AWS re:Invent 2025 - Architecting multicloud solutions from data mesh to generative AI (HMC210)

Architecting Multicloud Solutions: From Data Mesh to Generative AI

Defining Multicloud

• Multicloud refers to running workloads, IT solutions, or applications on more than one cloud service provider
• Key drivers for multicloud include:
  • Mergers and acquisitions (M&A) leading to disparate cloud environments
  • Leveraging differentiated cloud capabilities
  • Addressing regulatory requirements like data sovereignty and cloud concentration risk

Cloud Maturity Model

• Multicloud maturity is assessed across people, process, and technology dimensions
• Key focus areas:
  • Upskilling employees and leveraging AI/ML to address multicloud skill gaps
  • Establishing a centralized cloud center of excellence (CCoE) with specialized roles
  • Adopting multicloud architectural patterns and services

Data Strategy Definitions

• Gartner defines data strategy as a "highly dynamic process" for acquiring, organizing, analyzing, and delivering data
• AWS defines data strategy as encompassing people, process, technology, and rules for data as a strategic asset

Multicloud Data Challenges

• Data gravity: Data accumulation in a single cloud discourages multicloud adoption
• Data governance and control: Implementing consistent policies and controls across clouds
• Mergers and acquisitions: Integrating disparate data environments post-acquisition

Recommended Architectural Patterns

1. Materialized Views:
   • Pre-compute and store data aggregations to avoid repeated processing across clouds
   • Supported by cloud-native services like BigQuery Omni, Amazon Athena, and Apache Iceberg
2. Federated Queries:
   • Query data across clouds without moving the data, using services like Amazon Athena Federated Query
   • Requires careful management of metadata, access controls, and performance optimization
3. Data Mesh:
   • Decentralized data architecture with domain-oriented data products
   • Enables self-service data access and sharing, but requires upfront standardization
4. Metadata Catalogs:
   • Maintain a centralized view of data assets and lineage across cloud environments
   • Ensure consistent metadata synchronization and access control management

Multicloud Generative AI

1. Retrieval Augmented Generation (RAG):
   • Combines user queries with relevant knowledge base content to generate augmented responses
   • Supports structured RAG (SQL queries) and graph RAG (leveraging graph databases)
   • Requires consistent embedding models and resilient vector stores across clouds
2. Model Context Protocol (MCP) and LLM Gateways:
   • MCP provides a standard interface for agents to access data across clouds
   • LLM gateways enable routing requests to language models hosted in different cloud providers
   • Addresses model availability, capacity, and cost-based routing considerations

Key Takeaways

• Multicloud is a reality driven by M&A, differentiated capabilities, and regulatory requirements
• Effective multicloud data strategies require addressing challenges like data gravity, governance, and integration
• Architectural patterns like materialized views, federated queries, data mesh, and metadata catalogs can help overcome these challenges
• Generative AI in multicloud leverages techniques like RAG and MCP to access distributed data and models
• Careful planning around security, performance, and cost optimization is crucial for multicloud success