Microsoft’s carbon emissions climbed 25 percent last year. Amazon’s rose 16 percent. Google’s jumped 18 percent. All three companies are chasing the same thing, and it is the reason their climate math has stopped adding up: artificial intelligence.
The electricity bill nobody budgeted for
For years, Microsoft, Amazon and Google told investors and regulators they were on track for ambitious climate targets. Microsoft pledged to be carbon negative by 2030. Google set a net-zero goal for the same year. Amazon gave itself until 2040. Those pledges were built before the current AI boom, when data center footprints looked nothing like they do now.
New corporate disclosures from 2026 show what changed. Microsoft’s reported emissions from purchased electricity jumped 945 percent between 2024 and 2025, according to its own sustainability filings, while its total electricity consumption rose 24 percent. Some of that spike reflects an accounting shift, Microsoft says, as it moved away from relying on renewable energy certificates tied to existing projects and toward direct investment in new carbon-free power generation. But the underlying driver is simpler: training and running large AI models takes an enormous, and growing, amount of electricity, and Microsoft’s data centers are multiplying to keep pace with demand for Copilot, Azure AI services and OpenAI’s infrastructure needs.
Amazon and Google are seeing the same pressure from a different angle. Amazon’s emissions rose to roughly 81 million metric tons of carbon dioxide equivalent, up 16 percent from 2024, as AWS builds out capacity for AI workloads. Google’s emissions climbed 18 percent on an ambition-adjusted basis, with emissions from its own operations up 20 percent year over year, a figure the company has directly tied to data center expansion and the energy intensity of AI computing.
A rare bright spot in the numbers
Not every metric moved in the wrong direction. Microsoft says it returned more water to watersheds globally last year than it withdrew, and that its data center water-use efficiency improved 25 percent from a 2022 baseline. That matters because AI data centers are notoriously thirsty, relying on water for cooling systems that keep dense racks of GPUs from overheating. Efficiency gains there suggest the industry has found at least some ways to blunt AI’s environmental cost even as its carbon footprint grows.
Still, water efficiency does not offset a near doubling of purchased-electricity emissions. Climate researchers who track corporate disclosures note that the current trajectory puts Microsoft’s 2030 carbon-negative pledge in serious jeopardy unless the company can pair AI growth with a much faster buildout of new clean power, not just renewable energy credits.
Why this is happening now
The timing is not a coincidence. 2025 and early 2026 saw the industry move from experimenting with generative AI to deploying it at massive scale, embedding it into search, productivity software, cloud services and consumer devices. Every additional query, every model fine-tune, every new data center campus adds load to a grid that, in many regions, still leans heavily on fossil fuels to meet peak demand. Hyperscalers have signed a wave of nuclear and renewable power purchase agreements to try to keep pace, but construction of new clean generation capacity moves far slower than the rollout of AI products.
That mismatch is becoming a defining tension of the AI era. Tech companies want to keep scaling their models because the competitive stakes are high and customer demand is real. At the same time, the same companies made public commitments, often years ago, to cut their climate impact on a fixed schedule. Something has to give, and so far it has been the emissions numbers, not the AI roadmaps.
What to watch next
Expect more scrutiny of how hyperscalers count their emissions, particularly the use of renewable energy certificates versus direct investment in new power generation, an accounting distinction that can make the same physical footprint look very different on paper. Also expect continued investment in nuclear power, including small modular reactors, as tech companies search for a source of carbon-free electricity that can scale as fast as their data centers do. Whether any of the big three can still hit their original climate targets by the end of the decade looks increasingly like an open question rather than a settled commitment.
For more coverage of how artificial intelligence is reshaping energy and climate strategy, visit Mylistingo.
The Netherlands, for internationals
One email a week: the week's news for internationals in the Netherlands, new practical guides and what changed in the cities. No spam, unsubscribe any time.
