OpenStack

I am happy to share that I gave my first #KubeCon talk, together with my colleague Bertrand Souville, at KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026. We presented how AI agents can turn complex telecom requirements and specifications into reviewable engineering outcomes, helping identify bugs, gaps, and inconsistencies earlier while improving engineering quality and reducing manual effort. We shared our experience through real-world use cases with Kubernetes Operators and OpenStack Tacker. #KubeCon #CloudNativeCon #Telco #CloudNativeTelco #AIAgents

From 13% to 59% GPU utilization: Day 2 #OpenInfraSummit Asia keynotes showed how open source transforms AI infrastructure in production. Thierry Carrez discusses that if you do not run your own infrastructure, you pay for someone else's operating margins. As hyperscaler capex continues to rise, relying entirely on public clouds leaves organizations exposed to rising costs and limited control. Running production AI workloads on open infrastructure gives teams direct cost control, vendor independence, and regulatory compliance. During the Day 2 keynotes, speakers detailed how open source tools solve real-world scale and security challenges: NVIDIA's 🛡️ Zvonko Kaiser covered the path from container isolation to Confidential Containers, introducing bidirectional confidentiality. Full-stack attestation and hardware security layers allow teams to protect both input data and proprietary models in serverless GPU environments. Viettel Group shared how Vietnam's largest telecommunications group built a multi-layer AI platform on OpenStack Nova, Ironic, and Kata Containers. By moving from physical servers to tokenized access, they eliminated hardware silos, boosted GPU utilization from 13% to 59%, and delivered 3.5 times the work per card. Then fupan li from the Kata Containers community highlighted how Kata 4.0 provides hardware-level VM isolation per workload to protect infrastructure when AI agents execute untrusted code or access external APIs. Projects like OpenStack, Kata Containers, StarlingX, and Zuul provide a production-proven open source foundation for the modern AI stack.

"If you don't run your own infrastructure, you pay for someone else's margins," Thierry Carrez discusses the infrastructure that enables frontier intelligence, live from the #OpenInfraSummit Asia! Frontier intelligence requires more than open models and frameworks. It demands an open foundation built on compute, networking, and storage. Hyperscaler capex investments continue to rise, but relying entirely on public clouds means paying someone else's operating margins. Running production AI workloads on open infrastructure gives organizations direct cost control, vendor independence, and regulatory compliance. Projects like OpenStack, Kata Containers, StarlingX, and Zuul provide a production-proven, open source foundation for the complete AI software stack. #WeAreOpenInfra

Welcome to Day 1 of the #OpenInfraSummit! The OpenInfra team is excited to be joining our friends from the PyTorch and Cloud Native communities to kick off the day with Keynotes!

The OpenInfra community couldn't feel more welcomed to Shanghai! "Shanghai is Horace's hometown, so when he says welcome, he means it." Horace Li, Director of OpenInfra/CNCF/PyTorch Asia, and Jonathan Bryce, Executive Director of Cloud and Infrastructure, welcome the community to Shanghai and receive a warm welcome from the founder of open source in China, Professor Lu Shouqun!

It's not about OpenStack; it's not about Kubernetes; it's not about PyTorch. It's about all open source for the AI era we live in now. Jonathan Bryce and Horace Li open the keynote stage by introducing the complete open source AI stack, connecting models and inference down to physical hardware.