Your AI strategy might be quietly undermining your climate commitments
Most organizations adopting AI don't have the governance infrastructure to know whether their climate commitments are still holding. That's a problem, and it's about to become visible.
In 2020, Google CEO Sundar Pichai vowed to make the company "the first major company to operate carbon free — 24 hours a day, seven days a week, 365 days a year" by 2030. Last year, that commitment quietly disappeared from the company's sustainability homepage and reappeared in a report appendix. The company's own environmental report explains why: "Running the global infrastructure behind our products and services, including AI, takes considerable energy."
Google's position is an extreme version of a problem that is spreading far beyond Silicon Valley. As AI adoption accelerates across every sector, organizations are accumulating an environmental liability they mostly can't see, don't measure, and haven't accounted for in the commitments they've made to stakeholders.
This isn't primarily a story about Big Tech's energy consumption, though that's real enough. It's a story about governance, and about what happens when organizations deploy AI without the infrastructure to understand what it costs.
The gap most organizations don't know they have
In 2025, Capgemini surveyed 2,000 senior executives from organizations with active generative AI initiatives. Forty-eight percent said their GenAI use had increased their organization's greenhouse gas emissions. Forty-two percent said they'd had to re-examine their climate goals as a result. And yet only 12% said their organization actively measures its GenAI environmental footprint.
These aren't laggards. The survey targeted organizations with revenues above $1 billion that were already well into AI deployment, across manufacturing, financial services, retail, energy, and public sector. The governance gap it documents isn't an early-adopter problem. It's baked into how AI is being deployed at scale.
The most telling figure is this: 74% of executives cited lack of transparency from AI providers as the primary obstacle to measurement. They want to account for their AI usage. They largely can't, because the companies whose models they're running — OpenAI, Anthropic, Google, Microsoft — treat their energy and emissions data as proprietary. The infrastructure for accountability doesn't exist upstream, so it can't exist downstream either.
It's worth noting that this opacity is a choice, not an inevitability. In July 2025, Mistral AI published a full lifecycle environmental analysis of its Large 2 model — training emissions, inference costs, water consumption, hardware footprint — conducted with carbon accounting firm Carbone 4 and validated by France's environmental agency ADEME. The data exists when providers decide to release it. Most don't, which means organizations using any other major AI system are flying blind on the environmental cost of each deployment decision they make.
Why this matters for organizations with climate commitments
The governance gap would be manageable if it were simply a measurement challenge. What makes it a credibility risk is that many of the organizations driving AI adoption fastest are the same ones that have made the most prominent public climate commitments.
The 2025 Thomson Reuters Foundation and UNESCO AI Company Data Initiative — the largest global dataset of corporate AI governance disclosures, spanning nearly 3,000 companies — found that only 11% of companies report conducting Environmental Impact Assessments for their AI systems. Environmental and human rights considerations are consistently deprioritised in AI governance frameworks in favor of data privacy and compliance. The report's conclusion is blunt: "AI adoption is clearly outpacing governance."
Under the EU's Corporate Sustainability Reporting Directive, large organizations now face mandatory disclosure of climate-related data. The AI Act, now in phased rollout, adds requirements around environmental impact assessment for certain high-risk AI systems. CSRD and the AI Act are converging on exactly the terrain where AI governance is currently most absent. The compliance exposure is coming.
But the more immediate risk is reputational. Organizations that have made net-zero pledges, earned carbon certifications, or published sustainability roadmaps have staked institutional credibility on a specific standard of environmental integrity. When AI adoption quietly inflates emissions outside the governance perimeter, through untracked individual use, through third-party tools never assessed for environmental cost, through the embodied emissions of hardware that even a well-intentioned energy accounting misses, that standard erodes in practice while holding in public statements. The gap between what is claimed and what is real becomes the story.
This is structurally identical to what happens in every other social-issue crisis: the problem was never the communication. It was the governance failure that communication was eventually asked to cover.
The harder question isn't technical
The Capgemini report identifies something worth sitting with. Seventy-four percent of executives who aren't measuring their AI footprint say it's because AI providers won't give them the data. But 58% also say that driving AI efficiencies is a higher priority than measuring environmental impact. Those aren't the same obstacle. One is structural; the other is a choice.
A Carbon Direct analysis of AI infrastructure makes this concrete. For data centers powered by low-carbon electricity, which is where climate-conscious organizations prefer to route their workloads, embodied emissions from hardware manufacturing can represent up to 40% of total lifetime greenhouse gas emissions. Organizations that do manage to measure their operational energy are likely still systematically undercounting their actual environmental cost.
There's no clean technical fix for this. The path from current practice to genuine accountability requires decisions: deciding that AI's environmental cost is a governance issue, not a sustainability team issue; requiring that vendor selection criteria include environmental transparency; building measurement into deployment decisions before scale makes it prohibitive; and accepting that "we can't measure it because our provider won't tell us" is no longer a defensible position when regulators and investors are asking.
Organizations that make those decisions now, before the gap becomes public, are better positioned than those that make them under pressure. That calculus is as true for AI and climate as it is for any other area where governance and communications eventually have to account for each other.
I work with organizations navigating the intersection of governance, stakeholder expectations, and communications strategy, including how AI adoption sits within broader ESG and social-issue positioning. If you're trying to understand where your governance infrastructure might not be keeping pace with your public commitments, I'd like to hear from you.