Google’s Emissions Rise Amid AI Sustainability Claims

Big Tech faces accusations of exaggerating AI climate credentials

A March 2026 analysis has ignited a sharp debate about whether technology companies are overstating artificial intelligence’s environmental benefits. TechPolicy.Press published a detailed examination accusing major tech firms of greenwashing, claiming they couple energy-hungry consumer AI tools with older, narrower applications to create an inflated picture of climate progress.

The controversy centers on a specific practice. Companies like Google have promoted research suggesting AI could reduce global greenhouse gas emissions by 5-10% by 2030. That figure matches the entire annual output of the European Union. However, a February 2026 report by Beyond Fossil Fuels traced this statistic back to its origins and found something troubling. The number appears rooted in speculative projections of niche applications rather than verified real-world results.

For UK businesses evaluating their own digital strategies, this matters considerably. Many firms now face procurement requirements asking them to justify the carbon footprint of their technology choices. Meanwhile, they’re being told that AI adoption could deliver significant emissions reductions. Understanding which claims rest on solid evidence becomes essential for making informed decisions about technology investments and reporting credible sustainability progress.

The timing is particularly significant. Data center energy consumption continues rising across the UK and globally. Companies are under increasing pressure to demonstrate genuine environmental progress rather than simply shifting emissions elsewhere in their operations or supply chains.

What technology companies claim about AI and emissions

Google has been particularly vocal about AI’s climate potential. The company cited research in multiple forums, including papers, Fortune Magazine coverage, and an April 2025 EU policy roadmap. The central claim remained consistent: AI applications could mitigate between 5% and 10% of global emissions within this decade.

More recently, Google published detailed figures for its Gemini AI model. According to this data, a median text prompt uses 0.24 watt-hours of energy, produces 0.03 grams of CO₂ equivalent emissions using market-based accounting, and consumes 0.26 milliliters of water. The company compared this to running a television for less than nine seconds.

Over 12 months, Google reported that Gemini’s energy footprint dropped 33 times while carbon emissions fell 44 times through efficiency improvements. The company also highlighted that its data centers operate at 1.09 power usage effectiveness, a metric measuring how efficiently facilities use energy. Additionally, Google’s 2025 reporting claimed its Solar API enabled 6 million metric tonnes of CO₂ reductions in 2024 by helping Americans install rooftop solar panels.

Other companies have released similar figures. Sam Altman stated that ChatGPT queries average 0.34 watt-hours. Mistral, another AI provider, claimed 1.14 grams of CO₂ equivalent per prompt. These numbers vary considerably, making direct comparisons difficult for businesses trying to assess different platforms.

The technology sector has long argued that AI offers solutions for climate challenges. Applications include optimizing flight paths to avoid contrails, managing smart home energy consumption, improving industrial processes, and forecasting renewable energy generation. These use cases sound promising on paper.

Research reveals gaps between claims and evidence

Beyond Fossil Fuels conducted a systematic investigation into the origins of the 5-10% emissions reduction claim. Their February 2026 report traced the figure through various citations and found a concerning pattern. The statistic appears to rest on theoretical scaling of narrow applications rather than documented, large-scale impacts.

Crucially, the research found no evidence that consumer-facing generative AI tools like ChatGPT, Copilot, or Gemini have delivered material, verifiable, and substantial emissions reductions. This distinction matters because these consumer tools now dominate AI usage and energy consumption. The original climate benefit claims derived from different types of AI applications entirely.

TechPolicy.Press expanded on this analysis, identifying what they call “false-coupling.” This practice links resource-light traditional AI applications with energy-intensive generative models under a single narrative. Traditional AI often refers to narrower machine learning tasks like image recognition or recommendation algorithms. Generative AI encompasses large language models that create text, images, or code.

The difference in energy requirements between these categories is substantial. Therefore, combining them in climate accounting obscures the true environmental cost of the tools most businesses and consumers actually use. Independent analyses confirm growing concerns about AI’s footprint. Data centers globally have increased electricity consumption despite becoming more efficient, partly because AI workloads require significantly more computing power than previous internet applications.

Water consumption has risen noticeably. Google reported a 20% increase in water use between 2021 and 2022. Microsoft saw a 34% rise over a similar period. Data centers require substantial water for cooling systems, and AI training and operation intensify these demands. Meanwhile, many facilities still draw power from non-renewable sources, complicating emissions calculations.

Why measurement methods complicate the picture

Google’s Gemini figures rely on specific assumptions that affect how meaningful they are for comparison. These include hardware lifespan estimates, market-based rather than location-based emissions accounting, and calculations covering active compute, idle power, cooling, and embodied carbon from manufacturing equipment.

Market-based accounting lets companies count renewable energy credits they’ve purchased rather than the actual electricity mix powering their facilities at any given moment. Location-based accounting reflects the real grid composition where data centers operate. The difference between these methods can be substantial, particularly in regions with high fossil fuel dependence.

Prompt complexity creates another variable. A simple question uses different resources than a request to generate lengthy code or analyze complex documents. Google’s published figures represent median use, but real-world applications often exceed this baseline. Independent studies have suggested that actual per-query consumption may run higher than company-reported figures, though standardized measurement protocols remain absent.

Greenly, an emissions measurement platform, acknowledged Google’s transparency as a positive step forward while noting the limitations. The absence of standardized metrics across the industry makes it nearly impossible for businesses to compare platforms accurately or report AI-related emissions with confidence.

This measurement challenge extends beyond individual queries. Calculating the full lifecycle impact requires accounting for hardware manufacturing, facility construction, cooling infrastructure, electricity generation sources, water treatment, equipment disposal, and the carbon cost of training models before they ever answer a single query. Training large language models can consume as much electricity as hundreds of UK homes use in a year.

Five essential points for UK businesses

• Major technology companies claim AI could reduce global emissions by 5-10% by 2030, but this figure traces back to speculative projections of niche applications rather than verified evidence from widely-used generative AI tools.
• Recent transparency efforts show a median Gemini prompt uses 0.24 watt-hours and produces 0.03 grams of CO₂ equivalent, though measurement assumptions and prompt complexity variations limit direct platform comparisons.
• Water consumption at major tech companies increased 20-34% recently as AI workloads expanded, adding resource pressure beyond electricity use that businesses must consider in environmental assessments.
• The practice of “false-coupling” combines resource-light traditional AI with energy-intensive generative models in climate narratives, potentially obscuring the true environmental cost of tools businesses actually deploy.
• No standardized measurement protocols currently exist for AI emissions, making it difficult for UK firms to accurately compare platforms, report footprints, or verify supplier sustainability claims in procurement decisions.

Commercial implications for compliance and procurement

UK businesses face growing requirements to report technology-related emissions. PPN 06/21 requires suppliers bidding for central government contracts above £5 million to publish carbon reduction plans. Many buyers now extend this expectation to smaller contracts and private sector tenders. Consequently, firms must account for their digital infrastructure footprint, including AI tools.

The contested state of AI climate claims creates practical difficulties. If a business adopts AI tools based on vendor sustainability assertions, how should they report this in carbon accounting? Using vendor-supplied figures without understanding underlying assumptions risks overstating progress. Conversely, excluding AI entirely from reporting may undercount genuine emissions.

Scope 3 emissions present particular challenges. These indirect emissions from a company’s value chain include purchased cloud services and software. As AI capabilities become embedded in standard business software, from accounting platforms to customer service systems, Scope 3 calculations grow more complex. Businesses cannot simply avoid AI tools, as competitors are adopting them and efficiency pressures mount.

Procurement teams need better information to evaluate competing solutions. Currently, one platform might report emissions per query while another focuses on data center efficiency. A third might emphasize renewable energy purchases. These metrics don’t readily translate into comparable assessments of actual environmental impact. Therefore, buyers struggle to make informed choices even when they want to prioritize sustainability.

The situation also affects tender responses. If your business claims AI adoption as part of your carbon reduction strategy, buyers may now question whether those benefits are real or theoretical. Weak evidence supporting your claims could undermine credibility with procurement panels increasingly sophisticated about greenwashing.

Regulatory pressure appears likely to increase. The absence of standardized AI impact disclosure has prompted calls for mandatory reporting frameworks similar to those governing other environmental impacts. Businesses should anticipate that current measurement gaps will eventually close, potentially revealing that early AI adoption carried higher environmental costs than initially reported.

How this fits into broader sustainability strategy

Smart businesses recognize that technology decisions carry environmental consequences beyond direct operations. However, the current evidence gap around AI makes strategic planning difficult. Some applications genuinely appear to reduce emissions. Optimizing logistics routes, improving building energy management, or forecasting renewable generation can deliver measurable benefits.

The problem emerges when generalized claims about “AI” lump these specific applications together with general-purpose tools. A logistics optimization algorithm running on a single server differs fundamentally from a generative AI platform requiring massive data center infrastructure. Treating them identically in sustainability planning distorts decision-making.

Yale Environment 360 reported research on AI-optimized flight paths that avoid contrails, potentially offsetting all AI-related CO₂ emissions from 2020 if implemented at scale. Similarly, AI-managed smart homes could reduce household carbon by 40%. These examples demonstrate genuine potential. Yet neither involves the generative AI tools most businesses are currently deploying.

For UK SMEs developing carbon reduction plans for compliance purposes, this creates a dilemma. AI might genuinely improve efficiency in specific processes. Nevertheless, blanket claims about AI’s climate benefits lack the robust evidence needed for credible reporting. The solution involves specificity. Rather than claiming “AI reduces our emissions,” document particular applications with measurable outcomes.

Consider energy management systems using machine learning to optimize heating and cooling in warehouses. These deliver quantifiable reductions in electricity consumption. You can measure baseline usage, implement the system, and track actual changes. This approach provides defensible evidence for sustainability reports and tender responses.

Contrast this with adopting generative AI for customer service or content creation. These applications may improve productivity but likely increase your overall technology footprint. Acknowledging this honestly, while separately highlighting genuine efficiency gains elsewhere, builds more credible sustainability narratives than uncritically repeating vendor claims.

Businesses should also consider reputational risk. Greenwashing accusations damage trust with customers, investors, and procurement teams. As scrutiny of AI environmental claims intensifies, companies that relied on unsupported assertions may face awkward questions. Those that maintained rigorous, evidence-based approaches to sustainability reporting will prove better positioned.

Practical steps for assessing AI environmental impact

Until standardized metrics emerge, businesses can take several steps to better understand AI’s role in their environmental footprint. First, separate AI tools into categories. Distinguish between narrow applications delivering specific functions and general-purpose generative models. This clarifies where environmental costs concentrate.

Second, request detailed information from vendors. Ask how they calculate emissions figures, what assumptions underpin their reporting, whether they use market-based or location-based accounting, and how their data centers source electricity. Vendors reluctant to provide specifics may warrant additional scrutiny. Those offering transparent methodologies enable more informed decisions.

Third, consider alternative solutions for tasks that don’t require cutting-edge AI. Sometimes simpler technology delivers similar outcomes with lower resource demands. Generative AI often provides impressive capabilities, but not every business process needs those capabilities to function effectively.

Fourth, integrate AI considerations into existing sustainability frameworks rather than treating them separately. Your carbon reporting processes should account for technology infrastructure systematically. This prevents AI from becoming an overlooked emissions source as adoption grows.

Fifth, monitor developments in measurement standards. Organizations including Greenly have called for cross-sector alignment on science-based footprint calculations. When credible standards emerge, early adopters will gain competitive advantage in procurement processes requiring verified environmental data.

Training also matters. Staff responsible for technology procurement, sustainability reporting, and tender responses need to understand these issues. Professional development in digital sustainability helps teams ask better questions of vendors and avoid accepting unsupported environmental claims.

Finally, maintain perspective about what current evidence actually supports. AI may eventually deliver significant climate benefits through specific applications. However, widespread consumer generative AI tools have not yet demonstrated this at scale. Distinguishing between proven and theoretical benefits strengthens both decision-making and external reporting.

Further reading

Several sources provide ongoing analysis of AI environmental impacts. The Department for Energy Security and Net Zero oversees UK energy policy and may develop AI-specific guidance as part of broader net zero commitments.

Beyond Fossil Fuels maintains detailed analysis of technology sector environmental claims. Their February 2026 report on AI climate benefits provides one of the most thorough investigations into the evidence base for industry assertions.

Academic research continues examining AI energy consumption and potential applications. Yale Environment 360 regularly publishes accessible analysis of environmental technology developments, including both opportunities and challenges around AI deployment.

For businesses navigating these issues while maintaining compliance with procurement requirements and reporting obligations, the key involves balancing innovation against verifiable environmental performance. AI offers genuine capabilities that many firms will need to adopt for competitive reasons. However, sustainability credibility requires distinguishing between marketing claims and demonstrated outcomes, particularly as scrutiny of corporate environmental commitments continues intensifying across UK business and public sectors.