OpenAI and Google's AI climate promises lack evidence, op-ed argues

As countries meet in Antalya, Turkey for the United Nations climate conference, Mary Nichols and Jeremy Tamanini argue in an MIT Technology Review opinion piece that the climate case for AI has not been made. The conference host has already called on countries to commit to ensuring that AI is designed, powered and deployed in ways that support climate action. The authors say industries are not doing nearly enough to use the technology to cut greenhouse-gas emissions, despite optimistic claims from AI companies.

They cite two such claims. In 2024, OpenAI CEO Sam Altman declared that AI would unleash "astounding triumphs," including "fixing the climate." In a 2023 op-ed, the head of sustainability at Google and her coauthor wrote that AI could reduce global emissions 5% to 10% by 2030.

The authors then turn to academic work. A peer-reviewed study published last year, led by researchers at the London School of Economics' Grantham Research Institute and the firm Systemiq, estimated that AI could help cut global emissions by the equivalent of 3.2 to 5.4 billion tons of carbon dioxide a year by 2035. On the high end that is roughly on par with the US's annual energy-related emissions, and the figure is net of AI's own energy footprint. The gains are mostly concentrated in three sectors: power, meat and dairy, and vehicles.

The piece asks whether the largest companies in those sectors are integrating AI in line with the projection. An AI-assisted review of public information from significant market actors in each sector was conducted and reviewed by the piece's second author. The analysis itself was carried out by Anthropic's Claude Code, and its findings were checked against public documents and media accounts. It found that few companies have announced they are pursuing the specific emissions-reducing routes behind the study's projections, and fewer still claim to have used them for meaningful climate progress.

Power. This is where the earlier study's math is most concrete: AI-driven forecasting of supply and demand could raise the amount of renewable electricity reaching the grid from already-built solar and wind facilities by up to 20%, which could prevent an estimated 1.8 billion tons of emissions a year. Yet among the grid operators reviewed in China, the United States and Europe, only one deployed a known meaningful, working AI program of this sort. That was a partnership between the UK's National Grid ESO and Open Climate Fix, which halved the operator's solar-forecast error and avoided hundreds of thousands of tons of CO2 a year. Many other operators in the EU, the US and China had little to show in the public record.

Meat and dairy. The authors call this the starkest contrast between technology readiness and market reality. To reach the 2035 cuts, AI would need to close the taste and cost gap between traditional proteins and alternatives such as lab-grown or plant-based burgers and milk substitutes. One of the clearest efforts is NotCo, which recently formed a joint venture with Kraft Heinz using its platform Giuseppe; the platform analyzes plant molecular data to find combinations that replicate the taste, texture and smell of animal products. The review suggests few other alternative-protein companies use AI so deeply in product development. The sector also seems to face more fundamental problems, including slow mainstream consumer adoption and falling funding: venture capital flowing into cultivated meat fell from a $989 million peak in 2021 to around $83 million last year.

Vehicles. Here the authors see hopeful examples, particularly among battery makers and ride-sharing platforms. CATL, which controls roughly 40% of the global market for electric-vehicle batteries, launched an AI system trained on more than 100,000 battery design cases and 600 terabytes of test data. According to the company, it cut the time needed to screen candidate materials by two orders of magnitude. Independent researchers found that Didi Chuxing cut per-trip emissions 16% on average by using AI-optimized routing to minimize detours.

The authors add further problems. Oil and gas companies are also interested in AI tools, including for the potential to increase fossil-fuel production, and tech firms that committed to sustainability goals in the past should answer questions about how their tools could boost old, dirty energy as well as clean energy. Integrating AI is only a first step; it remains unclear how far the technology can speed up the envisioned improvements, and it does little to shorten the standard paths to certifying, permitting, building, packaging, marketing and selling. They concede that some companies and utilities may be pursuing relevant AI applications that have not been disclosed publicly.

Greater and better-targeted AI integration and efficiency gains may well materialize and bring the 2035 projections closer, they write, but not without deliberate and sustained effort. Meanwhile, rapid AI adoption and data-center buildout have driven a huge increase in energy consumption and emissions, and major tech companies such as Microsoft and Google have fallen well off track on their corporate climate goals. The authors say it would be surprising if the sector even manages to zero out its own climate pollution. Emissions associated with global data centers, which may handle AI and non-AI computing, could exceed 300 million tons annually in 2035 in the IEA's base forecast.

Their recommendations: industry leaders should implement the technology where it promises to cut emissions most; corporate customers should choose models with the lowest environmental impacts and work with AI developers doing the most to reduce their climate footprint; and states and regions need to do more to incentivize or require responsible operation.

Key facts

An AI-assisted review (run with Anthropic's Claude Code and checked against public documents and media accounts) found few firms in the power, meat and dairy, and vehicle sectors pursuing the AI routes behind a 2035 emissions projection.
The peer-reviewed LSE Grantham Institute and Systemiq study estimated AI could cut global emissions by 3.2 to 5.4 billion tons of CO2 equivalent a year by 2035, net of AI's own energy use.
Of the grid operators reviewed in China, the US and Europe, only one had a known meaningful, working AI program: National Grid ESO with Open Climate Fix, which halved its solar-forecast error.
Cultivated-meat venture capital fell from a $989 million peak in 2021 to around $83 million last year; CATL and Didi Chuxing are cited as hopeful vehicle-sector examples.
Data-center emissions could exceed 300 million tons annually in 2035 in the IEA's base forecast, and the authors say tech firms have fallen well off track on their own climate goals.

Why it matters

AI leaders have promised big climate payoffs: Sam Altman said AI would help with "fixing the climate," and a Google sustainability head wrote in 2023 that AI could cut global emissions 5% to 10% by 2030. With the UN climate conference under way and its host urging countries to commit to climate-supportive AI, the piece tests those promises against what companies in the three sectors the main study identifies are actually doing. Its verdict is that the evidence so far is thin, while AI's own energy use and emissions are already rising.

Who it affects

Energy and grid operators, alternative-protein and dairy-substitute companies, and battery makers and ride-sharing platforms are the firms examined. AI developers and big tech companies such as Microsoft and Google are affected through the claims made on their behalf and their missed corporate climate goals. The authors also address corporate customers who buy AI models, and states and regions that could incentivize or require more responsible operation. Oil and gas companies come up as another set of AI users.

How to use it

The authors give a short list of actions. Industry leaders should deploy AI where it promises the biggest emissions cuts. Corporate customers should select models with the lowest environmental impacts and prefer AI developers doing the most to reduce their climate footprint. States and regions should do more to incentivize or require responsible operation. Tech firms with past sustainability commitments should answer questions about how their tools could be used to boost fossil-fuel production.

How solid is it

This is an opinion piece by Mary Nichols (distinguished counsel at the UCLA School of Law and former chair of the California Air Resources Board) and Jeremy Tamanini (founder of Dual Citizen, a sustainability research and advisory firm). The underlying projection comes from a peer-reviewed study led by researchers at the LSE's Grantham Research Institute and Systemiq. The review was AI-assisted, carried out by Claude Code and checked against public documents and media accounts, and it relies on public information. Some figures come with their own limits: CATL's two-orders-of-magnitude speedup is the company's own account, while the Didi result was found by independent researchers. Several wording choices are hedged in the source, such as the review "suggests" and the sector "seems" to face deeper problems. The article reports no new quantitative results of its own beyond the review's findings.

Risks and caveats

The authors say some companies and utilities may be pursuing relevant AI applications that have not been disclosed publicly, so the review's finding reflects the public record only. They also say greater and better-targeted AI integration may well bring the 2035 projections closer, though not without deliberate and sustained effort. The IEA data-center figure covers data centers that may run AI and non-AI work, so it is not an AI-only number. The piece says it remains unclear how much AI can accelerate the improvements envisioned, and that AI does little to shorten certification, permitting, building and sales.

“projections like these are only as good as our willingness to test them against present-day reality.”

— Mary Nichols and Jeremy Tamanini, MIT Technology Review