A startup must decide between building infrastructure internally and adopting managed services. What should the evaluation actually weigh?

The factors that actually decide it

• Total cost of ownership, including the engineers. The comparison is never licence versus zero; it is licence versus the fully loaded cost of the people who will operate it, plus the opportunity cost of what they would otherwise have built. For a small team, one engineer's time exceeds most managed-service bills.
• Whether this is differentiating. Operating a database well is difficult and valuable and it is not what the customers pay for. Infrastructure that is not a competitive advantage should be bought unless the numbers are extreme.
• Failure modes and who handles them at 3am. A managed service's outage is out of your control and usually shorter than yours would have been. A self-hosted outage is in your control and depends on whether the person who understands it is reachable. Control is not the same as capability.
• Exit cost. How much of the design becomes specific to this provider, and what a migration would cost. This is where the honest answer is usually "more than we think", and it argues for keeping provider-specific features at the edges.
• Scale economics. Managed services are cheap at small scale and expensive at large scale, and the crossover is real. A workload that will be very large in two years may justify building — but only if the team will still exist to operate it.

The asymmetry that decides most cases

Buying is reversible more cheaply than building. A team that buys and later needs to self-host has a working system to migrate and knows exactly what it needs. A team that builds and later wants to buy has an integration project plus a decommissioning project plus a team whose skills were invested in the wrong place.

The judgement for a growing startup

Buy nearly everything, and revisit specific components when a measurable cost or capability threshold is crossed. The correct trigger is a number — this managed service now costs more than two engineers, or its latency is now a product problem — not a general sense that operating it ourselves would be better.

The counter-case is real: if your product is infrastructure, then operating infrastructure is your competitive competence and buying it means buying your competitor's constraints.