Generative AI and Digital Advertising: How to Measure the New Carbon Shock

The carbon footprint of generative AI in digital advertising has become an unavoidable issue in 2026. While AI optimizes campaigns and reduces media waste, it simultaneously increases the carbon intensity of the advertising supply chain — a paradox that current measurement frameworks still struggle to capture.

Generative AI in 2026: Mass Adoption, Energy Imbalance

In just a few years, generative AI has moved from research labs to mainstream culture: chatbots, AI-enhanced search engines, image generators, video creation, avatars, and voice synthesis. By 2026, it has become a mass-market technology — as routine as a web search or scrolling through a social feed.

This shift comes with an unprecedented energy imbalance:

• AI queries vs. traditional search: a single interaction with a generative model consumes roughly an order of magnitude more electricity than a standard web search — approximately 3 Wh compared to 0.3 Wh — according to analyses cited by the International Energy Agency (IEA) and multiple academic studies.
• Surging data center demand: the IEA projects that global data center electricity consumption could rise from approximately 460 TWh in 2022 to a range of 620 to 1,050 TWh by 2026, depending on the scenario — with the upper bound equivalent to the annual electricity consumption of a country like Japan.
• AI as the primary growth driver: servers equipped with GPUs and specialized AI chips are seeing consumption growth of around 30% per year, compared to 9% for traditional servers. AI could account for nearly half of all data center power consumption by the end of 2026.

Meanwhile, GPU production is accelerating — NVIDIA shipped several million AI-dedicated chips in 2025 alone — and the manufacturing emissions associated with these components are on an exponential trajectory.

Generative AI is no longer just optimizing campaigns or producing creative assets. It is reshaping the energy geography of the digital economy — and digital advertising, with its billions of daily bid requests and complex technical supply chain, sits squarely at the front line.

How AI Is Reconfiguring the Carbon Footprint of Digital Advertising

AI is now embedded across every layer of the advertising supply chain. Each integration delivers value — but it also adds compute, and therefore emissions.

Bidding and Optimization: Every Auction Is Now More Energy-Intensive

Demand-side platforms (DSPs) and supply-side platforms (SSPs) have always relied on optimization algorithms. But the current generation is shifting toward significantly heavier AI models:

• Conversion prediction, LTV modeling, churn detection, brand safety, and fraud prevention — often built on deep learning architectures.
• Near-real-time optimization of bids, budgets, and frequency capping across billions of impressions.
• Fine-grained inventory contextualization through semantic analysis of pages, video content, and social feeds.

A programmatic auction in 2026 has a fundamentally different energy profile than one in 2018. Where the industry once relied on traditional statistical computation running on CPUs, today’s pipelines are AI-intensive — sometimes running on GPUs or specialized ASICs.

The takeaway: the carbon footprint of an impression is no longer determined solely by creative file storage and delivery to the end device. It now also depends on the invisible compute that decides whether, when, and to whom that impression is served.

For more on integrating carbon signals into bid decisions, see our article on AI-driven carbon-aware bidding.

AI-Generated Creative: From Single Assets to Infinite Iteration

Advertising creative is undergoing its own revolution:

• AI-generated images and video (Midjourney, DALL·E, Sora, and others).
• LLM-generated copy, scripts, and headlines.
• Dynamic personalization on the fly: hundreds or even thousands of creative variants generated and tested automatically across micro-segments.

In energy terms, the orders of magnitude differ considerably depending on the type of content generated:

• Generating an image with AI is several dozen times more energy-intensive than generating text.
• Generating a few seconds of video can consume on the order of 100 Wh — equivalent to several dozen minutes of laptop use.

The issue is not any single prompt in isolation. It is the usage model: the industry is moving from a handful of pre-produced assets to continuous iteration loops — generate, test, regenerate, re-test — at massive scale.

AI-Powered Search and Conversational Assistants: Advertising Follows the Compute

With the arrival of AI Overviews, SGE-style experiences, and conversational assistants that integrate advertising, search is undergoing a deep transformation:

• A growing share of queries receives a generated response — sometimes enriched with images, tables, and comparisons.
• Advertising integrates into these responses through sponsored blocks, contextual commercial links, or conversational recommendations.

The model is shifting from “ten blue links plus a few text ads” to one where every query potentially passes through an LLM — and therefore through intensive computation. Even if raw query volume were to plateau, the carbon cost per monetized query is very likely to increase.

The AI Paradox: Less Media Waste, but a New Source of Emissions

Given this landscape, it would be tempting to oversimplify: AI = inevitably worse for the climate. The reality is more nuanced.

Where AI Genuinely Reduces Emissions

When deployed intelligently, AI is a powerful lever for reducing advertising emissions:

• Less media waste: better qualification of contexts, audiences, and timing — resulting in fewer wasted impressions.
• Multi-objective optimization: instead of optimizing solely for business performance, carbon signals (devices, geographies, formats, bid paths) can be integrated into the objective function.
• Rationalized ad pressure: limiting over-frequency and unmanaged “always-on” campaigns.

This is precisely Carbon Intelligence’s mission: leveraging data and AI to reduce the volume of emissions per euro of media spend, by steering toward the lowest-carbon inventory configurations.

Where AI Creates a Carbon Blind Spot

At the same time, AI introduces a new emissions category that is largely invisible in current dashboards:

• Emissions from optimization compute (bidding, targeting, orchestration) on the AdTech side.
• Emissions from AI-generated creative (text, image, video, dynamic personalization).
• Emissions from AI overlays on platforms (generative search, AI-powered ad assistants, generated recommendations).

Today, this component is rarely measured and even less frequently attributed to individual campaigns. An advertiser may believe they have “greened” their advertising by reducing impressions or optimizing formats, while significantly underestimating the AI-driven portion of their footprint.

GMSF v1.3: The Urgent Need to Account for AI’s Carbon Footprint

The Green Media Sustainability Framework (GMSF) brought much-needed standardization to carbon measurement for digital campaigns. But it was designed before the generative AI explosion we are experiencing in 2026.

What GMSF v1.2 Does (and Does Not) Cover

GMSF v1.2 provides a robust methodology for:

• Allocating emissions linked to data centers, networks, end-user devices, and formats — based on impressions, file sizes, and delivery chains.
• Distributing these emissions across ecosystem stakeholders (advertisers, agencies, publishers, platforms).

However, v1.2 remains largely silent on the specific footprint of AI inference:

• The optimization and recommendation models used by DSPs and SSPs are not modeled as a distinct emissions category.
• Creative production via generative AI is not explicitly accounted for.
• Generative search + ads experiences and AI-powered advertising assistants lack a methodological framework.

In a context where generative AI already accounts for a significant share of data center consumption, this gap has become a critical blind spot.

To understand the difference between the GMSF and spend-based approaches, see our comparative analysis: GMSF vs. spend-based methodology.

Cannes Lions 2026: A Unique Window for GMSF v1.3

The unveiling of GMSF v1.3 at Cannes Lions this year comes at a pivotal moment:

• The majority of major media groups and platforms have integrated AI into their advertising stacks.
• Advertisers are demanding more granular carbon reporting that includes the technological upstream of their campaigns.
• The public debate over AI’s energy sustainability is intensifying.

GMSF v1.3 has both the opportunity and the responsibility to open a new chapter: measuring emissions linked to AI inference within the advertising supply chain.

Toward an “AI Carbon Surcharge”: Making AI Visible in Carbon Reporting

To prevent AI from becoming a permanent blind spot in media carbon accounting, we propose an operational concept: the AI carbon surcharge.

The Principle: An Additive AI Coefficient

The goal is not to penalize AI, but to make it visible in carbon calculations:

1. Identify AI usage across the media supply chain:

   • AI-driven bidding and optimization (DSP, SSP, platforms).
   • AI-generated creative (text, image, video, dynamic assets).
   • AI-powered search and ad-serving assistants.

2. Assign each usage an additional emission factor (per impression, per query, or per generated asset), based on:

   • The estimated energy consumption of the model.
   • The actual volume of model calls used for the campaign.

In practice, this translates into an AI coefficient applied on top of existing GMSF calculations: a carbon surcharge proportional to the AI intensity of the campaign.

Three Concrete Use Cases

1. AI-Driven Bidding & Optimization

For a DSP running a real-time AI model, an additional emission factor is applied per thousand impressions processed or per thousand bid requests scored by the model.

2. AI-Generated Creative

When an advertiser generates hundreds of creative variants via AI for a campaign, an emissions module is integrated covering the generation phase and in-flight iterations.

3. AI-Powered Search & Assistants with Ads

For generative search experiences that incorporate ads, an additional factor is applied per sponsored query processed by the AI model, with differentiation between simple responses and rich responses (images, tables, video).

This mechanism enables an objective comparison between two configurations:

The advertiser can then make informed trade-offs, rather than absorbing an invisible carbon cost.

How Carbon Intelligence Is Preparing for This New Reality

At Carbon Intelligence, we believe the next frontier for responsible AdTech lies at the algorithm level — not just at the impression level.

In practice, this means:

• Detecting and quantifying the AI component of a campaign: model types, frequency, and intensity of usage.
• Delivering an “AI carbon surcharge” module that can be integrated into GMSF reporting and scores, as soon as v1.3 opens the door to this dimension.
• Collaborating with platforms, AdTech vendors, and agencies to:
  • Obtain the necessary data without compromising trade secrets.
  • Build realistic emission factors for the main AI use cases.
  • Test multi-criteria optimization scenarios: business performance, cost, and AI + non-AI carbon footprint.

This technological ambition unfolds against a regulatory backdrop that, despite political turbulence, continues to accelerate in both the US and Europe. Advertisers subject to carbon reporting obligations (SB 253 in California, CSRD in Europe) will need to measure all of their emissions — including those generated by the AI algorithms powering their campaigns.

Our goal: to build a sustainable AdTech ecosystem where AI genuinely drives down digital advertising emissions, rather than shifting the problem into an opaque technological layer.

Key Takeaways

Sources and References

• International Energy Agency (IEA), Electricity 2024 — Analysis and forecast to 2026, reports on data center, AI, and cryptocurrency electricity consumption — iea.org
• Ligozat, A.-L. & de Vries, A., “Generative AI: Energy Consumption Is Exploding,” Polytechnique Insights — polytechnique-insights.com
• Bon Pote, “Artificial Intelligence: The True Environmental Cost of the AI Race” — bonpote.com
• IRIS France, “Are New Technologies (AI, Data Centers) Compatible with Environmental Sustainability Goals?” — iris-france.org
• de Vries, A., “The growing energy footprint of artificial intelligence,” Joule (2023) — calculations based on NVIDIA and AMD chip consumption data
• Alterna Énergie, “What Is the Energy Consumption of Data Centers?” — alterna-energie.fr
• EX2, “AI Will Consume 50% of Data Center Energy by 2026” — ex2.com

This article reflects the state of available data as of February 27, 2026. The projections cited are based on scenarios published by the IEA and academic research, with the inherent uncertainties associated with such exercises.