{
  "id": "multicloud",
  "title": "Multicloud",
  "lang": "en",
  "source_lang": "en",
  "summary": "Multicloud is the use of cloud services from multiple providers, whether for separate workloads or as parts of an integrated computing environment.",
  "category": "computer-science",
  "subcategory": "software",
  "type": "concept",
  "aliases": [
    "Multi-cloud",
    "multicloud computing"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/multicloud/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/multicloud/",
      "json": "https://aiwiki.page/zh/multicloud.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Multicloud is the use of cloud computing services from more than one cloud provider. It may describe an organization that assigns different applications to different providers, or an architecture in which a single application uses services across providers. Multicloud does not inherently imply integration, portability, or automatic failover: these are separate capabilities that require deliberate design."
    },
    {
      "heading": "Scope and related concepts",
      "text": "In common technical usage, multicloud refers especially to the use of two or more public cloud providers. Broader definitions also include combinations of public and private clouds. The services involved can span infrastructure as a service, platform as a service, and software as a service. Some organizations use the term primarily for their infrastructure and application platforms, excluding routine use of multiple SaaS products.\n\nMulticloud differs from several related arrangements:\n\n- Hybrid cloud combines distinct cloud environments, commonly a private cloud and a public cloud, with mechanisms for integration or movement of data and applications. Hybrid cloud describes the relationship between deployment environments; multicloud emphasizes the involvement of multiple providers. An environment can be both.\n- Multiregion deployment places resources in different geographic regions, potentially within one provider. Multiple regions do not by themselves constitute multicloud.\n- Multiple accounts or subscriptions separate ownership, billing, or security boundaries within a provider. They do not create provider diversity.\n- Cloud federation involves coordination among clouds, such as shared identity or resource allocation. Simply purchasing services from multiple providers does not establish federation.\n\nAn ordinary on-premises data center is not necessarily a private cloud: private cloud generally implies cloud characteristics such as pooled resources, on-demand provisioning, and automation."
    },
    {
      "heading": "Development and adoption",
      "text": "Multicloud emerged as organizations expanded beyond isolated uses of public cloud services. Different business units selected different platforms, acquisitions brought previously separate environments together, and specialist services encouraged organizations to use providers alongside an existing primary platform.\n\nConsequently, multicloud can be either deliberate or incidental. A deliberate strategy specifies which providers support which requirements, how their environments are governed, and how applications can be moved or recovered. An incidental arrangement may consist of unrelated deployments with separate operational practices.\n\nThe growth of managed databases, analytics platforms, and machine learning services has made provider selection more than a choice of basic computing capacity. Organizations may adopt a second provider for a particular service without intending to distribute their entire application estate."
    },
    {
      "heading": "Architectural patterns",
      "text": "Independent workloads\n\nThe simplest pattern assigns separate workloads to different providers. For example, one provider may host business applications while another supports analytics. Each workload can use its provider’s native services with little cross-cloud communication.\n\nThis pattern offers organizational choice but does not make individual applications portable or resilient to the loss of their hosting provider. A company can be multicloud overall while each application remains dependent on one cloud.\n\nDistributed applications\n\nA single application may span providers, with its components exchanging requests or data across cloud boundaries. A processing service in one cloud might consume information held in a database in another.\n\nThis is a form of distributed computing. Network latency, bandwidth, routing, authentication, and partial failures become architectural concerns. Closely coupled components that exchange many small requests can incur substantial delays when separated across clouds.\n\nApplications organized as microservices are not automatically suitable for this arrangement. Independently deployable components can still have strong runtime dependencies.\n\nRecovery and redundant deployment\n\nA second cloud can hold backups, standby resources, or a parallel production environment for disaster recovery.\n\n- Backup and restore: data and deployment materials are retained for restoration elsewhere.\n- Active–passive: one environment serves production traffic while another is maintained for takeover.\n- Active–active: multiple environments serve production traffic concurrently.\n\nThese patterns differ in operating cost, recovery time, and potential data loss. Active–active systems require particular care with traffic routing, concurrent updates, and consistency. Database replication across providers may depend on database-specific capabilities or external replication software; equivalent managed database products do not necessarily interoperate directly.\n\nPortable deployment platforms\n\nOrganizations may standardize deployment through containers, Kubernetes, or shared application platforms. These can reduce differences in packaging and execution.\n\nPortability remains incomplete if applications depend on provider-specific identity systems, storage behavior, networking, or managed services. Moving a container image is usually easier than moving the application’s data, permissions, and operational dependencies."
    },
    {
      "heading": "Management and interoperability",
      "text": "Multicloud management coordinates environments with different resource models, terminology, and APIs. Common capabilities include:\n\n- resource inventories and ownership records;\n- deployment automation and configuration management;\n- identity federation and permission governance;\n- network connectivity and name resolution;\n- monitoring, logging, and incident handling;\n- billing allocation and cost analysis.\n\nInfrastructure as code can make deployments repeatable and auditable, but a common automation tool does not make provider resources equivalent. Templates may still require provider-specific definitions, and superficially similar services can have different limits or failure behavior.\n\nInteroperability and portability are related but distinct. Interoperability enables systems to work together; portability concerns moving software or data between environments. An application may communicate effectively with another cloud while remaining difficult to relocate there.\n\nA common abstraction layer can simplify operations by exposing a limited set of shared capabilities. However, restricting applications to that common subset can exclude useful provider-specific features. Alternatively, provider-specific implementations can preserve functionality while increasing development and maintenance work."
    },
    {
      "heading": "Data and networking",
      "text": "Data placement often constrains multicloud architecture more strongly than application packaging. Large datasets take time to transfer, may incur outbound transfer charges, and can become tightly connected to local processing services. This tendency is often called data gravity.\n\nCross-cloud connections may use the public Internet, encrypted tunnels, private connectivity services, or interconnection facilities. Private connectivity does not by itself imply end-to-end encryption, and connection security depends on the actual protocols and configuration.\n\nReplicated data introduces questions about update ownership, consistency, and recovery. Synchronous replication can add write latency because operations wait for remote confirmation. Asynchronous replication reduces that waiting but can leave replicas behind the primary system. Network partitions can force trade-offs between accepting operations and maintaining strong consistency, as formalized under particular assumptions by the CAP theorem.\n\nData-residency requirements can also constrain storage, processing, replication, and administrative access. Selecting a geographic region alone does not establish compliance with every applicable requirement."
    },
    {
      "heading": "Security and governance",
      "text": "Multicloud extends security responsibilities across multiple administrative systems. Providers differ in their identity models, default configurations, audit records, and service-specific controls. Uniform policy therefore requires more than copying configuration settings between platforms.\n\nImportant areas include authentication, access control, encryption-key management, secrets handling, and centralized audit collection. Federated identities can reduce duplicated accounts, but a shared identity service may also become a dependency affecting several clouds.\n\nThe shared-responsibility model continues to apply, but responsibility boundaries vary by service. A managed database, a virtual machine, and a hosted application leave different tasks to the customer.\n\nProvider diversity can isolate some failures or security incidents. Conversely, additional accounts, interfaces, credentials, and network paths increase the number of components that must be secured. Common deployment pipelines, privileged credentials, or management services can create correlated exposure across otherwise separate clouds."
    },
    {
      "heading": "Benefits, costs, and limitations",
      "text": "Potential benefits include access to specialized services, greater supplier choice, support for geographic requirements, and reduced dependence on a single provider. Multicloud may also support an exit strategy or improve resilience when applications are explicitly designed for it.\n\nThese benefits are not automatic. Important limitations include:\n\n- Operational complexity: teams must understand multiple platforms, support processes, and failure modes.\n- Transfer and integration costs: network charges, data duplication, and integration software can offset savings on individual services.\n- Reduced purchasing concentration: dividing expenditure can affect negotiated discounts or the utilization of committed capacity.\n- Persistent dependencies: vendor lock-in can remain through proprietary APIs, data formats, operational expertise, and migration costs.\n- Recovery uncertainty: standby infrastructure may lack sufficient capacity, current configuration, or valid access when needed.\n\nAvailability depends on application dependencies rather than provider count alone. If an application requires two clouds to operate simultaneously, failure of either can interrupt service. If either cloud can independently serve the application, redundancy can improve availability, but only when routing, data, identity, and recovery mechanisms also remain functional.\n\nLikewise, a provider’s service-level agreement does not establish an end-to-end guarantee for an application spanning multiple providers. Application-level reliability depends on how the individual services and shared dependencies are combined.\n\nA meaningful multicloud design therefore specifies what must work across providers, what may remain provider-specific, how much data loss or downtime is acceptable, and which recovery procedures have been tested."
    }
  ],
  "links": [
    "cloud-computing",
    "infrastructure-as-a-service",
    "platform-as-a-service",
    "software-as-a-service",
    "data-center",
    "machine-learning",
    "database",
    "distributed-computing",
    "disaster-recovery",
    "database-replication",
    "software-container",
    "application-programming-interface",
    "infrastructure-as-code",
    "interoperability",
    "internet",
    "encryption",
    "cap-theorem",
    "authentication",
    "access-control",
    "virtual-machine",
    "vendor-lock-in",
    "service-level-agreement"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}