Slide 1 of 8

llm-d Community

15 Organizations. One Mission.

From competitors to collaborators: the story of how cloud providers, hardware vendors, and researchers built a vendor-neutral platform for Kubernetes-native LLM serving, together.

15+

Organizations

100+

Contributors

8

Integrations

CNCF

Sandbox Project

Slide 2 of 8

The Partner Network

No single company should own the routing plane for LLM serving. Red Hat, Google Cloud, and IBM Research co-founded llm-d with this conviction.

How 15 Organizations Unite

3 founding partners set governance and the routing/EPP framework
NVIDIA and CoreWeave contributed GPU optimization and scale testing
6 ecosystem partners bring hardware, models, and production workloads
3 academic/governance institutions anchor research and IP

Slide 3 of 8

From Launch to CNCF Sandbox

13 months from first commit to CNCF acceptance. Every milestone built on the last.

13

Months to CNCF

6

Conference Demos

4

Major Releases

Slide 4 of 8

Cloud-Native AI Landscape

llm-d connects to the broader cloud-native AI ecosystem through 8+ integrations.

vLLM -- High-throughput inference with PagedAttention (v0.6+)
KServe -- Model serving on Kubernetes (v0.13+)
Gateway API -- K8s-native routing with Inference Extension (v1.2+)
NIXL -- NVIDIA GPU-to-GPU KV cache transfer (v0.3+)
Kubernetes -- Container orchestration with CRDs (v1.28+)
Prometheus -- Observability and autoscaling metrics (v2.45+)
KEDA -- Event-driven autoscaling by queue depth (v2.12+)
Envoy -- Data plane for Gateway API traffic (v1.28+)

Slide 5 of 8

Vendor-Neutral by Design

CNCF governance ensures no single company controls the roadmap. Contributions flow up, governance flows down.

CNCF

Cloud Native Computing Foundation, vendor-neutral home

Technical Oversight Committee

Evaluates projects, sets technical direction

CNCF Sandbox

Accepted March 2026, IP protection and CI infrastructure

Maintainers

Red Hat, Google Cloud, IBM Research, merge authority

Contributors

15+ organizations, PRs evaluated on technical merit

Slide 6 of 8

One PR. Infinite Ripple.

See how a single contribution ripples through the entire ecosystem to serve millions.

A single code fix is reviewed by 3 maintainers, merged into a release used by 42 organizations, deployed across 180 GPU clusters, routing 2.4M requests per day, and ultimately serving 12M end users worldwide.

Score 1,284

New Feature Impact

Score 1,580

Perf Optimization Impact

Slide 7 of 8

From First Issue to CNCF Contributor

Every maintainer started where you are now. Here is the path.

A Contributor's Story

"My first PR was a two-line fix to a config parser. The review was thorough but kind. They helped me understand the project structure along the way."

Becoming a Maintainer

"It was not about writing the most code. It was about consistently helping others succeed and earning trust through reliable, thoughtful contributions."

Slide 8 of 8

Your GPU Clusters Need Better Routing.
We Need Your Expertise.

We started with 5 organizations. We are now 15+. The next contributor could be you, and we mean it. Every PR gets reviewed. Every voice gets heard.

📝

File Your First Issue

Found a bug? Have a feature idea? Issues are where every contribution starts. No idea is too small.

📞

Join the Weekly Community Call

Thursdays, 10am ET. Meet the maintainers, hear what is in progress, and shape the roadmap.

🎯

Pick Up a Good First Issue

We label beginner-friendly tasks specifically for new contributors. Real work, not busywork.

💬

Share Your Production War Stories

Running LLM inference at scale? Your hard-won lessons help shape what we build next.

View on GitHub Community Call CNCF Listing Website