Silicon Valley has adopted the greenwashing playbook perfected by oil majors, according to a forensic examination of how technology companies promote artificial intelligence as a climate solution whilst their energy-hungry datacentres devour ever-larger shares of the electricity grid.
An analysis of 154 industry statements reveals a systematic conflation between traditional machine learning and the generative AI tools now driving datacentre expansion. The research, commissioned by nonprofits including Beyond Fossil Fuels and Climate Action Against Disinformation, found precisely zero examples where consumer-facing products like ChatGPT, Google's Gemini, or Microsoft's Copilot produced material, verifiable emissions reductions.
The tactics mirror those of fossil fuel companies advertising modest solar investments whilst drilling deepwater wells. But tech firms have refined the approach, according to Ketan Joshi, the energy analyst who authored the report.
The evidence gap
When companies tout AI's climate credentials, they're typically referring to decades-old machine learning applications — predictive models for building energy efficiency or grid optimisation — rather than the large language models and image generators currently reshaping the industry's infrastructure demands. This muddling of technologies serves a purpose: it allows firms to claim climate benefits whilst side-stepping questions about the emissions profile of their fastest-growing products.
The evidentiary standards underpinning these claims would struggle to pass peer review. Only 26% of the green claims examined cited published academic research. A striking 36% provided no evidence whatsoever. The remainder pointed to corporate websites or industry reports, including an International Energy Agency analysis that leading tech companies reviewed before publication.
Consider the widely circulated assertion that AI could mitigate 5% to 10% of global greenhouse gas emissions by 2030. Google repeated this figure as recently as April 2024. Its source? A BCG report commissioned by Google, which traced the claim back to a 2021 blogpost the consulting firm wrote citing its own "experience with clients". What appears to be robust analysis is actually a circular reference dressed up as research.
Energy demands obscured by partial disclosure
Sasha Luccioni, who leads AI and climate research at Hugging Face, an open-source AI platform, draws a sharp distinction that industry communications deliberately blur. "When we talk about AI that's relatively bad for the planet, it's mostly generative AI and large language models," she noted. The climate-friendly applications are "often predictive models, extractive models, or old-school AI models" — precisely the technologies that predate the current boom.
The energy consumption gap between these categories couldn't be starker. A simple text query to ChatGPT might consume roughly a minute of lightbulb electricity, according to partial industry disclosures. Complex functions like video generation and Microsoft's deep research features require exponentially more power. The operative phrase here is "partial industry disclosures" — companies release selective data that obscures rather than illuminates the true scale of consumption.
Datacentres currently account for just 1% of global electricity demand. But BloombergNEF projects their share of US electricity will surge to 8.6% by 2035, more than doubling from present levels. The International Energy Agency estimates datacentres will drive at least 20% of electricity demand growth across developed economies through the decade's end. These figures assume current trajectories; they don't account for potential acceleration as AI capabilities expand.
The greenwashing playbook, version 2.0
What's striking about Joshi's characterisation of these tactics as "diversionary" is how precisely they replicate fossil fuel industry strategies from previous decades. Oil companies promoted carbon capture technology that might address a fraction of 1% of their emissions whilst positioning themselves as climate leaders. Tech firms now advertise traditional AI's efficiency gains whilst their core business — the energy-intensive generative models attracting billions in investment — expands unchecked.
"These technologies only avoid a minuscule fraction of emissions relative to the massive emissions of their core business," Joshi observed. "Big tech took that approach and upgraded and expanded it."
Google defended its methodology in response to the analysis, stating that its "estimated emissions reductions are based on a robust substantiation process grounded in the best available science" with transparent principles and methodology. Microsoft declined to comment. The IEA did not respond to enquiries.
The response highlights the central tension: what companies describe as robust substantiation looks rather different when examined by independent analysts tracking evidence chains back to their origins.
What happens when the narrative meets reality
The datacentre expansion now underway will reshape electricity infrastructure across developed economies. Utilities are delaying coal plant retirements and exploring new gas capacity to meet projected demand. Some jurisdictions are prioritising AI datacentre connections over residential developments. These are tangible infrastructure decisions being made on the basis of industry claims about AI's net climate impact.
Investors pumping capital into AI infrastructure and the policymakers writing regulations around datacentre development need a clearer picture than industry communications currently provide. The question isn't whether machine learning can optimise building energy systems — it demonstrably can. The question is whether the generative AI tools now dominating investment and infrastructure planning produce climate benefits that justify their electricity consumption, or whether we're witnessing an expensive categorical error with decades of consequences.
The speed at which datacentres are being built suggests the market has already answered that question. Whether the answer is correct depends on evidence that, according to this analysis, the industry has yet to provide.
