From Better Models to Better Enterprise AI Data Context
Enterprise AI data context is the structured, shared understanding of an organization’s data, relationships, rules, and goals that allows AI systems to behave like informed insiders rather than isolated tools reacting blindly to prompts. At Microsoft’s Build conference, this concept moved to center stage. Microsoft argued that the hard part of enterprise AI is no longer picking a powerful model, but giving agents access to reliable, governed context about how the business works. Amir Netz, CTO of Microsoft Fabric, framed the difference as AI behaving like an employee, not “a stranger on the outside.” This shift recasts the enterprise AI race: once foundation models reach a baseline of quality, competitiveness hinges on how well organizations integrate operational data, analytics, and semantics into a consistent context layer that all AI agents can share.
Azure HorizonDB: A Database Built for AI-Scale Context
Azure HorizonDB, now in public preview, is Microsoft’s new PostgreSQL-compatible database designed to keep enterprise AI context close to both transactions and models. It scales storage elastically up to 128 TB and compute to 3,072 vCores, while delivering sub-millisecond multi-zone commit latency for demanding workloads. HorizonDB folds vector search, integrated AI model management, and direct connections to Microsoft Foundry and Fabric into one platform, so teams can store transactional data and AI-friendly embeddings in a single system. According to NASDAQ’s Mohsin Shafqat, HorizonDB “brings transactional data, vector search, and AI capabilities into a single platform, which simplifies the architecture without forcing a complete rethink.” The goal is AI agents that can query live operational data, search over semantic representations, and call models without jumping across brittle integrations, reducing context loss at every step.
GPU-Accelerated Fabric Data Warehouse and Low-Latency AI Workloads
To keep shared organizational context usable at scale, Microsoft is adding GPU acceleration to Fabric Data Warehouse in early access preview. The company says it has integrated NVIDIA accelerated computing directly into the warehouse, with no query rewrites, delivering up to 7x faster performance than three unnamed competitors at 64-user concurrency in internal benchmarks from May 2026. Netz contrasted this with usual expectations: in warehousing, a 10 percent yearly gain is celebration-worthy, but GPUs have yielded 5x to 100x speedups in Microsoft’s research, which won Best Industry Paper at ACM SIGMOD 2026. UNC Health reports up to 5x faster queries, freeing staff to focus on insights over tuning. For AI agent context management, this matters because many agents and humans query the same semantic layer at once; GPU-backed warehousing helps keep that shared context responsive under heavy load.
Fabric IQ and Ontologies: Encoding Shared Organizational Memory
Fabric IQ, now generally available, is the semantic and ontology layer that turns scattered enterprise data into a consistent organizational memory. It extends Power BI semantic models, already used by hundreds of thousands of organizations, with richer business entities, relationships, rules, real-time signals from Fabric Real-Time Intelligence, and explicit definitions of which actions agents may take. Operations agents, now GA, sit on top of this layer to monitor live data and act on predefined business logic, while upcoming ontologies broaden the coverage of domain-specific concepts. Microsoft is wiring Fabric IQ into its broader agent ecosystem: it appears as a knowledge source in Microsoft Foundry, a first-party MCP tool for Microsoft Agent 365, and a grounding layer for Microsoft 365 Copilot, including Cowork and Copilot Chat. This gives agents a governed, consistent context graph instead of ad hoc prompt engineering.
Unified Context Across Past, Present, and Future Workflows
Beyond individual components, Microsoft is pitching Fabric as the unified platform where enterprise AI agents store, reason over, and act on shared context. Graph in Fabric models relationships between systems and business entities, while new planning features can write forecasts back into Fabric. Netz describes the result as temporal coverage: a context layer that spans past data in OneLake, live operational signals, and forward-looking plans, so agents see not only what happened and what is happening, but also what is supposed to happen. A Database Hub in Fabric will centralize HorizonDB, Azure Database for PostgreSQL, and Azure Cosmos DB, with data mirrored into OneLake, while Cosmos DB gains semantic reranking and an agent memory toolkit. Together with the Rayfin SDK for building backends directly on Fabric, these moves show how shared organizational context is becoming the main design surface for enterprise AI, rather than raw model horsepower.
