Skip to content

Independent marketplace for verified cloud accounts

Multi-Cloud: When It Helps and When It Adds Cost

Multi-cloud is usually an accident described later as a strategy. Which reasons for running two providers hold up, which do not, and the costs nobody itemises.

Multi-cloud is one of those terms that sounds like a strategy and usually describes an accident. Two teams picked different providers, a company was acquired, a workload was moved for one reason and the old environment never shut down. Three years later the architecture diagram has two logos on it and someone writes a slide calling it a deliberate posture.

That is not an argument against running on more than one provider. There are solid reasons to do it. It is an argument for knowing which reason applies to you, because the ones that justify the overhead are narrower than the pitch suggests, and the overhead is larger than the diagram implies.

The term covers four different architectures

Before deciding anything, establish which of these you are actually discussing. They have almost nothing in common operationally.

  • Split estate. Different applications live on different providers, each fully resident on one. No single workload spans providers.
  • Split tier. One application has components on two providers, for example compute in one place and object storage in another. Every request crosses the boundary.
  • Active failover. A full copy of the workload sits ready on a second provider, kept current, waiting to take traffic.
  • Portable by design. One deployment, built on abstractions that could run anywhere, currently running in one place.

Cost and complexity differ enormously across these. A split estate is close to free in architectural terms and mostly costs you in tooling and skills. Split tier is the expensive one, because every call pays network latency and transfer charges. Active failover doubles infrastructure and more than doubles operational work. Portable by design costs you capability rather than money, which is a subtler bill.

Reasons that hold up under scrutiny

Some of these come up repeatedly in real estates and survive the cost conversation.

Data residency obligations. If a regulator requires certain records to stay in a jurisdiction and your primary provider has no suitable region there, you need a second provider. This is not a preference, it is a constraint, and it typically produces a clean split estate with a narrow scope. The detail is in European cloud providers and data sovereignty.

A capability that genuinely exists in one place only. A managed service with no real equivalent, or accelerator capacity you cannot get elsewhere. Worth it when the capability is central to the product rather than convenient.

Latency to a specific population. If your users are concentrated in a market where your main provider has thin presence, a regional provider closer to them can be straightforwardly better. This usually shows up as an edge or regional tier rather than a full second estate.

A large cost delta on a bulk workload. Steady, predictable, bandwidth-heavy or storage-heavy work can be dramatically cheaper on a value-focused provider, while the control plane and the complicated parts stay where they are. This is one of the few cases where the split pays for itself quickly.

Concentration risk that a board actually tracks. Some organisations are required to demonstrate they are not wholly dependent on one supplier. If that requirement is real and audited, it is a reason. If it is a slide, it is not.

Reasons that usually do not survive the maths

Provider outage insurance. This is the most cited justification and the weakest in practice. Full provider failure is rare. The failures that actually take people offline are regional, or they are self-inflicted: a bad deploy, a certificate expiry, a misapplied firewall rule, a deleted record. A second provider protects against none of those, and the extra moving parts increase the odds of the failures you do get. Multi-region on one provider addresses the realistic scenarios at a fraction of the cost, as covered in how cloud regions and availability zones work.

Bargaining power. The theory is that a credible threat to leave lowers your price. The practice is that bargaining power comes from contract timing and committed spend, and that splitting your spend across two vendors reduces the volume discount you qualify for with each. Some organisations do extract value here, but they are large and they have procurement teams who do this full time.

Avoiding lock-in. A real concern, but multi-cloud is an expensive way to address it and often does not. Running on two providers using each one’s managed services means you are now locked into two. The cheaper approach is covered in avoiding vendor lock-in without giving up managed services.

The costs nobody puts on the slide

Infrastructure spend is the visible part and usually the smaller part. The rest shows up in places that are harder to attribute.

Cost area What it looks like in practice
Cross-provider traffic Data leaving one provider is billed, and split-tier designs generate far more of it than anyone modelled
Identity and access Two permission models, two audit trails, two sets of credentials to rotate and review
Observability Metrics and logs in two formats, usually requiring a neutral tool on top to get one view
Expertise Your team needs working depth in two platforms, including the on-call rota at three in the morning
Discount eligibility Split commitments qualify for lower tiers than consolidated ones
Incident response Cross-provider issues have no single support owner, and triage takes longer

The expertise line is the one that quietly dominates. Infrastructure costs scale with usage; human cost scales with the number of distinct systems people must hold in their heads, and it does not go down when traffic does.

Where to put the seam

If you are going ahead, the single most consequential decision is where the boundary falls. Good seams are places where traffic is already low, asynchronous and tolerant of delay. Bad seams cut through a synchronous request path.

Workable boundaries include backups and archive storage on a second provider, batch processing that reads once and writes once, static asset delivery, and an entire self-contained application. Boundaries that cause pain include a database on one provider with its application on another, a cache separated from the service that reads it, and anything where a user request crosses the seam more than once.

Test the seam under load before you commit. Latency across providers is not just higher, it is more variable, and tail latency is what users feel.

If you do it, run both properly

Half-maintained second environments are worse than not having them. The failover region that has not been patched in eight months will not save you; it will fail during the one incident where you needed it and consume hours proving it.

Hold both environments to the same standard: same deployment pipeline, same monitoring coverage, same alert routing, same patch cadence, same access review. If a second provider cannot meet that bar given your team size, that is a complete answer on its own. Automated, repeatable provisioning is close to mandatory, and infrastructure as code and the provider model explains how to structure definitions so the second target does not become a hand-maintained copy.

Keep the accounts separate and clearly owned. Mixing providers inside one shared login arrangement makes attribution and offboarding harder, and the reasoning in why account isolation beats one shared account applies with extra force when two platforms are involved.

A test before you commit

Ask four questions and answer them in writing. What specific outcome does the second provider deliver that one provider cannot? What is the all-in annual cost including staff time and cross-provider traffic? Who owns the second environment by name, and is that person’s capacity actually free? If this were a new company starting today with no history, would anyone propose this architecture?

If the first answer is vague or the fourth answer is no, you probably have an accident rather than a strategy, and the right move is consolidation rather than formalisation.

Starting small when the answer is yes

When the case holds, keep the first step narrow. One workload, one clear boundary, a defined period to evaluate, and a decision point at the end where continuing is an explicit choice rather than a default.

Pairing a hyperscaler with a value-focused provider for bulk work is the most common pattern that pays off, and our AWS and Hetzner comparison shows the shape of that trade-off clearly. On TrustCloudHub you can review verified listings for a Buy Hetzner Account, a Buy OVH Account or a Buy Vultr Account if you want a second environment running without a procurement cycle. We are an independent marketplace rather than a provider partner, so confirm region availability, account limits and any credit position inside the provider’s console after handover. Promotional credits are controlled by the provider, expire on its schedule and exclude some services, so never build a cost case on a balance you have not seen there yourself.

Related comparisons

Hetzner vs Digital Ocean
vs

Hetzner vs Digital Ocean

Hetzner sells hardware value from a mostly European footprint. DigitalOcean sells a developer platform with managed add-ons and deep documentation.…

Hetzner vs OVH
vs

Hetzner vs OVH

Two large European providers that own their data centres but sell very different things. This compares product breadth, bare metal…

Azure vs Digital Ocean
vs

Azure vs Digital Ocean

DigitalOcean optimises for a developer getting something running today. Azure optimises for an organisation running a thousand things for a…

Azure vs Civo
vs

Azure vs Civo

Civo treats managed Kubernetes as the whole product. Azure treats it as one service among hundreds. That difference in focus…

Keep reading

Handing a Cloud Account Over to a Client Cleanly
Cloud Accounts 7 min read

Handing a Cloud Account Over to a Client Cleanly

Transferring an account you built to the client who paid for it. Covers scoping the handover, separating your tooling, switching billing, and…

Monitoring and Alerting That Engineers Do Not Ignore
DevOps and Deployment 7 min read

Monitoring and Alerting That Engineers Do Not Ignore

Most monitoring systems fail not because they miss problems but because they report too many. This guide is about signal quality and…

Need the account, not just the theory?

Browse documented configurations across every provider in the catalog, or ask us which one fits your workload.

Telegram