Their 2026 sustainability reports show total emissions up 25% and 16% year over year as data center buildout for AI accelerates, and the pollution per dollar and the practice of matching electricity use to wind and solar purchases both companies
Clean-cloud computing was a story about matchable electricity. AI has turned it into a story about what that electricity now costs.
For a decade, Big Tech sold the public a tidy picture: every search, every email, every streamed movie running on power matched, dollar for dollar, to wind and solar. The setup worked because the load was small, the matching was cheap, and the accounting was uncontested. That picture is now visibly straining, and not because the press has changed its mind. The companies' own 2026 sustainability reports make the case.
Google's 2026 sustainability report shows total emissions up roughly 25% year over year. Amazon's matching report puts its total up 16%. Both rises trace primarily to Scope 3, the indirect emissions from the supply chain and the products the companies sell, and inside Scope 3 the dominant driver is data center construction: the steel, concrete, chips, and turbines that go into the buildings behind AI. Google's own accounting puts Scope 3 up by roughly 2.1 million tonnes year on year, with capital goods the main mover.
The two firms still loudly defend their green credentials. Amazon's Climate Solutions page advertises that the company's electricity use has been 100% matched with renewables for two consecutive years. Google points to similar matching on its Data Centers: Operating Sustainably page. Both companies have reiterated their net-zero pledges, Google by 2030 and Amazon by 2040. And both now lean heavily on carbon intensity, pollution per dollar of revenue, as the metric that demonstrates progress even as absolute emissions climb.
The trouble is that the sustainability reports themselves now treat these promises as under strain. The 2026 Google report states that its 2030 net-zero ambition has become "more complex and challenging across every level" and that navigating them requires flexibility. That language is notably more candid than the previous edition. Amazon has not been as explicit, but the direction of travel is the same. Absolute emissions are climbing even as the companies' headline metrics stay tidy.
Why the disconnect? Two reasons the ledgers alone do not capture.
The first is physical. Google data center energy use roughly doubled in four years, according to TechCrunch's read of the company's earlier disclosures, and that pace is if anything accelerating as AI training and inference workloads scale. Meeting that load means building more generation, and increasingly that means natural gas turbines, the on-demand power equipment data centers depend on when the grid is tight. Wood Mackenzie reports gas turbine prices have soared 195% on a supply-demand squeeze. Renewable matching does not change the fact that the underlying megawatt-hours come from somewhere.
The second reason is conceptual. Carbon intensity is a ratio, not an absolute. A company can hold or even improve the pollution-per-dollar figure while absolute pollution rises, as long as revenue grows faster than emissions. That trick works until regulators, investors, or neighbors start asking about the input side of the equation: the actual tons of CO2, the actual liters of cooling water, the actual molecules of unburned methane leaking from the turbines the data centers run on. Carbon intensity was, as the new TechCrunch analysis notes, the same metric China preferred in climate treaty talks for exactly that reason.
None of the companies themselves attribute their emissions increases to AI. The link is inferential, drawn from the Scope 3 capital goods bucket and the known scale of AI infrastructure spend. That caveat matters: it is the difference between a warning sign and a confession.
What to watch: whether the 2027 reports add an explicit AI-attributed emissions line item, whether gas-turbine capex shows up directly in capital expenditure disclosures, and whether regulators in the EU and California start probing renewable matching certificates rather than accepting them at face value. The clean-AI story is not yet dead. The 2027 reports will reveal whether the companies have to choose between AI growth and their stated net-zero dates, or whether the accounting can still hold them up.