The first generation of cloud programs was often justified on a blend of agility, resilience, security and cost.
In practice, cost savings were frequently harder to capture than expected because migration created consumption transparency faster than it created consumption discipline.
Many enterprises are now "mostly migrated" but not necessarily cloud effective.
Their portfolios include lifted-and-shifted workloads, duplicated tooling, unclear product ownership, manual tagging and chargeback rules, and data architecture that was never designed for AI-native use cases.
Cloud created flexibility but also made it easier for teams to scale spend faster than governance could respond.
Implications for AI-Readiness:
AI raises the stakes of poor workload placement. Training and inference workloads may have different economics.
Data-intensive workloads may be constrained by egress cost, latency or sovereignty rules. Agentic workflows may require secure access to multiple systems and real-time observability.
A cloud estate optimised for hosting applications may not be optimised for orchestrating autonomous digital work.
The costing model changes from service procurement to transaction or tokenization usage monitoring, leading to possible increased budget/costing models.