**Kubernetes is the Foundation for AI: NVIDIA’s Commitment to an Open, Community-Governed Future**
Kubernetes has become the de facto operating system for artificial intelligence (AI). According to the Cloud Native Computing Foundation’s (CNCF) 2025 Annual Cloud Native Survey, 82% of container users now run Kubernetes in production, and 66% of organizations leveraging generative AI use Kubernetes to manage some or all of their inference workloads. However, there is a significant gap between adoption and operational maturity: only 7% of organizations deploy models daily, while 47% do so only intermittently.
This gap is primarily operational. As AI moves beyond training into inference, fine-tuning, and agentic workloads, it must run reliably and continuously—something that demands distributed systems expertise. Fortunately, the cloud-native community has spent over a decade mastering exactly that. The opportunity now is to make GPU workloads as routine on Kubernetes as any other workload. At NVIDIA, we believe the future of AI will be built in the open, and we are investing deeply in the community driving that vision.
—
### Kubernetes is Already the Foundation for AI
For more than a decade, the cloud-native community has built the infrastructure underpinning the modern internet. Kubernetes and the CNCF ecosystem provide a common way to orchestrate complex, distributed applications across heterogeneous environments at global scale—precisely what AI workloads need. These workloads require dynamic scheduling, observability, networking, and resilience that only mature distributed systems can offer.
What remains a challenge is first-class support for GPUs within that foundation. While Kubernetes excels at scheduling, networking, observing, and securing distributed workloads, GPUs have historically been treated as static, indivisible resources. That worked for early experiments but breaks down at scale when organizations run large-scale training, latency-sensitive inference, and multi-tenant AI platforms simultaneously on shared clusters.
Bridging this gap is an orchestration challenge—and orchestration problems are best solved openly, through shared standards rather than proprietary extensions. NVIDIA is committed to that approach.
—
### What the Community Has Been Telling Us
Platform teams and maintainers consistently highlight the same issue: operating GPUs efficiently at scale requires deeper collaboration because current tools were not designed for it. Traditionally, Kubernetes allocated GPUs as static resources assigned at scheduling time. That model is inefficient for dynamic, large-scale AI workloads that demand flexibility in GPU utilization.
The result is a “tax” on operations: overprovisioned hardware, idle accelerators, and applications that pass tests but fail in production. This tax explains why many organizations limit model deployments to intermittent intervals rather than continuous delivery. Solving this requires community-driven orchestration innovation.
—
### Deepening Our Commitment to CNCF
NVIDIA already contributes significantly to the cloud-native ecosystem through open-source projects like the NVIDIA Container Toolkit and the GPU Operator. Our commitment is now deepening—not only by contributing code but by investing in governance and shared infrastructure.
This year, NVIDIA joined the CNCF Governing Board and committed $4 million over three years to support CNCF project CI and testing on real GPU infrastructure instead of emulators. Our approach remains consistent: contribute upstream first, share governance rather than control, and stand with the projects we help initiate.
Three key contributions illustrate this commitment:
– **NVIDIA GPU Dynamic Resource Allocation (DRA) Driver**: Now upstream in Kubernetes SIG-Node, this reference implementation replaces static GPU assignment with real-time, on-demand allocation. It supports MIG (Multi-Instance GPU) device sharing (in alpha) and ComputeDomains, enabling safe and fast GPU memory sharing across nodes via Multi-Node NVLink. In practice, this allows large multi-GPU jobs to request exactly the resources they need, when they need them.
– **KAI Scheduler**: Accepted as a CNCF Sandbox project at KubeCon + CloudNativeCon Europe, KAI handles complex AI scheduling needs such as gang scheduling with pre-simulation to avoid unnecessary evictions, hierarchical queues with Dominant Resource Fairness for multi-team environments, and asynchronous binding tested on clusters exceeding 10,000 GPUs. By placing it under community governance, NVIDIA ensures the ecosystem can shape and evolve this critical technology.
– **Kubernetes AI Conformance Program**: This program has expanded from 18 to 31 certified platforms, providing consistent verification that AI-ready infrastructure works reliably across providers. New requirements in v1.35 focus on agentic workflow support and in-place pod resizing for inference serving.
Underlying all these efforts is a simple principle: the foundational infrastructure for running AI should be a community asset.
—
### An Open Future, by Choice
The future of AI infrastructure is not yet written. It could consolidate around a few proprietary, closed platforms—or it could follow the path of the internet, driven by open foundations like Linux and Kubernetes that generate broader innovation. NVIDIA is betting on the latter—not out of ideology, but because history shows that open, interoperable, community-governed platforms build larger ecosystems, move faster, and deliver benefits to more people.
As Jensen Huang has said, “We want AI to diffuse into every industry and every country, every researcher, every student. And if everything is proprietary, it’s hard to do research and it’s hard to innovate on top. So open source is fundamentally necessary.”
That conviction guides NVIDIA’s commitment to CNCF. The future of AI infrastructure will not belong to any single company—it will be built collectively. If you work in SIG-Node, on DRA, the AI Conformance Program, or anywhere in AI scheduling and inference, the roadmap is public, and the work is upstream. Come build it with us.
—
### FAQ
**Q: Why is NVIDIA investing in CNCF now?**
A: NVIDIA joined the CNCF Governing Board to deepen its commitment to open, community-driven AI infrastructure. By investing $4 million over three years and contributing upstream projects, we aim to accelerate GPU orchestration standards that benefit the entire ecosystem.
**Q: What problem does the NVIDIA GPU Dynamic Resource Allocation (DRA) Driver solve?**
A: It replaces static GPU assignment with real-time, on-demand allocation, enabling multi-tenant clusters to share GPUs efficiently. Features like MIG device sharing and ComputeDomains allow flexible, safe GPU memory sharing across nodes.
**Q: How does the KAI Scheduler improve AI workload management?**
A: KAI provides gang scheduling with pre-scheduling simulation to reduce evictions, hierarchical queues with fair resource sharing, and asynchronous binding tested at massive scale—over 10,000 GPUs.
**Q: What is the Kubernetes AI Conformance Program?**
A: It is a certification program ensuring AI-ready infrastructure works consistently across providers. It now includes requirements for agentic workflows and in-place pod resizing for inference, promoting reliability and interoperability.
**Q: Why does NVIDIA believe open source is necessary for AI?**
A: Open source enables broad diffusion of AI across industries and geographies, fosters research and innovation on top of existing work, and avoids the limitations of proprietary systems.
—
### Conclusion
The convergence of Kubernetes and AI represents a pivotal shift in how intelligent workloads are orchestrated and scaled. NVIDIA’s active participation in the CNCF ecosystem underscores a clear commitment: building the future of AI infrastructure collaboratively. By open-sourcing critical tools, investing in community governance, and aligning with open standards, NVIDIA is helping ensure that GPU workloads become as routine and reliable as any other cloud-native workload. The path forward is open, community-driven, and built on the proven foundations of Kubernetes—offering scalability, flexibility, and innovation for everyone involved.
