Amazon met its renewable-energy target again in 2025.

The company said it matched 100% of the electricity used across its global operations with renewable energy for the third consecutive year. Its electric delivery fleet surpassed 52,700 vehicles, while the number of packages delivered by electric vehicles reached 2.4 billion. Carbon emissions per delivery declined 7% from the previous year.

Yet Amazon’s total carbon footprint increased.

According to Amazon’s 2025 Sustainability Report, the company’s absolute carbon emissions reached 80.85 million metric tons of carbon dioxide equivalent, up 16% from 69.55 million metric tons in 2024. Carbon intensity, measured as emissions per dollar of revenue, also increased 3%.

The figures expose one of the central contradictions facing Big Tech in the AI era.

Artificial intelligence can help companies become more efficient. It can optimize delivery routes, reduce warehouse inventory, detect energy waste in buildings, improve packaging decisions and automate maintenance.

But AI is also an energy-intensive industrial system.

Training and operating generative AI models requires massive data centers, high-performance chips, dense server racks, advanced cooling systems, networking infrastructure and a growing supply of electricity. Every chatbot query, AI agent workflow, image-generation task and enterprise-model deployment ultimately runs on physical infrastructure powered by energy.

For Amazon, the collision between AI expansion and climate commitments is becoming increasingly visible.

The company has pledged to reach net-zero carbon emissions across its global operations by 2040. But as AWS expands its cloud and AI capacity, Amazon must now demonstrate that efficiency improvements and renewable-energy procurement can keep pace with the absolute growth of its business.

The Climate Paradox of AI

AI is often presented as a climate solution.

It can help optimize logistics networks, identify energy waste, reduce unnecessary transportation, forecast demand, improve inventory management and detect equipment failures before they become costly.

Amazon says it is already using AI and machine-learning tools across hundreds of buildings to identify water leaks, heating and cooling problems, and operational-efficiency opportunities.

In that sense, AI can make business operations cleaner.

But AI also makes the underlying infrastructure larger.

Generative AI models require far more computing power than many traditional software applications. They depend on GPUs, specialized AI chips, large storage systems, high-speed networks and increasingly complex cooling equipment.

The more companies adopt AI, the more data-center capacity they need.

Amazon’s report showed that emissions associated with purchased electricity increased 34% in 2025. The company attributed the rise to growing electricity consumption in data centers, the electrification of its delivery network and the electrification of buildings.

AWS alone added more than 1.2 gigawatts of data-center capacity in the fourth quarter of 2025, according to Amazon.

That figure captures the scale of the challenge.

AI may be software from the user’s perspective. But from an infrastructure perspective, it is becoming a heavy industrial sector. It consumes electricity, water, land, semiconductors, construction materials and transmission capacity.

The global AI race is therefore also becoming a race for energy.

Across the United States, Europe, India and the Middle East, companies and governments are competing for data-center sites, grid connections, renewable-energy projects, nuclear-power agreements and new transmission infrastructure.

The question is no longer only which company will build the best model.

It is also which company can secure the energy required to run it.

Efficiency Is Improving — but Demand Is Growing Faster

Amazon argues that it is addressing the problem through efficiency and cleaner power.

AWS reported an average Power Usage Effectiveness, or PUE, of 1.14 for its data centers in 2025. PUE measures how much energy a data center consumes beyond the computing equipment itself, including cooling, lighting and power conversion.

The closer the figure is to 1.0, the more efficient the facility.

Amazon says its data centers are more efficient than typical enterprise-operated facilities and more efficient than the broader public-cloud industry average.

The company also highlights improvements in AI hardware.

Amazon says its Trainium3 AI chip can generate more than five times as many AI responses per megawatt of power as the previous generation and improves performance per watt by four times.

This is an important strategy.

Instead of trying to reduce AI demand, Amazon is trying to make each unit of AI computing more efficient.

But efficiency alone may not solve the emissions problem.

When computing becomes cheaper and faster, organizations may use more of it. As AI tools become more efficient, they may also become easier to deploy across more products, services and workflows.

Economists describe this as the rebound effect.

A technology becomes more efficient. The cost of using it falls. Demand increases. Total consumption may rise rather than fall.

Amazon’s 2025 figures illustrate that possibility.

Emissions per delivery fell.

Data-center efficiency improved.

Renewable-energy matching reached 100%.

But total emissions still rose because the scale of cloud computing, AI services and e-commerce operations grew even faster.

Renewable-Energy Matching Is Not the Same as Real-Time Clean Power

Amazon’s renewable-energy achievement is significant, but it also highlights a distinction that is becoming more important in the AI era.

Matching 100% of electricity consumption with renewable energy on an annual basis means that Amazon has procured or contracted for renewable generation equal to the electricity it uses across the year.

But that does not necessarily mean that every Amazon data center is operating on carbon-free electricity at every hour of every day.

A data center may consume power at night, during periods of low wind or during times when solar generation is unavailable. The electricity physically delivered through the local grid may still include power generated from fossil fuels.

This is why the next stage of corporate climate accountability is likely to move beyond annual renewable-energy matching.

The more difficult question will be:

When and where was the electricity used?

Amazon has indicated that it will begin separately disclosing the share of carbon-free energy used in its operations.

That could signal an important shift in how AI companies are evaluated.

The standard may no longer be only how many renewable-energy contracts a company has signed. It may increasingly become whether the company can operate data centers with carbon-free electricity in the same locations and at the same times that computing demand occurs.

For AI companies, the future climate test may be hourly and regional rather than annual and global.

Nuclear Power Is Becoming Part of the AI Energy Strategy

Amazon is responding to rising power demand not only through wind and solar projects, but also through nuclear energy.

The company has announced investments and agreements connected to nuclear power, including plans tied to new nuclear-generation capacity and long-term procurement from existing plants.

The logic is clear.

Solar and wind are essential for decarbonization, but they are variable. AI data centers need stable, continuous power. Nuclear energy offers a source of low-carbon electricity that can operate around the clock.

For Amazon and other large cloud companies, the future energy mix may increasingly involve a combination of renewables, batteries, transmission upgrades, grid-scale storage, natural-gas backup and nuclear generation.

This makes AI decarbonization more complex than earlier corporate sustainability efforts.

It is no longer a question of installing a few solar panels or purchasing a fleet of electric delivery vehicles.

It is a question of how quickly data centers can be expanded, how efficiently AI models can run, how electricity grids can be strengthened and whether business growth can be separated from rising environmental pressure.

Carbon Intensity Has Fallen Since 2019 — but Absolute Emissions Still Matter

Amazon argues that its long-term progress should not be judged only by the 2025 increase.

The company says its revenue has grown 156% since 2019 while carbon intensity has declined 38% over the same period.

That means Amazon is producing less carbon per dollar of revenue than it did several years ago.

This matters. Efficiency gains can reduce the environmental cost of economic activity.

But climate change is driven by total greenhouse-gas emissions, not only emissions intensity.

A company can become more efficient while still producing more overall emissions if it grows fast enough.

That is the difficult reality facing Amazon.

Its business spans global logistics, cloud computing, data centers, consumer devices, entertainment, content production, retail operations and AI services. Each part of that system may become more efficient over time. But expansion across all of them can still increase the company’s total climate impact.

The challenge is especially difficult because Amazon is not only responding to AI demand. It is helping create it.

By expanding AWS infrastructure, offering AI chips, providing managed foundation-model services and making AI tools easier for companies to deploy, Amazon is increasing the capacity of the broader AI economy.

That may generate revenue and innovation.

It may also increase the electricity demand that makes net-zero targets harder to achieve.

The Real Test Is Whether AI Can Become Cleaner Faster Than It Becomes Bigger

Amazon’s sustainability report is both a progress report and a warning.

It shows that the company is adding renewable energy, expanding electric delivery fleets, improving data-center efficiency and using AI to reduce waste.

But it also shows that those gains are occurring inside a business model that is rapidly increasing its energy demand.

AI can be a tool for sustainability.

It can reduce unnecessary travel, improve industrial efficiency, optimize energy use and help companies identify emissions they previously could not see.

But AI can also become a new source of environmental strain if its infrastructure expands faster than clean energy and efficiency improvements can keep up.

The central question is therefore not whether AI is good or bad for the climate.

It is whether AI can become cleaner faster than it becomes bigger.

Amazon’s 2025 results do not provide a final answer.

They show a company moving in two directions at the same time.

It is building more renewable-energy capacity.

It is deploying more electric vehicles.

It is making data centers more efficient.

And it is expanding AI and cloud infrastructure at a pace that is pushing total emissions upward.

That is the climate paradox of the AI era.

The technology may help solve environmental problems.

But the infrastructure required to run it may become part of the problem first.