Summary
- Microsoft reported a 25 percent year-on-year increase in total emissions in its 2026 sustainability reporting.
- The increase was linked to data centre expansion and changes in renewable energy certificate accounting.
- The report includes facility-relevant metrics around electricity matching, PUE, water, recycling, and infrastructure efficiency.
Microsoft has reported a sharp rise in emissions in its 2026 environmental sustainability reporting, exposing the strain that AI infrastructure growth is placing on earlier corporate climate targets.
The company’s 2026 Environmental Sustainability Report covers fiscal year 2025 and measures progress against its 2020 baseline. Microsoft says the growth of AI is changing the responsibilities that come with building technology at scale, with data centre infrastructure now central to its environmental performance.
Trade reporting on the report identified a 25 percent year-on-year increase in total Scope 1, 2, and 3 emissions, driven predominantly by data centre expansion and a change in how renewable energy certificates are counted. The report also restates Microsoft’s targets to become carbon negative, water positive, and zero waste.
AI growth strains carbon targets
New data centres carry emissions before a server is switched on. Steel, concrete, electrical equipment, cooling plant, backup systems, batteries, servers, and grid infrastructure all add supply-chain and construction impacts. Fast AI build-out therefore pushes emissions through procurement and construction as well as electricity consumption.
As AI workloads scale, efficiency gains can be overtaken by absolute growth. High-density compute can require more power per rack, more complex cooling, larger electrical systems, and accelerated hardware cycles. A facility may operate efficiently while the company’s total infrastructure footprint still expands quickly enough to increase emissions.
Microsoft says it matched 100 percent of its annual electricity consumption with renewable energy in FY25. Annual matching remains an important procurement measure, but the sector is moving towards more demanding questions around additionality, hourly matching, grid location, and the alignment between renewable supply and data centre load.
Those questions are becoming more exposed in Europe, where governments want more AI infrastructure but are also dealing with electricity network constraints, reporting obligations, and local scrutiny of large industrial loads.
Facility metrics come into view
The report’s useful data centre material sits in its facility and supply-chain metrics. Power usage effectiveness, water usage, hardware reuse, battery recycling, construction waste diversion, and local water projects give a more concrete account of infrastructure performance than broad climate targets alone.
Those metrics show how sustainability scrutiny is becoming operational. A hyperscaler can finance renewable energy, but it still has to obtain grid connections, manage local water impact, reduce construction emissions, handle retired equipment, and explain how new AI capacity fits into national and regional energy systems.
Microsoft’s report includes examples of water and ecosystem projects near data centre locations, including work in Europe. Its Amsterdam section refers to rainwater collection and aquifer storage designed to collect more water than is needed for cooling, with excess volumes supporting local reserves.
Such projects will become more visible as data centres face closer scrutiny over water use, especially in markets where cooling choices, climate pressure, and local resource availability collide. A water-positive target will be judged less by global accounting language and more by evidence near the facilities that draw public attention.
Hyperscaler sustainability performance is becoming harder to read through single headline targets. Renewable matching, carbon removal, efficiency improvements, water replenishment, embodied carbon reduction, and grid investment all move at different speeds. AI infrastructure growth can outpace progress in several of those areas at once.
European policy is likely to sharpen that pressure. The EU is moving towards more formal data centre reporting on energy and water performance, while national governments are trying to encourage AI infrastructure without ignoring grid and environmental constraints. Hyperscaler disclosures will increasingly sit beside permitting disputes, grid-connection queues, and local environmental objections.
Microsoft’s latest figures do not show a retreat from sustainability commitments. They show the physical cost of scaling AI and cloud infrastructure at speed. The burden is now visible in emissions accounts, not just in planning documents, supply-chain warnings, and substation queues.
