MOUNTAIN VIEW, Calif. — The servers that run Google’s artificial-intelligence models are getting more efficient. The factories that build them, and the data centers that house them, are not. Google’s eleventh annual environmental report shows total emissions rose 18% in 2025, with supply-chain emissions — the concrete, steel, servers, and chips purchased from thousands of suppliers — up 25%, according to data cited by ESG Dive and Impakter. Emissions from Google’s own operations fell 2% over the same period, a split that tells the story of the company’s AI strategy in one sentence.
The gains in efficiency are real. Google has spent a decade squeezing more computing out of every watt, and its newest AI-optimized data centers use dramatically less energy per calculation than the generation they replace. The company signed record volumes of renewable power purchase agreements last year and has pushed its suppliers to do the same. But the scale of the AI build-out is overwhelming those improvements: the number of data centers under construction, the volume of servers being manufactured, and the energy embodied in both are growing faster than efficiency gains can offset.
The report attributes the increase directly to AI infrastructure. Data-center construction, server manufacturing, and the cooling systems that keep AI chips from overheating account for the bulk of the supply-chain growth, and the company says those categories will keep expanding as long as demand for AI computing continues to climb. Google is building new data centers at a pace unmatched in its history, and each one carries an emissions footprint that is mostly realized before the first model is trained.
The pattern is not unique to Google. Amazon reported a similar dynamic in its own sustainability disclosures, and Microsoft has acknowledged that its AI investments are making its climate targets harder to hit. The entire industry faces the same contradiction: the products driving revenue growth require physical infrastructure whose construction emissions count against the same budgets the companies have pledged to reduce. Google’s pledge to reach net-zero emissions across its operations and value chain by 2030 — one of the most ambitious in the industry — was made in 2020, before the AI build-out began in earnest.
The numbers show how far the company has to go. Emissions have climbed sharply since 2019, when Google set its 2019 baseline, and the trajectory has steepened as AI workloads have grown. Google’s own operations — the data centers it runs, the offices it occupies — are becoming cleaner, and the 2% decline in operational emissions is the payoff of years of efficiency work. But operational emissions are the smaller half of the ledger; the supply chain accounts for the majority of the company’s footprint, and that is the half that is growing.
The report has become a focal point for critics who argue that the AI boom is incompatible with the industry’s climate promises. Environmental groups have pointed to water consumption in data-center regions, to gas plants built to power new campuses, and to the carbon cost of manufacturing the chips that AI depends on. Google counters that AI itself will help solve climate problems — optimizing grids, improving weather prediction, and accelerating materials science — and that its data centers are among the most efficient ever built.
The accounting question is where the two sides part ways. Google counts emissions where they occur, which means the electricity generated by a gas plant supplying a new data center shows up in the company’s numbers even if the plant is owned by a utility. Suppliers are being asked to decarbonize, and Google says its purchasing power is accelerating that transition, but the transition takes years and the construction is happening now.
The emissions are counted on Google’s ledger even when they occur elsewhere. The company’s supply-chain footprint includes the electricity used by its contract manufacturers, the materials that go into its servers, and the construction of the data centers it builds with partners — emissions Google can influence but not directly control. The 25% rise in that category reflects the sheer volume of building: more fabs, more server plants, more data centers than at any point in the company’s history.
Google is not alone in the bind. Amazon’s disclosures show a similar pattern, and Microsoft has said its AI investments complicate its climate goals, and the industry’s response has been to buy more renewable power and push suppliers to do the same. Investors have so far treated the reports as a cost of doing business rather than a reason to question AI spending, but the trend line is being watched: if emissions keep rising at this pace, the companies’ climate targets will need to change, and that will be a reputational event for an industry that has made sustainability a selling point.
For investors, the report is a footnote to a much larger story about AI spending. Google’s capital expenditures are at record levels and rising, and the environmental cost is a byproduct of that spending. The company has not said whether the 2030 net-zero target is at risk, and it continues to describe the goal as achievable. But the trend line in this year’s report suggests that between the pledge and the data centers under construction, one of them will have to give.
