Microsoft Cut Carbon Removal 80% as Its AI Capex Heads to $175 Billion

Microsoft is on track to spend about $175 billion on AI infrastructure this year. It also just bought about 80% fewer carbon removal credits than it did last year, in its first retreat in three years from the market it dominates. The company says its decarbonization "ambition" is unchanged. Those three facts are the same story, and the contradiction between them is the news.

The 80% drop shows up in a BloombergNEF market report cited by Gizmodo and independently confirmed by the Seattle Times. Microsoft, the single largest corporate buyer of engineered removals (the kind of tech-based capture projects, as distinct from forest- and soil-based credits), has effectively been the demand signal for the entire young industry. A pullback at that scale is not a procurement footnote. It is the market's largest customer reallocating its dollars.

The reallocation lines up with the AI buildout. In the most recent quarter, Microsoft spent about $41 billion in capex, with the majority going to AI initiatives and data center construction, and guided to roughly $175 billion for the full year, per the company's latest earnings report. That buildout is what is driving the company's emissions up. Microsoft told Bloomberg that its roughly 25% jump in company emissions last year is tied to AI, and new data centers tend to be heavy, around-the-clock loads. The cheapest way to feed them in many U.S. regions is still natural gas turbines, per the same reporting.

Microsoft's response to the credit pullback, relayed through Bloomberg and confirmed by ESG Dive, is that carbon removal is not the only decarbonization strategy and that the move does not signal a shift in the company's stated "ambition." The rebuttal is the test of the thesis, and it is worth taking seriously. Microsoft's climate plan does not live or die on its carbon removal line. The company also buys nature-based credits, funds internal efficiency programs, and has signed multi-year advance purchase contracts with removal startups. Cutting the removal line does not, on its own, prove the climate plan is failing. What it shows is where incremental climate dollars stopped going this year.

The harder question is whether the AI buildout is itself compatible with the climate math. A Nature study published last week, as summarized in the Gizmodo report, found that the climate-optimization benefits AI can deliver to renewables and the grid are dwarfed by the benefits the same technology delivers to the fossil fuel industry. Oil and gas executives have said, in industry coverage, that they are eager to adopt AI specifically to make extraction cheaper. Microsoft is not the only company that uses AI, and the Nature finding is not about Microsoft specifically, but the pattern is the relevant one. AI helps the clean transition and the dirty one, and without policy steering, the dirty application tends to win on unit economics.

Latitude Media has argued for some time that the carbon removal market is, in practice, a single-buyer market with Microsoft at its center. The BloombergNEF report is the first real test of that dependence. The market held; Microsoft is the one that pulled. Carbon removal suppliers priced their contracts against the assumption that Microsoft's demand curve was the floor. The 80% pullback does not cancel that floor, but it bends it.

The contradiction is not that Microsoft is lying about its climate plan. The contradiction is that capital allocation is the operational definition of a corporate climate commitment, and this year's capital is going to AI infrastructure. The next read on whether the gap widens or narrows is the company's full-year sustainability disclosure, due early next year. If AI capex is held flat or trimmed, the 2025 carbon numbers and the 2026 capex pattern start to look like a one-year stress test. If AI capex keeps climbing toward or past the $175 billion guidance and the 80% removal pullback holds, the "ambition" line becomes harder to defend as anything other than a slogan.