Manufacturing

Plant data is split between operational technology (controllers, machines, historians), engineering systems such as PLM and CAD, and enterprise planning and finance systems. Microsoft's own manufacturing overview concedes that most manufacturers piloting generative AI have yet to see significant cost reductions and that proving return on investment remains a challenge, which is why it puts AI-ready data ahead of agents.

Design and process know-how is often a manufacturer's most valuable asset, and it is a target. ASM, a semiconductor equipment maker whose sites and customers span Asia, Europe and North America, describes threats from opportunistic attackers and state-sponsored adversaries, including espionage, insider risk and advanced phishing, handled by a security team that had to investigate faster without growing.

Brownfield factories also run long-lived equipment and edge servers that were never designed for cloud connectivity, so every new AI use case raises questions about segmentation, latency and who maintains the factory edge.¹²

Microsoft for Manufacturing, one of the Microsoft Industry Solutions, combines Microsoft and partner solutions across four areas: digital engineering, resilient supply chains, intelligent factories and connected customers. Microsoft says its manufacturing strategy rests on agentic AI, a digital thread of AI-ready data built in Microsoft Fabric, and edge-to-cloud infrastructure through Azure's adaptive cloud approach. Its manufacturing page names Azure IoT Operations, Microsoft Foundry, Microsoft Fabric, Azure Machine Learning, Dynamics 365 Supply Chain Management and Power BI among the products involved.

A sensible sequence starts with data rather than models: connect machine and quality data at the edge, land it beside ERP and PLM data in one governed store, then add copilots for engineers and agents for narrow frontline tasks such as maintenance triage. Security runs alongside. ASM connected Microsoft Security Copilot across its Microsoft security tools so analysts can query in plain language, combine signals and follow suggested next steps, and it wrote its own investigation methods into Security Copilot workflows. ASM calculates that this saves around 337 analyst hours a week; the figure is its own estimate with no published baseline.

Much of this is unnecessary for a single plant with a working MES and historian, whose built-in dashboards often cover line monitoring and OEE reporting. The Microsoft platform layers earn their place when several sites must share one data model, when agents need governed access to engineering and operations data at once, or when a small security team must cover many sites.¹²³