ArcelorMittal is expanding its relationship with Microsoft as it moves core IT systems, enterprise data, analytics, and artificial intelligence onto a more unified cloud foundation.

The Luxembourg-headquartered steelmaker has designated Microsoft Azure as its primary cloud computing platform under a strategy it calls “Cloud First, Data Centric”. The company plans to use the platform to modernise core systems, consolidate information, strengthen resilience, and deploy analytics and AI across its global business processes.

ArcelorMittal will integrate services including Microsoft Fabric, Purview, and Foundry, although it has not disclosed a timetable, contract value, or the first operational workloads moving under the expanded arrangement. It said the programme would also reduce dependence on legacy IT and support more efficient decision-making.

Nik Puri, chief information officer at ArcelorMittal, said “technology is central to how we compete”, linking the infrastructure programme directly to operating performance rather than presenting it as a separate innovation exercise.

A cloud programme inside a physical industry

Industrial cloud adoption differs from an ordinary office software migration because the systems involved can stretch from finance and procurement to maintenance, logistics, production planning, quality control, and plant operations. Some workloads can move relatively easily, while others remain tightly connected to specialist equipment, local networks, and processes that cannot tolerate prolonged disruption.

ArcelorMittal operates in 60 countries and has primary steelmaking operations in 14, creating a large and uneven technology estate. The group generated revenue of $61.4 billion in 2025 and produced 55.6 million tonnes of crude steel, so common data standards and shared platforms could influence purchasing, energy use, inventories, customer orders, and asset maintenance.

Yet consolidation does not automatically create standardisation. Plants built at different times, operating under different national requirements, and using equipment from numerous suppliers can encode processes in incompatible formats, while information that appears comparable at group level may have been collected differently on the factory floor.

The difficult work will involve deciding which data can be trusted, who owns it, how it is classified, and where local operating requirements should override a global template. Without that work, analytics can reproduce inconsistencies more quickly, while AI systems may generate plausible outputs from information that has never been reconciled.

ArcelorMittal said Microsoft Purview would form part of the programme, suggesting that governance, discovery, and control are intended to accompany the technical migration. Those controls become particularly important where operational, employee, supplier, commercial, and equipment data are combined inside the same analytical environment.

Industrial AI depends on reliable systems

Manufacturers have numerous potential uses for machine learning and generative AI, ranging from equipment monitoring and quality inspection to production planning, engineering support, purchasing, and the retrieval of technical knowledge. However, those systems only become useful when connected to live operational processes and when their recommendations can be checked against physical conditions.

Inside a steel business, an inaccurate office summary may waste time, while a poor recommendation affecting maintenance, materials, or production scheduling can create a much larger operational consequence. Deployment should therefore depend on the risk of the task, the quality of available information, and the point at which a qualified employee must review or override a result.

The programme also concentrates more of ArcelorMittal’s technology operations on a large US cloud provider. Microsoft can supply global infrastructure and an integrated set of data and AI services, but ArcelorMittal will still need to manage commercial dependence, exit planning, portability, service resilience, and the distribution of technical skills between its own teams and external suppliers.

Those questions become sharper as European policymakers and businesses debate control over strategic digital infrastructure. A steel producer cannot treat its data estate as interchangeable with a short-lived software product, since systems may be expected to support plants, assets, and regulatory obligations for decades.

The expanded partnership is nevertheless more substantial than a corporate AI trial because ArcelorMittal is tying model deployment to the replacement of legacy systems and the consolidation of enterprise data. That sequence reflects a wider shift among large organisations, which are finding that the limiting factor for AI is often not access to a model but the condition of the systems and information around it.

ArcelorMittal has not published the operational targets by which the programme will be assessed. Useful measures would include the number of retired legacy applications, system reliability, the speed of deploying changes across plants, reductions in manual data handling, security performance, and evidence that production or maintenance decisions have improved.

Plant-level adoption will also determine whether the common platform becomes an operating system for the group or remains an enterprise layer above local processes. Engineers, operators, maintenance teams, and plant managers will need incentives and authority to change established workflows, rather than being asked to feed another central reporting environment.

Until those measures emerge, the expanded Microsoft relationship remains a statement of architecture and intent. Its commercial value will become visible when shared platforms alter how individual mills, mines, offices, and supply chains operate, without creating new points of fragility inside an industry where digital systems ultimately answer to physical production.