a year since launch.
Scaling Agent Execution on Kubernetes
Governing agents is only part of the challenge. Organizations also need to deploy them at scale.
NVIDIA is contributing to Agent Substrate, a Google-originated project in the CNCF Sandbox, and ensuring it works with NVIDIA Dynamo for model inference and KAI Scheduler for shared cluster resource allocation. This effort brings together agent execution, inference serving and resource scheduling on Kubernetes.
Agent Substrate preserves agent environments between tasks while releasing compute when agents are idle. By suspending and resuming those environments across a shared pool of agents, it lets teams scale agent workloads without keeping dedicated compute running for every session.
NVIDIA and Google are also collaborating on an integration of OpenShell and Agent Substrate, so enterprises can secure and govern agents across clouds and on-premises systems.
Building an Open Foundation for AI Factories
AI factories need to offer customers managed Kubernetes services while retaining control of their infrastructure, software and customer experience. Three new open source projects extend the DSX AI Factory Platform to make Kubernetes easier to run and inference faster to scale.
The NVIDIA DSX AI Factory Platform software stack, from GPU infrastructure and Kubernetes management to workload orchestration and model serving.
NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR) — now available in public preview — provide open source building blocks for delivering managed Kubernetes services on GPU infrastructure. They automate cluster lifecycle management and provide signed, version-pinned images.
NVKE and NVKR are launching with ecosystem support that extends their reach across AI infrastructure platforms and cloud services. Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, together with Firmus and SK Telecom, have aligned with NVIDIA on integration plans to bring NVKE and NVKR capabilities into their offerings.
NVIDIA Snapshot addresses the challenge of bringing additional inference capacity online quickly. Loading models and warming up workloads can delay scaling and recovery. The open source, modular component restores checkpointed replicas in seconds, without model reload or warmup, supporting faster startup, recovery and elasticity for inference and agentic workloads on Kubernetes.
Design partners Crusoe and Saturn Cloud have evaluated Snapshot on their own workloads using NVIDIA HGX B300 systems. Crusoe, an upstream contributor to Snapshot, tested GLM-5.3 on Crusoe Cloud and reduced the time for the workload to become ready from roughly eight minutes to 31 seconds. Saturn Cloud, a token factory platform, benchmarked Snapshot for scale-from-zero inference and reduced vLLM startup times from between two and 8.5 minutes to six and 34 seconds.
Learn more by joining NVIDIA at KubeCon.
原文版权归 NVIDIA Blog 及原作者所有，本站转载仅供阅读。如需删除，请联系 support@aisa.one。
Studio