The promise was made in sunnier times for the climate math. Google committed to running its operations on net-zero carbon by 2030, and Amazon signed the Climate Pledge, promising to reach net-zero emissions by 2040, with its founder calling the target ambitious but achievable. The data centers those companies have built since then are testing whether the promises can survive contact with artificial intelligence.
The electricity demands of AI have grown faster than anyone predicted when the pledges were written. Training large models consumes power on a scale that was previously the preserve of small cities, and running those models for customers, the inference side of the business, is projected to consume more. The result, analysts say, is that the companies’ power consumption is climbing at a pace that their clean-energy plans were never designed to match.
Google’s trajectory illustrates the problem. The company has reported that its emissions have climbed sharply since 2019, the baseline for its net-zero commitment, driven in large part by the data centers that power its AI products. It has bought record volumes of renewable energy and signed contracts for new solar, wind and geothermal projects, but the gap between the power its facilities draw and the clean power it can procure keeps widening.
Amazon faces the same arithmetic with a longer runway and a larger footprint. The company’s cloud unit, AWS, is one of the largest buyers of data center capacity in the world, and its commitments to AI customers have multiplied its power needs. Amazon has said it remains committed to the Climate Pledge, and it has invested in nuclear power and renewable projects, but the pace of new construction is straining the timelines its own pledge implies.
The power problem is compounded by geography. The regions where AI data centers are being built fastest, from Virginia to Texas to the desert Southwest, often have grids that run heavily on fossil fuels, and the companies have had to pair their facilities with new renewable projects that take years to come online. In the meantime, the emissions are counted and reported, and the reports have made for uncomfortable reading.
The carbon accounting has its own complications. Companies can buy renewable energy credits and claim progress on paper, a practice that critics say inflates the appearance of decarbonization without changing the power that actually reaches their facilities. Both Google and Amazon have been pressed by investors and environmental groups to move beyond purchased credits toward power they control, and both have argued that their procurement is real and growing.
The cost of the conflict is not just environmental. Power is now a line item in the AI arms race, and the companies that secure reliable, cheap electricity have an advantage over those that do not. Google and Amazon have responded by buying into nuclear generation, signing long-term power agreements and building data centers next to the sources of energy they need, a strategy that protects their AI ambitions even as it complicates their climate accounting.
The reputational dimension is harder to price. Both companies have marketed their sustainability programs aggressively, to customers, employees and regulators, and every report showing rising emissions gives their critics ammunition. European regulators, in particular, have begun demanding more detailed disclosure of data center energy use, and the companies face the prospect of compliance regimes that treat their power consumption as a liability.
The tension has produced creative responses. The companies have funded carbon capture projects, invested in advanced geothermal, and explored small modular nuclear reactors, technologies that were fringe when the pledges were made. Each bet carries its own risks, and none will deliver at the scale the data centers need within the timelines the pledges require.
Utility companies have become the unlikely beneficiaries of the conflict. Power providers in data center hubs have reported surging demand and rising profits, and their stocks have outperformed as investors priced in the AI build-out’s appetite for electricity. The companies buying that power may be struggling with their climate math, but the utilities selling it are not complaining.
Investors are divided on what the trade-off means. Some argue that clean power spending is a necessary cost of the AI build-out, and that the companies’ commitments demonstrate the discipline to manage it. Others say the rising emissions figures undercut the sustainability narrative that has supported their valuations, and that the companies will eventually have to choose between their climate promises and their computing ambitions.
TechCrunch’s analysis, which brought the conflict into focus this week, points to a third possibility: that the pledges are being redefined rather than abandoned. Companies can adjust baselines, change accounting methods or reinterpret what net-zero means, and regulators have shown limited appetite for punishing failures that were, in fairness, made before the AI boom’s power needs were visible.
Neither Google nor Amazon has said it is walking away from its commitment, and both continue to announce clean-energy deals. But the direction of the numbers is hard to argue with: the power is rising, the clean supply is not keeping pace, and the distance between promise and performance is growing. The 2030 and 2040 targets were set in a different computing era, and the AI era is rewriting the timeline.
