Ex-Microsoft Paper: AI's Fossil-Fuel Boost Dwarfs Data Centers

A new paper in NPJ Climate Action argues that the standard way of counting AI's climate footprint misses most of the damage. Futurism reported on the study by Will and Holly Alpine, two former Microsoft managers, which calculates that AI could enable 'anywhere from three to 13 times more fossil fuel emissions' than the pollution from running the data centers themselves.

The framing they push is a contrast between 'enabled emissions' (the extra oil and gas produced because AI made extraction cheaper or more efficient) and 'avoided emissions' from AI's use in renewables and efficiency. Their finding is that the enabling side is winning by a wide margin. 'It is difficult and frustrating for us to see the conversation around the true climate impacts of AI just stop at operational emissions,' Holly Alpine told Futurism.

The paper points at direct commercial ties as evidence, not abstract theory. Microsoft has a 20-year contract with Chevron tied to a 'massive natural gas plant,' and Amazon is described as building 'the largest fossil fuel plant in the entire United States.' 'There are teams of engineers and salespeople at these tech companies that are explicitly for the fossil fuel industry,' Alpine said. Two AI experts in our Who's Who directory have shared the paper, which tracks with a growing appetite among practitioners for a broader emissions accounting than the operational-only story that has dominated coverage.

The obvious limitation is that a three-to-13x range is a wide one, and the authors are not neutral observers; they are ex-Microsoft insiders now pushing publicly for stricter guardrails on the fossil-fuel side of these partnerships. The Futurism write-up also does not spell out how the multiplier was derived or say how much of the 'enabled' figure is attributed to the specific Chevron and Amazon deals versus a modelled industry-wide effect.

If the central claim survives peer scrutiny, the strategic implication is that any hyperscaler pitching a net-positive climate story on the strength of its AI-for-renewables work has to defend the other half of the ledger publicly, and any investor or regulator using data-center power draw as their sole AI-emissions proxy is measuring the wrong thing.