Sommaire
Artificial intelligence is turbocharging the growth of Google and Amazon, and their carbon footprints along with it.
As both companies race to build and power the data centers needed to train and run AI models, their reported CO2 emissions are climbing, according to reporting cited byLa Presse. The surge is putting real pressure on Big Tech’s glossy “net-zero” promises, because electricity demand is rising faster than power grids are getting cleaner.
The problem isn’t hard to explain: AI takes a lot of computing power. More computing means more electricity. And in many places, more electricity still means more fossil fuels.
Google’s AI boom is colliding with its climate goals
Google publishes an annual environmental report that tracks both its direct emissions and the much larger indirect ones, especially emissions tied to purchased electricity and its supply chain. Those categories matter because AI is changing the company’s energy math in a big way.
Training large AI models requires massive clusters of specialized chips, often GPUs, running around the clock for long stretches. Then comes “inference,” the everyday work of serving AI responses to millions of users, which demands always-on capacity and redundancy. The result: Google has to expand and pack more computing gear into data centers, buy more hardware, and lock in more electricity.
That creates a timing problem. Google’s long-term climate strategy leans heavily on renewable energy purchases and long-term contracts. But if a data center is plugged into a grid still powered partly by natural gas or coal, ramping up computing loads can still push real-world CO2 emissions higher, even if the company can claim cleaner energy on paper through contracts and accounting rules.
And the footprint goes beyond the electric meter. Manufacturing servers, constructing buildings, and producing materials like concrete and steel all add emissions, often counted in “scope 3,” the supply-chain category. If AI accelerates hardware replacement cycles, those upstream emissions rise too. Google’s challenge is now twofold: clean up the electricity feeding its data centers and reduce the emissions embedded in the equipment itself, without slowing the AI rollout that’s become central to its business.
Amazon’s emissions pressures are tied to AWS and relentless scale
Amazon’s climate equation is even more complicated because it spans everything from delivery vans to warehouses to cloud computing. But the AI-driven emissions growth is heavily concentrated in Amazon Web Services, the cloud division that hosts and trains AI models for Amazon and for outside customers.
As AI workloads expand, AWS needs more compute and more storage, meaning more servers, more internal networking gear, and more data centers. Even when each server gets more efficient, Amazon can still lose the emissions battle through sheer volume: demand grows so fast it can overwhelm efficiency gains.
This is the rebound effect in action. Make AI tools easier and cheaper to use, and people use them more, chatbots, content generation, predictive analytics, driving up total electricity consumption.
Where those data centers are built also matters. Companies tend to locate them where power is reliable, fiber connections are strong, and land is available. But the carbon intensity of the local grid varies widely. If the grid is still heavy on fossil fuels, every additional megawatt-hour increases emissions. Amazon can sign renewable energy contracts, but matching clean generation to 24/7 consumption, hour by hour, is still a thorny technical and political challenge because the grid is shared and electricity demand never stops.
Data center electricity demand is rising faster than grids are decarbonizing
Google and Amazon share the same core constraint: data centers are guzzling more electricity as AI expands. Compared with traditional web services, AI requires sustained computing power and more robust cooling.
Operators track efficiency using metrics like PUE (power usage effectiveness), which compares a facility’s total energy use to the energy used by IT equipment. Best-in-class sites can approach about 1.1, but squeezing out additional gains gets harder as facilities near physical limits.
The bottleneck isn’t only climate-related, it’s industrial. Utilities can’t always upgrade transformers, substations, transmission lines, and interconnections as fast as new data centers come online. In parts of North America and Europe, grid hookup timelines are stretching out, pushing some operators toward stopgap solutions that can be more polluting, like on-site generators or power arrangements tied to dirtier mixes.
Cooling adds another layer. Data centers consume electricity and, depending on the system, significant water. Air cooling, water cooling, and immersion cooling each come with tradeoffs based on local climate and resources. AI’s dense computing loads generate more heat, raising the stakes, and sometimes wiping out part of the gains from better cooling tech simply because the total amount of computing keeps climbing.
That growth is forcing policymakers into uncomfortable choices. Data centers bring investment and jobs, but they also claim large chunks of the power supply, sometimes competing with other needs. The debate is increasingly about how much grid capacity should be dedicated to the digital economy, and what performance standards and emissions transparency should be required in return.
Net-zero claims face scrutiny: renewable contracts, offsets, and the transparency gap
To counter rising emissions, major tech companies typically point to three levers: buying renewable electricity through long-term power purchase agreements, improving efficiency and shifting computing to times or places with cleaner power, and offsetting remaining emissions with carbon credits.
But each approach is under a brighter spotlight. Renewable contracts can improve a company’s reported footprint, yet they don’t always guarantee that the electricity powering a specific data center at a specific hour is actually carbon-free. Carbon credits raise their own questions, quality, permanence, and whether the reductions are truly “additional.” A forest project can burn. A land-use plan can change. And some credits have been criticized as delivering little real-world impact.
That’s why transparency is becoming the new battleground. Critics and researchers want more granular reporting, hour-by-hour data showing whether data center consumption is matched with low-carbon generation. They also want clearer disclosure of supply-chain emissions tied to server manufacturing, raw material extraction, and construction. Corporate reporting is improving, but methodologies still vary widely, making apples-to-apples comparisons difficult.
The bigger question now reaches beyond accounting: if ever-more-powerful AI models become the default, how do companies keep energy demand from spiraling without choking off innovation? Options include smaller specialized models, better software optimization, more efficient chips, shared infrastructure, and shifting workloads to regions with cleaner electricity. But the gap between AI’s commercial acceleration and the slower pace of grid decarbonization helps explain why Google and Amazon can see emissions rise even while publicly committing to net-zero targets.
