By Stuart Kerr, Technology Correspondent, LiveAIWire

Seventy-four percent of the claims Big Tech makes about AI’s climate benefits are unproven, according to the first systematic analysis of the industry’s own statements, published in February 2026. Researchers examined 154 public claims that AI will deliver a net benefit to the fight against climate change, sourced from companies including Google and Microsoft and from institutions such as the International Energy Agency. Only 26 percent of those claims cited a published academic paper. More than a third cited no evidence at all. Machine greenwashing, in other words, is not a fringe accusation from activists with an axe to grind. It is what you get when you actually go and check the receipts.

The report was commissioned by a consortium of environmental groups, Beyond Fossil Fuels, Climate Action Against Disinformation, Friends of the Earth U.S., the Green Screen Coalition, the Green Web Foundation and Stand.earth, and authored by independent climate and energy analyst Ketan Joshi. Its most striking finding is not the 74 percent figure itself but what sits underneath it: the analysis could not identify a single case in which a consumer generative AI product, ChatGPT, Gemini or Copilot among them, had delivered a material, verifiable, substantial reduction in emissions anywhere in the world.

The report’s release was timed deliberately, arriving two days ahead of the AI Impact Summit 2026 in New Delhi, where government and industry leaders were due to discuss exactly the kind of AI-for-climate partnerships the report says have gone unverified.

Machine Greenwashing’s Bait and Switch: What “AI Helps the Climate” Really Means

The mechanism behind that gap has a name, and the report gives it one: bait and switch. Claims about AI’s climate benefits routinely blur two technologies that behave nothing alike. Traditional machine learning, the kind used to forecast wind patterns, optimise a power grid, or fine-tune irrigation timing on a farm, has a real and often well-documented climate case. Generative AI, the large language models and image generators that produce the “wow” demos tech companies show off at keynotes, carries a substantially larger and still-growing environmental cost, and has yet to produce a verified case study of comparable climate benefit at scale.

When a company cites AI’s climate potential in a sustainability report or a public statement, the report found it is often unclear which of these two technologies is actually doing the work being credited. That ambiguity is not an accident of loose language. Ketan Joshi put it directly: “It appears tech companies are using vagueness about what happens within energy-hogging data centres to greenwash a planet-wrecking expansion.” The vagueness, in other words, is doing commercial work. It lets a genuine, narrow success in agricultural forecasting or grid balancing stand in for a much broader and far less substantiated claim about generative AI’s climate credentials.

What This Means for You

If you are trying to work out whether a company’s AI-and-climate claim is worth taking at face value, the report’s own methodology gives you a usable filter. Ask which type of AI is actually being described. A specific, named application, wind forecasting, crop irrigation scheduling, grid dispatch optimisation, with a cited, published study behind it is a meaningfully different claim from a general assertion that “AI will help solve climate change.” The second kind of claim is exactly the sort that made up more than a third of the statements the report examined with no evidence attached at all.

It is also worth remembering that scepticism should run in both directions. Some widely repeated claims about AI’s environmental harm are just as loosely sourced as the industry’s claims about its benefits. The figure that a single ChatGPT query uses ten times the energy of a Google search traces back to an offhand 2023 remark by Alphabet chairman John Hennessy, not a measured study.

OpenAI’s own disclosure, published by chief executive Sam Altman, puts a typical query at roughly 0.34 watt-hours, in the same range as the search estimate it is usually compared against, which itself dates to 2009 and predates a decade and a half of hardware efficiency gains on both sides of the comparison. The fog here cuts both ways, and neither figure has been independently peer reviewed.

The Specific Promises Companies Have Made, and How They Are Holding Up

Google states that its data centres run on electricity that is roughly 90 percent matched to renewable sources on an annual accounting basis. Meta has set a target of net-zero emissions across its operations by 2030. Microsoft has gone further than either, pledging to be carbon negative by 2030 and to remove its entire historical carbon footprint by 2050. None of these commitments are fabricated, and all three companies have put real money behind renewable power purchase agreements and carbon removal contracts.

The strain is showing regardless. Microsoft’s own 2025 environmental sustainability report disclosed that total emissions have risen by roughly 23 percent since 2020, driven overwhelmingly by the Scope 3 emissions embedded in building the data centres its AI ambitions require, which now account for more than 97 percent of the company’s total carbon footprint. Microsoft points out, fairly, that the increase looks modest against a 168 percent rise in energy use and 71 percent revenue growth over the same period, and it has cut its own direct Scope 1 and 2 emissions by nearly 30 percent since 2020. The Scope 3 number is the one that actually reflects the data centre build-out, and it has moved in the opposite direction.

In April 2026 the company paused new purchases of carbon removal credits, before signing a fresh seven-year, 650,000 tonne removal deal with a Danish startup weeks later, a sequence that reads less like abandonment of the carbon-negative target and more like a company visibly struggling to reconcile that pledge with the physical footprint of the infrastructure it is racing to build. Microsoft describes itself as “pragmatically optimistic” about still hitting carbon negative by 2030. That phrase is doing a great deal of work for a target four years away, on a trajectory currently pointed the wrong direction.

Annual renewable matching is a real commitment, but it answers a different question from the one that matters for a specific AI training run happening on a specific afternoon. A data centre can be matched to 90 percent renewable electricity across a full year and still draw heavily from a coal or gas-heavy grid during the exact hours a large model is training, if that training happens to fall on a still, cloudy day when the wind and solar contracted to the company are not actually generating.

The annual figure is accurate. It is also close to meaningless for judging the carbon intensity of any individual burst of AI compute, which is precisely the level of detail the industry has been reluctant to disclose. None of the companies examined in the February 2026 report currently publish hour-by-hour or workload-specific emissions data for their AI systems, which means the headline renewable percentage is, at present, the only number the public actually gets to see.

Where AI’s Climate Case Is Genuinely Strong

None of this means every AI-and-climate claim is hollow. The distinction the greenwashing report draws between traditional and generative AI holds up under scrutiny specifically because the traditional side has real, audited wins behind it. Voyage optimisation systems that recalculate a container ship’s speed and routing against live weather data produce metered reductions in fuel burned, verifiable against vessel logs rather than modelled projections. AI-guided irrigation and fertiliser timing has produced documented reductions in nitrous oxide emissions, a gas roughly 300 times more potent than carbon dioxide per molecule, across multiple crop systems tracked by international agricultural bodies.

LiveAIWire’s own examination of AI’s genuine emissions payback found this same pattern across grid management, building climate control and logistics: real, auditable savings that exist alongside, not in place of, a real and growing energy cost from the generative side of the industry. Weather and climate forecasting systems built on machine learning, the kind LiveAIWire covered in an earlier look at AI’s climate applications, are among the strongest examples on record: DeepMind’s flood and cyclone forecasting tools and Google’s GraphCast weather model have both been validated against independent meteorological benchmarks, not just company press releases.

That is the accounting Big Tech’s public statements have mostly failed to do. Crediting the whole AI category with the wind-forecasting win while quietly expanding the generative AI infrastructure that carries none of its climate benefit is exactly the substitution the February 2026 report set out to document, and exactly why the distinction between the two kinds of AI needs to survive the trip from a research paper into a corporate press release.

Why the Industry Keeps Getting Away With It

Part of the answer is structural. The carbon cost of AI training and inference remains largely invisible to the people generating it. A user typing a prompt into ChatGPT receives no indication of the energy that response consumed. A business running AI workloads in the cloud typically sees that consumption folded into a general computing bill that obscures the AI-specific contribution entirely. Where costs are invisible, benefits are much easier to overstate, because there is no equally visible cost sitting alongside the claim to contradict it.

Regulation has been slow to close that gap. The International Energy Agency, whose own Energy and AI report was among the sources the greenwashing analysis scrutinised, has itself acknowledged that data centre electricity demand is growing faster than the renewable capacity being added to serve it in most major markets, even as some of its own framing has been criticised for blurring the same traditional-versus-generative distinction the wider report flags. Jill McArdle of Beyond Fossil Fuels put the core problem in blunter terms: “Big Tech companies are writing themselves a blank cheque to pollute on the empty promise of future salvation. We cannot bet the climate on these baseless claims.”

Not every corporate response has been purely rhetorical. Google’s 2025 nuclear energy agreement with Kairos Power, aimed at supplying its AI data centres with round-the-clock, zero-carbon electricity from the early 2030s, is a genuine attempt to solve the underlying power problem rather than simply claim credit for it. The gap between that kind of concrete, decade-long infrastructure commitment and a vague sustainability-report line crediting “AI” for an unspecified climate benefit is precisely the distinction the February 2026 report is asking the industry to be honest about.

What Genuine Transparency Would Actually Require

The fix the report’s authors point toward is not complicated in principle, even if it is politically difficult in practice. Companies would need to disclose energy consumption at the level of a specific deployment, not just an annual company-wide average. They would need to state plainly which category of AI, traditional or generative, is actually responsible for any climate benefit being claimed. And they would need third-party verification of energy and emissions figures before those figures appear in a sustainability report, the same standard already applied to financial accounts.

None of that requires new technology. It requires companies whose entire public pitch depends on AI’s transformative potential to accept a much less flattering, much more granular form of disclosure about what their own AI infrastructure is actually costing the planet in the meantime. Until that disclosure exists, the honest position on any sweeping claim that AI will save the climate is the one the researchers themselves reached after reading 154 of them: unproven, and worth checking before you repeat it.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.