Multi-Cloud Is Almost Always a Mistake (And the Rare Exceptions)
The multi-cloud pitch is appealing: avoid vendor lock-in, negotiate better pricing, survive regional outages. The pitch is also mostly wrong for most companies. Here's why.
Key takeaways
- Multi-cloud doubles operational surface for marginal benefit in most cases.
- "Vendor lock-in" rarely costs as much as multi-cloud complexity.
- The real exceptions: regulated multi-region requirements, acquisition integration, specific service-level needs.
- Pick one cloud, master it.
Why multi-cloud usually backfires
Two of every operational concern: two IAM models, two networking stacks, two billing systems, two on-call runbooks, two security postures, two ways everything works slightly differently. Most engineers know one cloud well; multi-cloud forces them to know both badly.
When multi-cloud is right
Regulated geography
Some regulators require data residency in specific regions where your primary cloud has weak presence. Multi-cloud bridges the gap.
Specific best-of-breed services
You need GCP's BigQuery + AWS's IAM + Cloudflare's edge. Sometimes the technical reasons are real.
Post-acquisition integration
You acquired a company on a different cloud. Migrating is expensive; sometimes leaving them where they are is right.
True high availability
Cross-cloud failover for systems where downtime is catastrophic. Rare.
When it isn't
"To avoid lock-in." Lock-in is real but quantifiable, and the cost of multi-cloud is usually higher than the risk of lock-in.
"For better pricing." Cloud reps will negotiate. Multi-cloud doesn't unlock pricing the way people think.
"Resilience." Region-redundant on one cloud handles 99% of resilience needs at 1/10 the complexity of multi-cloud.
What we recommend
Pick one cloud. Master it. Leave the door open for adding services or regions as specific needs arise. Don't make multi-cloud the architecture.
FAQs
Hybrid (cloud + on-prem)? Different beast, often necessary for legacy or regulatory reasons.
Cloudflare counts as multi-cloud? Mostly no, CDN/edge is complementary, not parallel.
What if our customer demands multi-cloud? Sometimes the price is right. Sometimes it isn't.