Without data, enterprise AI isn't going to be successful.
Getting all the data in one place and having the right type of data tools, including connections to different types of databases is a critical aspect of having the right data for AI.
There are multiple vendors all vying to be the data platform of choice for enterprises today, with Databricks, Snowflake, Google and Amazon among the big options. Microsoft has increasingly been active in the space with its Microsoft Fabric technology first announced in 2023 and expanded in the years since with data tools to accelerate AI workflows. In 2024, Microsoft claimed that 70% of the Fortune 500 used Fabric, now in 2025 that figure has moved higher to 80%.
Microsoft Fabric is a unified data platform that combines data lakes, databases, data warehouses, real-time analytics and business intelligence into a single service. It eliminates the complexity of managing multiple data tools while providing the Microsoft OneLake, which is a virtualization layer that can connect to data across clouds without requiring migration.
Microsoft is now integrating LinkedIn's proven graph database technology into the platform. The graph database addition addresses a fundamental challenge plaguing enterprise AI deployments. Vector databases excel at semantic search, but they struggle to understand relationships between data entities. Graph databases fill this gap by modeling connections between customers, suppliers, network devices, or any business entities. This creates a knowledge graph that provides crucial context for AI applications.
"Graph databases are incredibly important because you know there is only so much data you can put into perfectly structured tables," Arun Ulag, corporate vice president for Azure Data at Microsoft, told VentureBeat. "The actual world that we live in is full of relationships, relationships between people, relationships between customers, relationships within suppliers, the way supply chains work the way cloud systems work, everything is connected and you really need a very good graph database."
LinkedIn's enterprise graph engine comes to Fabric
The graph capability isn't built from scratch. Microsoft moved a substantial portion of LinkedIn's graph database team into Azure Data approximately 18 months ago. The goal was to adapt the technology that powers LinkedIn's massive social network for enterprise use cases.
"The graph database allows you to collect the set of entities that matter for you to be able to run your vector search on," Ulag explained. "Graph databases narrow the solution space into the ones that matter, and then the vector index allows you to go zoom in even further."
The technical implementation reveals Microsoft's strategy for optimizing AI performance through a two-stage data narrowing process. First, the graph database identifies relevant entities based on relationships. For example, all suppliers connected to a specific customer or all network devices linked to a particular data center. Then, vector search operates within that constrained set to find semantically relevant information.
This approach could significantly improve AI response accuracy and speed. Rather than searching across an entire data lake for relevant information, AI systems can focus on a pre-filtered set of connected entities. This reduces both computational overhead and the risk of retrieving irrelevant data.
The graph database supports standard GraphQL queries and integrates with Fabric's existing data lake architecture with all data remaining available in open-source data formats.
Beyond social networks: Enterprise graph use cases
Graph databases have traditionally been associated with social networks and fraud detection. However, Microsoft and industry analysts see broader applications emerging, particularly for agentic AI systems that require persistent memory and context.
"Graph databases can help with some very old, but some very well established use cases like fraud detection, which they're great at that, but they can also serve as some very modern, forward looking capabilities, such as bringing a memory to an agentic system," Brad Shimmin, VP and Practice Lead for Data and Analytics at Futurum Group, told VentureBeat.
The graph database also strengthens Microsoft's competitive position against Databricks, Snowflake and Google Cloud's data platforms. According to Futurum Group's analysis, Microsoft Fabric ranks in the "Elite category" alongside Google and Databricks. But the graph capability provides a differentiator that competitors currently lack.
"Microsoft bringing Graph into fabric, it's a no brainer," Shimmin noted.
Microsoft's approach integrates graph capabilities directly into the data platform rather than offering it as a separate service. This aligns with the broader industry trend toward unified data intelligence platforms. The integration means enterprises can work with graph, vector, geospatial and traditional relational data within a single platform. They avoid complex data movement or synchronization.
More data sources come to Fabric
Beyond graph databases, Microsoft announced several other Fabric enhancements that strengthen its enterprise positioning:
-
Expanded data source integration: New mirroring capabilities for Oracle databases and Google BigQuery allow enterprises to virtualize data from these sources in near real-time.
-
Enhanced geospatial capabilities: Native mapping functionality, powered by Azure Maps technology, enables large-scale geospatial analysis integrated with real-time data streams. This could prove valuable for logistics, retail and IoT applications.
-
Developer experience improvements: The new Fabric Extensibility Toolkit and Model Context Protocol (MCP) integration make it easier for developers to build custom applications and integrate with AI development tools.
Enterprise evaluation framework
Industry analysts offer specific guidance for enterprises evaluating data platforms.
From a technical perspective, Shimmin identifies five critical capabilities enterprises should evaluate:
-
Unified Lakehouse Architecture: A single, cohesive platform that merges the scalability of a data lake with the performance and transactional reliability of a data warehouse
-
Native Support for Open Table Formats: Deep, first-class support for standards like Apache Iceberg and Delta Lake to prevent vendor lock-in
-
Integrated MLOps/LLMops Framework: Built-in capabilities to manage the entire AI lifecycle, including data vectorization, model development, deployment, and ongoing governance and monitoring
-
Hybrid and Multi-Cloud Portability: A container-native architecture that ensures a consistent operational experience and unified governance layer across any cloud or on-premises environment
-
Decoupled, Multi-Engine Compute: An elastic architecture that separates storage from compute and supports multiple workloads on the same data, enabling performance isolation and efficient, consumption-based pricing
Gartner which also ranks Microsoft Fabric highly also suggests that enterprise alignment is a critical part of the decision making process.
"Enterprises evaluate data management vendors by aligning their data and analytics strategy to technology choices, including database management, data integration, metadata management, and related categories," Gartner analyst Thornton Craig, told VentureBeat. "Data management platforms, including Microsoft Fabric, offer a comprehensive vision for overall data management."
Strategic implications for AI adopters
Having a flexible data platform that handles all data type should be table stakes for any enterprise serious about its AI strategy.
The open approach that Microsoft has taken with Fabric, enabling it to connect, mirror and use all types of data is an attractive feature to many enterprises. The graph database integration signals Microsoft's broader strategy of building AI-ready data infrastructure rather than simply adding AI features to existing platforms. For enterprises already committed to Microsoft's ecosystem, the graph capability provides immediate value without additional vendor relationships or data movement.
However, organizations should approach platform decisions strategically.
"The trick is to be able to kind of align the solutions you're looking at with both the outcome of what you're trying to do, are you trying to save money or trying to make money, and how well that aligns with the estate that you're sitting on," Shimmin advised. "If you're sitting on a data swamp and everything's a mess and you can't get to it, you're going to need to invest in a platform that's very flexible in terms of being able to adapt and bring in a lot of different and disparate data sources."
