Accepted at PlatformCon 2026
I am excited to share that my session “Lessons Learned Orchestrating Multi-Tenant GPUs on OpenShift AI with NVIDIA KAI (G/H200)” has been officially accepted for PlatformCon 2026 — the largest platform engineering conference in the world.
🎟️ Register for free at platformcon.com
PlatformCon is a virtual event that brings together thousands of platform engineers, DevOps practitioners, and infrastructure leaders from across the globe. Being selected to present alongside the best in the industry is a huge honor.
The talk
This session covers the real-world lessons from building a safety-first, multi-tenant GPU platform using NVIDIA hardware and Red Hat OpenShift AI. If you have been following my work, you will recognize this as an evolution of the talk I delivered at KubeCon EU 2026 in Amsterdam — now refined with additional production insights and deeper coverage of the NVIDIA KAI (Kubernetes AI) stack.
What I will cover
GPU sharing is the hard problem. Everyone talks about running AI workloads on Kubernetes. Few talk about what happens when multiple teams need to share the same GPU infrastructure safely, fairly, and without burning through your budget.
Here is what the session addresses:
- Multi-tenant GPU isolation patterns — MIG, MPS, and time-slicing on NVIDIA G/H200 hardware
- NVIDIA KAI on OpenShift AI — how the Kubernetes AI stack manages GPU lifecycle, scheduling, and workload placement
- Safety-first architecture — quotas, policies, and guardrails that prevent one team from starving another
- Platform engineering patterns — self-service GPU provisioning that data scientists actually want to use
- Lessons learned the hard way — the failures, surprises, and production incidents that shaped our approach
- Cost optimization — chargeback models and utilization strategies for expensive GPU infrastructure
Who this is for
- Platform engineers building internal developer platforms with GPU support
- MLOps teams running shared AI/ML infrastructure
- Infrastructure architects planning GPU investments for enterprise AI
- Engineering managers trying to balance GPU demand across multiple teams
Why PlatformCon
PlatformCon has become the definitive event for platform engineering. What started as a community initiative has grown into a conference that attracts tens of thousands of attendees and features talks from engineers at the largest technology companies in the world.
The virtual format means anyone can attend for free, regardless of location. No travel budget required, no visa applications, no jet lag. Just deep technical content from practitioners who build platforms every day.
PlatformCon 2026 key details
| Detail | Information |
|---|---|
| Event | PlatformCon 2026 |
| Format | Virtual (free to attend) |
| My Session | Lessons Learned Orchestrating Multi-Tenant GPUs on OpenShift AI with NVIDIA KAI (G/H200) |
| Speaker | Luca Berton |
| Website | platformcon.com |
A growing conference journey
PlatformCon joins an exciting 2026 speaking schedule:
- KubeCon EU 2026 (Amsterdam) — Multi-Tenant GPUs on Bare Metal ✅ Delivered
- Cloud Native Rejekts EU 2026 (Amsterdam) — Master of Ceremonies ✅ Delivered
- Red Hat Summit 2026 (Atlanta) — GPUs Take Flight: Safety-First Platform Engineering 🔜 May 12
- PlatformCon 2026 (Virtual) — Multi-Tenant GPUs on OpenShift AI with NVIDIA KAI 🆕 Accepted
Each conference reaches a different audience with a different angle on the same core expertise: making GPU infrastructure work safely and efficiently at enterprise scale.
Stay updated
I will share the recording link once the session is published. In the meantime:
- 📺 YouTube — Luca Berton
- 💼 LinkedIn — Luca Berton
- 📅 Book a consultation if you are building a multi-tenant GPU platform
