Microsoft, Amazon and Google reported a combined 119 million metric tonnes of carbon-dioxide-equivalent emissions for the financial year ending in March 2026, up from roughly 101 million a year earlier. The Guardian identified data-center construction as a principal driver of the increase. The total makes cloud expansion physical, but it does not assign every tonne to artificial intelligence. [1]
Friday's account of Meta's Alberta campus and its linked gas plant argued that an AI project becomes real through power, water and cost records rather than a capacity target. The new disclosures widen that ledger from one proposed campus to three operating companies. They do not resolve Meta's permits, water use or ratepayer allocation.
The aggregate is useful because 119 million tonnes is a scale that corporate climate language can otherwise divide among separate reports. It is also limited. Each company sets reporting boundaries across direct operations, purchased energy and supply chains. Adding the disclosed totals creates a broad physical picture; it does not make every scope definition or fiscal series interchangeable.
Microsoft reported 20 million tonnes, a 25 percent increase, which it said was driven primarily by expanding data-center infrastructure. Google reported an 18 percent increase and pointed to supply-chain activity supporting business growth. Amazon reported a 16 percent overall rise and a 20 percent increase in supply-chain emissions, including data-center building and construction. [1]
Those descriptions matter more than a single AI label. Concrete, steel, chips, power equipment and construction services can raise emissions before a server trains or runs a model. Cloud services also store ordinary business data and host non-AI workloads. AI demand is part of the boom described by the companies and the Guardian, not an audited allocation of all 119 million tonnes.
The customer side creates another accounting problem. University College London economist Cecilia Rikap told the Guardian that businesses moving data and AI work into the cloud can shift the reported digital footprint to the cloud provider. The emissions do not disappear when they leave a customer's own inventory; they move into another company's disclosed system. [1]
That shift is why the aggregate matters to buyers as well as shareholders. A company can report a lighter internal technology footprint after outsourcing servers while still consuming the same or greater computing capacity. Readers need supplier and customer accounts that show the boundary, avoid double counting and prevent physical emissions from vanishing between ledgers.
All three companies retain net-zero targets: Microsoft and Google for 2030 and Amazon for 2040. Google also says its AI systems helped avoid 41 million tonnes of emissions elsewhere. Such claims may describe real benefits, but avoided emissions and emitted tonnes are different series. One cannot be silently subtracted from the other without compatible methods and verification. [1]
No verified topic-matched X status surfaced, so boom and hypocrisy cannot be presented as a measured online consensus. The Guardian's comparison supplies the stronger record: 119 million tonnes, roughly 18 million more than the prior year, with construction prominent and company boundaries still unlike.
The next useful disclosure is not another global promise. It is a comparable account of construction, electricity, supply chains, offsets and customer use, tied to the grids and communities carrying the load. The cloud can relocate a company's emissions report. It cannot outsource the atmosphere.
-- DARA OSEI, London