Hybrid and Multi-Cloud
Hybrid cloud combines an organisation's own infrastructure or private cloud with one or more public clouds, run as a connected whole; multi-cloud means using public cloud services from two or more providers. Financial firms adopt one or both to limit dependence on a single provider, meet data-location rules and keep a credible exit route open.
- Also known as
- Hybrid cloud, Multi-cloud, Multicloud, Hybrid multi-cloud
- Definitions
- NIST (SP 800-145, 2011) describes hybrid cloud as two or more distinct cloud infrastructures bound together by technology that allows data and applications to move between them; ISO/IEC 22123-1 defines multi-cloud as using public cloud services from two or more providers
- Regulation
- DORA requires EU financial firms to weigh concentration risk before contracting a provider that is not easily substitutable, and to hold tested exit strategies for services supporting critical or important functions; firms may set out a multi-vendor strategy explaining their mix of providers
- Switching
- under the EU Data Act, cloud providers may charge only reduced, cost-based switching fees until 12 January 2027 and none after it; contracts must let customers start a switch on no more than two months' notice, with a transition of up to 30 days
- Typical patterns
- core ledger on premises or in a private cloud with digital channels in public cloud; a warm standby for critical services with a second provider; different providers for different workloads such as AI training, analytics and payments
- Trade-off
- resilience and bargaining power against duplicated engineering, security tooling and skills, and the loss of each provider's proprietary services
How hybrid and multi-cloud works in practice
Few financial firms run one application live across two providers; the cost is high and the benefit narrower than it looks. More common is a division of work — the core ledger and card processing on premises or in a private cloud the firm controls, customer-facing apps and analytics in a public cloud, AI workloads wherever the accelerators are — plus a standby arrangement for critical services in a second environment. The design question is which services must survive the loss of a provider, how fast, and at what cost.
Every step towards portability costs something: containers and infrastructure-as-code instead of the provider's managed services, a second set of security controls, engineers trained on two platforms, and data egress fees. Every step away from it deepens lock-in, which supervisors now treat as a risk to be managed rather than a commercial choice. When a major AWS outage on 20 October 2025 took down platforms including Coinbase, customers with no second environment could only wait.
For a buyer, cloud architecture is a line in the technology diligence and sometimes in the price. A target built entirely on one provider's proprietary services may need expensive re-platforming to meet the acquirer's resilience standards; a hybrid estate with its own data centres brings capital expenditure and staff that a cloud-native acquirer will want to retire.
Example
A digital lender with €2bn of loans runs everything on one public cloud at €3.1m a year. Its supervisor flags concentration risk: there is no tested exit, and the loan-servicing platform depends on the provider's proprietary database. Full active–active multi-cloud is costed at €2.4m a year extra plus €4m of re-engineering. The lender chooses a narrower plan: move the servicing database to an open-source engine on containers for €1.3m, keep a warm standby with a second provider that holds replicated data and can take over servicing within four hours for €700k a year, and leave marketing and analytics on one cloud. An exit test in month nine moves servicing to the standby in 3 hours 20 minutes. The cloud bill rises by 23%, and the finding is closed.
Common mistakes
- Calling two providers for unrelated workloads resilience, when neither could take over from the other. - Building on proprietary managed services and discovering the exit cost only when a supervisor asks for the plan. - Replicating applications but not data, identity or DNS, so the standby cannot run. - Leaving egress fees and duplicated staff out of the business case. - Treating on-premises as automatically safer; it concentrates the risk in the firm's own operations.