Summary
- Microsoft opened 31 data centres across five continents in its fourth quarter and 88 during the year.
- Approximately 1GW of capacity was added during the final quarter.
- The build rate increases competition for power equipment, contractors, cooling systems, and commissioning staff.
Microsoft brought 88 data centres online during its 2026 financial year, including 31 facilities across five continents in the final quarter.
The technology group added approximately 1GW of capacity during the quarter and said it remained on course to roughly double its overall capacity within two years.
Microsoft also cut the period between receiving new graphics processors and making them operational in its largest regions by almost 50%, drawing building delivery, electrical commissioning, cooling readiness, networking, and hardware installation into a tighter programme.
Buildings and hardware arrive on one schedule
The 88-site figure covers different facility types, expansions, and regions rather than 88 identical hyperscale campuses. Microsoft has not published the location, capacity, ownership model, or development structure for every opening.
Its full year earnings material still provides an unusually detailed view of the speed at which a hyperscaler is converting capital into operational infrastructure.
Finance leases reached $5.6 billion during the quarter, primarily associated with large data centre sites, while cash paid for property and equipment amounted to $35.8 billion.
The group said its Cobalt 200 server racks would be installed in more than 25 data centres by the end of July. Each deployment depends on available power, cooling, network fabric, physical security, and operational acceptance before the equipment can serve customers.
AI accelerators have increased the cost of poor sequencing. A processor waiting for a completed hall produces no cloud revenue, while an energised building without hardware leaves power, cooling, and capital underused.
The emphasis on reducing dock-to-live time extends the project beyond construction completion. A building may be structurally finished while integrated testing, network deployment, rack installation, firmware work, and software validation remain outstanding.
Customer hardware schedules can also move between regions. A hall designed for one accelerator generation may need changes to power distribution, liquid cooling, or network architecture before it reaches full utilisation.
Europe competes inside the global programme
Microsoft did not provide a European split for the 88 openings, although the global build rate affects regional contractors and suppliers because equipment and engineering resources are allocated across one international programme.
Transformers, switchgear, generators, UPS systems, chillers, coolant distribution units, busway, and controls are often sourced through global frameworks. A major deployment in the US or Asia can draw on factory capacity needed by projects in Ireland, the UK, Germany, France, Spain, or the Nordics.
The same pressure reaches specialist labour. Hyperscalers and delivery partners move project managers, commissioning engineers, controls specialists, and operations staff between markets, while local electrical and safety requirements still need to be met.
Power remains the least standardised element. Microsoft can repeat hardware, software, and building designs, but each grid connection sits within a different queue, reinforcement plan, tariff structure, and planning regime.
Sites with deliverable substations and confirmed network work are likely to progress faster than projects dependent on uncertain transmission upgrades. Capital alone cannot create physical grid capacity where the required infrastructure has not been built.
Cooling architecture is changing during the programme as rack density rises. New facilities increasingly need liquid-ready or hybrid designs, while existing sites may require pipework, coolant distribution, heat rejection, and control upgrades.
Standardisation helps procurement, but local climate and water conditions still influence cooling. A design suited to the Nordics may require different heat-rejection equipment in Spain or southern France.
The scale of investment also raises utilisation risk. Microsoft needs enough capacity to prevent customer constraints, yet overbuilding would tie up capital in facilities and hardware that depreciate before reaching the planned load.
Azure surpassed $100 billion in annual revenue, while Azure and other cloud services grew 43% in the fourth quarter. Demand is therefore supporting both infrastructure spending and operating growth.
European planning and grid systems will be measured against regions where Microsoft can bring comparable capacity online more quickly. A slow market risks losing future phases even when customer demand remains strong.
Eighty-eight openings record the breadth of the programme, but the more demanding measures will be delivered megawatts, utilisation, energy efficiency, and the length of time between construction completion and productive customer service.
