A GPU cluster can remain operational long after the engineers who configured it have moved on, but that does not guarantee the workload running inside it will remain supportable. This distinction deserves greater attention as neocloud providers compete to supply specialised computing capacity for increasingly demanding artificial intelligence applications. Customers typically evaluate accelerator availability, networking performance, storage throughput and commercial terms before committing production workloads to a provider. Yet the people who understand how those components interact may hold operational knowledge that never reaches a formal support process. An infrastructure environment can satisfy its original performance requirements while carrying undocumented dependencies that complicate future maintenance. Those dependencies become particularly relevant when the engineers responsible for initial deployment leave the organisation or move into different roles. For customers relying on predictable inference performance or lengthy training schedules, the concern extends beyond personnel changes. The real question is whether the provider has developed an operational capability that survives the departure of its original technical specialists.