Google’s carbon emissions jumped 25% last year. Amazon’s climbed 16%. The two tech giants released their latest sustainability reports within days of each other this week. Neither pointed a finger at artificial intelligence. Yet the data tells its own story.
AI systems demand vast computing power. That power translates into surging electricity use, water for cooling, and emissions from building and running data centers. The reports expose a widening gap between ambitious net-zero pledges and the practical realities of scaling AI infrastructure.
Both companies have long touted progress on buying renewable energy. Those purchases once offset emissions from offices and smaller facilities with relative ease. AI changed the equation. Energy consumption rose sharply as AI workloads expanded. Scope 3 emissions, which cover supply chains and purchased goods, drove most of the increase. For Google that category more than doubled from its 2019 baseline after a 2.1 million metric ton rise last year. Amazon cited adding more than 1.2 gigawatts of data center capacity in the fourth quarter of 2025 alone.
Tim De Chant laid out the numbers in TechCrunch. “It’s no secret that AI is a hog, consuming energy and water like no digital technology before it,” he wrote. “Now we know just how much Big Tech’s pursuit of AI is costing the environment.” The article notes that neither firm directly blames AI. Indirect signals abound. Pages devoted to AI’s potential environmental benefits read like overcompensation. Carbon intensity metrics appear prominently. These measure pollution per dollar of revenue. They offer one way to frame growth amid rising totals.
Energy purchases themselves stayed relatively clean thanks to years of renewable contracts. That advantage may not hold. Tech firms including Google have turned to natural gas plants to meet immediate AI-driven demand. Construction of new facilities adds another layer. Steel and cement production emit heavily. Chip manufacturing for GPUs relies on grids in Asia still powered largely by fossil fuels. Chemicals used in fabs act as potent greenhouse gases. Their warming effect dwarfs equivalent carbon dioxide volumes.
Broader industry data paints an even starker picture. Global data center electricity consumption hit 415 terawatt-hours in 2024, about 1.5% of world supply. The International Energy Agency projects that figure could reach 945 TWh by 2030, with AI as the dominant force. In the United States alone, data centers could claim between 6.7% and 12% of total electricity by 2028, up from 4.4% in 2023, according to a Lawrence Berkeley National Laboratory analysis cited in a Brookings Institution report from April. Brookings also highlighted forecasts that U.S. data center demand may rise 130% by 2030.
But. These projections carry real consequences for consumers and grids. Residential electricity prices have climbed more than 36% since 2020. Some regions near heavy data center activity saw jumps as high as 267% over five years. Sen. Elizabeth Warren and other Democrats pressed Amazon, Google, Microsoft and Meta on whether companies pass infrastructure costs to ratepayers. A March Consumer Reports piece described one Virginia homeowner’s bill spiking from $100 to $281 in a single month. He blamed nearby AI facilities. “They’re building them like it’s ‘Field of Dreams’—build it and the electricity will come—but we don’t see how that’s going to happen,” the resident told the publication.
Amazon’s own water usage report, released recently, showed its data centers consumed 2.5 billion gallons globally in 2025. The company claimed efficiency seven times better than the industry average at 0.12 liters per kilowatt-hour. Still, the absolute volume underscores cooling demands. A Consumer Reports article from March noted a typical hyperscale center can draw 100 megawatts, enough for 100,000 households. Bloom Energy predicted U.S. data center demand nearly doubling from 80 to 150 gigawatts between 2025 and 2028. That’s equivalent to adding Spain’s entire electricity needs in three years.
Amazon leads current U.S. self-built data center power consumption at roughly 9 gigawatts. Google and Microsoft each sit around 5 gigawatts. A Wall Street Journal article from late June examined how the race for power favors Amazon’s focus on cost and reliability while Google emphasizes clean sources. Both strategies face limits. BloombergNEF forecasts U.S. data center power demand more than doubling to 78 gigawatts by 2035, with average hourly consumption nearly tripling.
Some efficiency gains offer hope. Google’s latest Ironwood TPU delivers 30 times the efficiency of its 2018 version. The company reports a fleet-wide power usage effectiveness of 1.09, among the best in the sector. It also cut energy per median Gemini prompt by a factor of 33 over the past year. Yet overall demand still climbs faster than these improvements. Training frontier models could require 5 gigawatts each by 2027, Anthropic estimated. The firm projected the U.S. AI sector alone needing 50 gigawatts of new capacity by 2028, roughly twice New York City’s peak demand.
Former Google CEO Eric Schmidt told Congress that data centers will need 29 gigawatts more by 2027 and another 67 by 2030. Capital spending reflects the stakes. Hyperscalers poured over $200 billion into AI data centers in 2025. Amazon, Google, Meta and Microsoft alone plan hundreds of billions more. One analysis pegged 2025 spending by those four at $364 billion. PPA prices for renewables rose 35% in 2024 amid this buying spree. Big Tech accounted for 43% of global clean energy purchase agreements that year.
So the tension grows. Net-zero targets look harder to hit. Fossil fuel backups increase. Supply chain emissions from chips and concrete resist easy fixes. Startups pursue low-carbon steel and cement, but scale remains years away. Carbon removal credits will likely fill gaps, at significant cost. The TechCrunch piece concludes that none of these problems are insurmountable. “Amazon, Google, and their peers have their work cut out for them,” it states. “Their embrace of AI hasn’t made it any easier.”
Recent discussions on X echo the pressure. One post noted token costs rising even for Google, forcing fundraising despite past profitability. Another highlighted Google Cloud efforts to cut inference expenses through specialized storage. These moves address economics directly tied to energy and hardware demands. Yet they do not erase the physical constraints. Grids strain. Communities push back on new plants and transmission lines. Policymakers demand transparency on who bears the expense.
Industry leaders once framed AI as an efficiency tool that could reduce overall emissions. The sustainability reports complicate that narrative. AI helps optimize some operations. Its own footprint expands faster. Capital goods and energy-related Scope 3 categories dominate the rises. Data centers and GPUs sit at the center. As one recent Medium analysis framed it, inference costs have dropped dramatically while total energy use doubles. Companies that manage this balance gain advantage. Those that ignore it face higher bills, regulatory scrutiny and reputational damage.
The reports from Google and Amazon serve as early signals. They reveal how quickly AI infrastructure alters long-standing environmental strategies. Power procurement shifts. Emissions categories balloon. Water and land use draw fresh attention. Executives must now weigh model performance against these mounting physical and financial costs. The data leaves little room for doubt. Scaling AI carries a heavier price tag than many forecasts anticipated. And that price keeps growing.
WebProNews is an iEntry Publication