WeAreDevelopers World Congress 2026: Why Infrastructure Forecasting Fails
I'll be speaking at WeAreDevelopers World Congress North America on September 24 about why infrastructure forecasting fails and how to build a self-serve forecasting platform.
I'll be speaking at the WeAreDevelopers World Congress 2026 – North America on Thursday 24 September, 15:30–16:00, Stage 4, on the DevOps & Platform Engineering track. The session is “Why Infrastructure Forecasting Fails – Building a Self-Serve Forecasting Platform.”
Infrastructure forecasting fails in a predictable way. A central team owns the model, the model needs inputs only the service teams have, the service teams have no incentive to keep those inputs current, and by the time the forecast is wrong enough to notice, the hardware lead time has already closed. Adding more accuracy to the model does not fix any of that.
This session is about the alternative: making forecasting self-serve, so the teams who own the demand also own the forecast, and the central platform's job becomes providing the model, the data, and the guardrails rather than the answer.
What I'll cover
- Why centralised forecasting breaks down as the number of services grows
- Separating the forecast from the capacity decision, and why conflating them is expensive
- What a self-serve forecasting platform needs to expose – and what it must not let teams change
- Handling GPU and accelerator demand, where lead times and unit costs punish being wrong
