Multi-cloud cost strategy: when two clouds make sense (and when they don't)
Multi-cloud doubles your discount negotiations but also doubles your egress risk and tooling sprawl. A clear-eyed look at when a second cloud actually saves money.
Every scaleup eventually hears the pitch: go multi-cloud for leverage, resilience and negotiating power. Sometimes that is true. Often it doubles complexity while the promised savings evaporate into egress fees and duplicated tooling.
The honest cost model
A second cloud adds four recurring costs: cross-cloud data transfer (the silent killer), duplicated observability and security tooling, a second set of commitment discounts to manage, and engineering time spent maintaining two deployment paths. If those four exceed the discount delta you negotiated, multi-cloud is a cost, not a saving.
When multi-cloud genuinely wins
Three scenarios hold up. First, genuinely independent workloads: a data platform on one cloud and a transactional app on another, with minimal data flow between them. Second, regulatory or customer requirements that mandate a specific provider or region. Third, best-of-breed services where one provider's offering is decisively better for a specific workload — and the data stays put.
When it doesn't
Splitting one workload across clouds for resilience rarely survives the cost analysis: the cross-cloud traffic and duplicated infrastructure usually exceed the price of a multi-region setup on a single provider. And leverage in negotiations is overrated below roughly seven figures of annual spend — the switching cost is your leverage, not the architecture diagram.
A decision framework
Before adding a second provider, answer three questions. Which workloads move, and how much data flows between them per month? What tooling and skills get duplicated? What is the concrete saving, measured against the four recurring costs? If you cannot quantify the third answer, you do not have a multi-cloud strategy — you have a multi-cloud expense.
The takeaway
One cloud done well beats two clouds done halfway for most scaleups. Choose deliberately, per workload, with data-transfer math in hand.