Architecting Cloud-Native AI Ecosystems

Cloud-native AI describes an emerging approach to creating and deploying artificial intelligence capabilities alongside the emerging approaches and best practices of cloud-native software development, which itself is following and expanding upon the DevOps and Agile lines of thought. Cloud-native AI combines the power of scalable storage and compute capabilities with cloud-native software principles and best practices in order to create AI models and to deliver AI functions to end users.

Conventional (or “non-cloud-native”) AI development platforms and ecosystems are characterized and driven by a number of factors: relatively slow and therefore relatively costly model-building processes; monolithic services that combine many functions and that are deployed within the same process; the use of virtual machines to deploy services; complex and monolithic production environment constructs; the use of platform as a service (PaaS) renditions such as IBM Watson; and the general absence of AI development tools—services and pipelines—into the AI model-building process. Cloud-native AI employs cloud-native design principles to minimize and often to eliminate these external constraints. Taking advantage of containerization, microservices hosting platforms and orchestration frameworks such as Kubernetes, serverless compute, and DevOps-based AI model and AI service development guidelines, Cloud-native AI—typically deployed in AI developer laboratories—enables significantly faster and less costly AI model-building activities. The APIs and containers of the cloud-native AI ecosystem can then be made available for deployment in the more productionized environment of the AI model execution environment.