AI's environmental footprint is exploding as the technology moves beyond simple questions to longer, more complex tasks that require significantly more computing power. A new analysis from Vals AI, an independent benchmarking company, found that lengthier tasks requiring AI models to "think" for extended periods, such as building a software application, can have an environmental impact 10,000 times greater than basic queries answered almost instantly.

Why Does Task Complexity Matter So Much for AI's Energy Use?

The shift in how people use AI is fundamentally changing its environmental toll. Users are moving away from treating AI like a search engine for one-off answers and instead having extended conversations with chatbots, asking them to complete complex projects, and relying on AI agents to handle multi-step tasks. Each additional computational step required to complete these longer tasks multiplies the energy demand.

To put this in perspective, Vals AI's research found that building a web app using some AI models consumes energy equivalent to powering a home for 2.5 hours. The firm evaluated 16 models using its proprietary benchmarking index combined with EcoLogits, an open-source tool that estimates AI's environmental impact, measuring carbon emissions, water consumption, and electricity use across real-world tasks.

"Usage is transitioning from just using ChatGPT as a Google Search alternative to actually having conversations with these chatbots. Footprints scale dramatically because these are no longer just single-shot questions; they're a lot more carbon intensive," said Omar Almatov, a founding engineer at Vals.

Omar Almatov, Founding Engineer at Vals AI

Which AI Models Have the Largest Environmental Footprint?

The analysis revealed significant variation in how different AI models impact the environment. Alibaba Group's most powerful model, Qwen3.8 Max, scored highest for energy, carbon, and water intensity across the tasks evaluated, while Ling 3.0 Flash 2607, a model from Ant Group designed for simpler tasks, scored lowest. Notably, 14 of the 16 models analyzed were developed by Chinese companies, with one from US-based Thinking Machines Lab and one from France's Mistral AI.

The researchers focused on open-weight models, whose training parameters are publicly available and can be directly analyzed. Most leading US models, including those from OpenAI and Anthropic, are closed-weight, meaning their underlying parameters remain private and cannot be evaluated using the same method.

A particularly striking finding emerged around performance gains: Kimi K3, the most accurate open-weight model on the Vals Index, has an outsized environmental footprint compared with DeepSeek V4 Flash, which is only marginally less accurate. The report authors describe this difference as equivalent to charging your laptop once versus 20 times, or drinking a glass of water versus flushing a toilet.

How Are Tech Companies Responding to AI's Growing Energy Demands?

Major technology companies are falling significantly short of meeting their climate commitments as AI energy consumption surges. A recent report from the United Nations International Telecommunication Union (ITU) and the World Benchmarking Alliance analyzed data from 200 leading technology firms in 2024 and found a critical gap between corporate climate pledges and actual progress.

The findings paint a sobering picture of the technology sector's environmental impact. Emissions from four prominent AI and cloud providers jumped by as much as 239 percent between 2020 and 2024. Collectively, the 200 assessed companies generated 301 million tonnes of carbon dioxide equivalent, representing 0.8 percent of global energy-related emissions, and consumed approximately 500 terawatt-hours of electricity, or 1.7 percent of worldwide electricity supply.

The top ten electricity consumers alone accounted for 269 terawatt-hours, surpassing Australia's total annual electricity usage. China Mobile led individual consumption at 63 terawatt-hours, followed by Alphabet and Samsung at 32 terawatt-hours each, and Microsoft at 30 terawatt-hours.

What Progress Are Companies Actually Making on Climate Goals?

Despite widespread climate commitments, progress remains uneven across the technology sector. The UN ITU and World Benchmarking Alliance assessment revealed that only 25 of the 200 evaluated companies source 100 percent renewable electricity. While 151 firms have established near-term emission reduction targets, largely influenced by investor pressure and emerging regulations, merely 85 are currently on track to meet their goals.

• Companies Meeting Targets: Only 85 of 200 assessed firms are currently on track to meet their near-term emission reduction goals, despite 151 having established such targets.
• Renewable Energy Adoption: Just 25 companies source 100 percent renewable electricity, while the remaining 175 continue to rely partially on conventional power sources.
• Transition Planning: Only 81 companies have developed comprehensive transition plans to decarbonize their operations and supply chains.

The assessment ranked companies based on their targets, data transparency, and actual performance. Swisscom achieved a perfect score, leading alongside Accenture, Deutsche Telekom, Vodafone, Capgemini, and Telefonica. In contrast, the bottom 30 performers included Huawei, Spotify, Nintendo, Weibo, Xiaomi, Zoom, and Toshiba TEC, while 18 entities, including X and SpaceX, received no score due to insufficient disclosure.

How Can Organizations Better Account for AI's Environmental Impact?

A major barrier to progress is the lack of transparency and standardized measurement across the AI industry. Businesses that use AI tools currently lack a clear way to account for AI's carbon footprint in their own emissions reporting. Some AI firms are now under pressure to disclose more information because of a new California law on climate disclosure coming into effect.

However, AI companies reveal little about their model architecture, the types of hardware they use, and the cooling and electrical infrastructure powering their data centers; information that would make it easier to measure their footprint accurately. American AI companies have disclosed only a handful of estimates for water and electricity consumption. OpenAI Chief Executive Officer Sam Altman wrote in 2025 that an average ChatGPT query uses about 1/15th of a teaspoon of water, and estimates by Altman and Alphabet's Google the same year put the energy toll of the average text query between 0.24 and 0.34 watt hours.

Separate Microsoft research found that long-reasoning and agentic requests can increase energy consumption by more than an order of magnitude, consistent with Vals' analysis. Yet companies have provided far less information about the footprint of these longer, more complex tasks that are becoming increasingly common.

"Policy conversations should be rooted in real evidence and data," stated Rayan Krishnan, Vals co-founder and CEO.

Rayan Krishnan, Co-founder and CEO at Vals AI

Industry leaders emphasize that environmental resilience must be embedded into the digital infrastructure of tomorrow. The UN ITU chief Doreen Bogdan-Martin stressed this necessity, while World Benchmarking Alliance executive director Gerbrand Haverkamp noted the urgent need for firms to decarbonize their extended supply chains. As digital infrastructure continues to expand, the technology sector faces mounting pressure to accelerate renewable adoption and implement rigorous emission controls, or risk undermining global climate objectives.