AI Energy Use Soars, But Gemini Apps' Energy Footprint Stays Surprisingly Low

Google's software efficiency efforts have slashed the carbon footprint of a median Gemini Apps text prompt by 44x in just one year, yet the company's overall carbon emissions still surged by 18% in 2025. While individual AI interactions become greener, the sheer volume of AI deployment and related infrastructure expansion drives a net increase in environmental burden. The rising demand for AI services, from generative models to predictive analytics, pushes energy consumption higher across data centers and their extensive supply chains.

The carbon footprint of individual AI interactions is dramatically decreasing, but Google's aggregate carbon emissions are increasing. Micro-level improvements are being overwhelmed by macro-level growth, exposing a critical challenge for sustainability efforts. Companies face pressure to demonstrate environmental responsibility, often emphasizing per-unit efficiency gains.

Companies are likely to continue prioritizing per-unit efficiency while the aggregate environmental impact of AI grows unchecked, making true sustainability harder to achieve. The strategic focus on per-unit efficiency obscures the broader environmental costs associated with the accelerating buildout of AI infrastructure and its global supply chain, creating an incomplete picture of the industry's true carbon footprint.

Key Statistics on AI's Environmental Impact

• 44x — The carbon footprint of a median Gemini Apps text prompt dropped over one year due to efficiency efforts, according to Arxiv.
• 18% — Google's overall carbon emissions increased year-over-year in 2025, as reported by ESG Dive.
• 25% — Google's Scope 3 emissions, which cover its supply chain and indirect impacts, increased in 2025 from the prior year, according to ESG Dive.
• 3% — Google's Scope 2 emissions, stemming from purchased electricity, decreased last year, according to ESG Dive.
• 65% — Google used an average of approximately 65% carbon-free energy across its operations in 2025, according to ESG Dive.
• 0.03 grams — The median Gemini Apps text prompt emits 0.03 grams of carbon dioxide equivalent (gCO2e), according to Cloud.
• 33x — The energy impact of the median Gemini Apps text prompt dropped over a recent 12-month period, according to Cloud.

The Granular Cost of AI

The median Gemini Apps text prompt consumes 0.24 Wh of energy in a large-scale, AI production environment, illustrating the specific resource requirements of individual AI operations. This metric, alongside others, provides a detailed view of the per-unit environmental cost companies strive to minimize. While these granular metrics confirm significant per-unit efficiency improvements, showcasing technological progress in reducing the individual environmental cost of AI interactions, their aggregate impact remains a critical concern as overall energy demands continue to climb.

Note: Data pertains to 2025 unless otherwise specified. Specific figures are reported by named sources.

Why Are AI Emissions Rising?

Google's overall carbon emissions increased by 18% in 2025, despite significant per-prompt efficiency gains, primarily due to the unchecked scaling of AI infrastructure. The exponential growth in demand for AI services necessitates building and operating more data centers, which require substantial energy. This expansion occurs even as individual AI models become more efficient, proving that the sheer volume of new AI deployments and increased usage can easily outpace any per-unit reductions. The focus on individual prompt efficiency, while technologically impressive, inadvertently masked the larger environmental burden created by the industry’s rapid expansion.

A significant factor in this aggregate increase is the 25% surge in Google's Scope 3 emissions in 2025. Scope 3 emissions encompass indirect emissions from a company's value chain, including manufacturing servers, networking equipment, and construction materials for new data centers. The environmental burden of AI's rapid expansion is increasingly pushed into complex supply chains, challenging the industry's ability to achieve genuine net-zero targets without radical transparency and control over its entire value chain. Companies frequently find it easier to manage and reduce direct operational emissions (Scope 1 and 2) than to exert influence over the extensive and often opaque supply chains that support their growth.

Even with Google using an average of approximately 65% carbon-free energy across its operations in 2025, the 18% increase in overall emissions suggests that current clean energy procurement strategies are insufficient to keep pace with the exponential energy demands of AI at scale. While sourcing renewable energy is a critical step, the sheer volume of new energy required by expanding AI operations means a substantial portion of new demand still meets grid electricity that includes fossil fuels. AI growth outstrips the rate at which clean energy infrastructure can be deployed to support it, confirming a fundamental challenge.

The slight 3% decrease in Google's Scope 2 emissions last year, representing direct energy purchases, further exposes this imbalance. While the company made progress in decarbonizing its direct energy consumption, the dramatic rise in Scope 3 emissions indicates a strategic shift in the environmental impact. The environmental cost is not disappearing; it is simply being displaced further up the supply chain, making it more challenging to track, quantify, and ultimately mitigate. This displacement allows companies to report improvements in direct emissions while the total environmental footprint expands.

Who Is Affected by Rising AI Energy Consumption?

The escalating aggregate carbon footprint of AI, despite per-prompt efficiencies, directly affects environmental advocacy groups and global climate targets. Organizations dedicated to monitoring and reducing global emissions find their efforts complicated by AI's obscured true impact. When major technology companies emphasize individual efficiency metrics while their overall emissions climb, it creates a disconnect that makes comprehensive environmental accounting difficult. Stakeholders seeking transparent, holistic environmental impact assessments are left with an incomplete picture, hindering effective policy-making and corporate accountability.

Regulators and policymakers also face challenges in addressing AI's environmental impact. The complexity of tracking Scope 3 emissions, often diffuse and embedded within global supply chains, makes it difficult to implement effective regulations or incentives for sustainable AI development. Without clear, comprehensive data on the full lifecycle emissions of AI infrastructure, governments may struggle to craft policies that genuinely mitigate the environmental burden, rather than merely shifting it to less transparent parts of the value chain. This lack of clarity can delay necessary interventions and allow the environmental impact to grow unchecked.

Furthermore, communities hosting data centers experience increased localized environmental stress. The significant energy demands of AI data centers necessitate robust grid infrastructure, often leading to increased reliance on existing, sometimes carbon-intensive, power sources in specific regions. As Data Center Frontier reports, AI data centers are becoming integral parts of regional energy systems, placing new strains on local resources. This localized impact, while not always reflected in global aggregate emissions reports, directly affects local air quality, water resources, and energy security, creating tangible consequences for residents near these expanding facilities.

What's Next for AI's Carbon Footprint?

The industry faces increasing pressure to adopt more comprehensive reporting standards beyond per-unit efficiency metrics.

• Companies touting individual AI efficiency metrics inadvertently mask the true environmental cost of the AI boom; Google's 18% overall emissions increase proves that scaling AI usage can easily outpace even dramatic per-prompt reductions.
• The 25% surge in Google's Scope 3 emissions, reported by ESG Dive, indicates that the environmental burden of AI's rapid expansion is increasingly pushed into complex supply chains, challenging the industry's ability to achieve genuine net-zero targets without radical transparency and control over its entire value chain.
• The continuous demand for more powerful AI models, which require greater computational resources, will likely sustain the upward trend in aggregate energy consumption. irrespective of per-prompt improvements.

These trends suggest that future sustainability efforts must move beyond optimizing individual AI interactions to address the systemic growth in infrastructure and supply chain emissions. This will require a fundamental re-evaluation of how AI's environmental impact is measured and reported, potentially leading to new regulatory frameworks that demand full lifecycle assessments. Companies may need to invest significantly in greening their entire supply chains and developing truly energy-efficient AI architectures that can scale without proportional increases in carbon footprint. The current trajectory indicates that without these broader changes, the environmental cost of the AI boom will continue to escalate.