IT owns the bill, business owns the usage — a split that breaks down for token spend specifically. Chargeback and showback models adapted from cloud FinOps, and what accountability actually needs to look like.
Model right-sizing, prompt and context engineering, caching, and retrieval design are cost levers as much as performance levers. Optimization has to be owned at design time, not applied after a bill shocks someone.
Most enterprises cannot answer what they spent in tokens this week, on what, and whether it was worth it. Real-time token metering is an architecture decision made at design time, not a reporting problem solved after the fact.
Token cost is becoming the defining AI cost-control problem for enterprises, the way cloud spend was a decade ago. Most organisations have neither the real-time visibility nor the operating model to manage it. This is the conversation that needs to start now.
AI is a powerful execution engine. But the expertise required to know whether its output is right, where it will fail, and what to do about it remains irreducibly human. Domain knowledge is not a legacy requirement in the age of AI. It is the foundation that makes AI work.
AI is becoming the execution layer for an expanding range of work. But the architectural function, defining what should be built and why, remains irreducibly human. The enterprises that understand this distinction will deploy AI far better than those still debating job displacement.
DORA compliance is routinely justified on regulatory grounds — the cost of non-compliance exceeds the cost of the programme. This is a sufficient justification but not the strongest one. The operational resilience capability that DORA requires has business value well beyond avoiding fines.
AI transformation programmes are failing to deliver business value across European enterprises at rates that should concern every technology leader planning or executing one. The failure patterns are consistent and preventable. These are the ones I see most often.
Enterprise AI deployments fail to scale for a consistent set of architectural reasons that could have been identified and addressed before the scaling investment was committed. This is the readiness checklist that surfaces those gaps.
KubeCon EU 2026 produced clear signals about the convergence of cloud-native infrastructure and AI workload requirements. Three themes have specific architectural implications for enterprise teams planning AI infrastructure investment.
