Operations | Monitoring | ITSM | DevOps | Cloud

GPU utilization is one of the most expensive metrics in cloud infrastructure to get wrong. A GPU running at 30% utilization costs the same as one running at 90%, but it's doing a third of the useful work. For workloads measured in tens of thousands of GPU-hours, the difference between average utilization in the 30s and average utilization in the 70s is hundreds of thousands of dollars across the life of the workload.

Your AI calls already emit OpenTelemetry: your LLM gateway exports it, and it’s the open standard your own services can speak. But you don’t have anywhere to turn those spans into spend you can allocate to an outcome. Now you can. CloudZero exposes an OpenTelemetry endpoint that doesn’t care what’s on the other end.

Operational visibility is becoming increasingly important as infrastructure teams are asked to support AI initiatives, automation goals, cost accountability, modernization efforts, and growing operational complexity at the same time. Most are expected to do it without expanding headcount, introducing additional risk, or rebuilding the environment from scratch. Those expectations are changing the role of infrastructure operations.

Mission-critical networks are changing fast. Utilities, transport operators, and critical infrastructure providers are under pressure to deliver more data, more automation, and more resilience—without ever compromising reliability. The challenge is simple: legacy SDH/SONET networks were built for a different era. They still deliver reliability. But they can’t support what comes next.

Three days, 20 talks at Devoxx France 2026. The through-line wasn't AI hype - it was discipline. Context engineering, code review under AI volume, and the local-vs-remote question now shaping security, cost, and sovereignty. Fabien is a senior software engineer at Qovery. He writes about platform engineering, AI tooling, context engineering, and the practical realities of running modern developer infrastructure.