“Dangerous consequences” – how AI’s climate framing lets Big Tech off the hook

As tech giants race to build out AI and the sprawling infrastructure it depends on, climate concerns have tended to focus on one thing: power-hungry data centres.

Their electricity use is growing so fast that by 2030, it’s projected to be nearly three times more than the combined annual consumption of Pakistan, Bangladesh and Nigeria. With the explosion in the construction of data centres driving new investment in fossil fuels, especially in the US, greenhouse gas emissions generated by data centres – now standing at less than 1% of the global total – are set to soar.

But this narrow focus on electricity has let AI’s supporters and the International Energy Agency (IEA) make a convenient case: that rising emissions can be more than offset by the technology’s green applications, like optimising renewables or boosting efficiency. That story conceals how AI’s real climate danger lies elsewhere: in the oil fields, where it’s helping fossil fuel companies extract planet-heating oil and gas faster and more cheaply.

As a senior manager at Microsoft, Holly Alpine was shocked by this blind spot. In 2024, she and her husband Will – also a Microsoft manager – quit their jobs and launched a campaign to hold Big Tech accountable for the emissions its technology enables.

Over the past two years, they have teamed up with two researchers to quantify just how deep the fossil fuel industry’s embrace of AI tools runs.

Their peer-reviewed study, published last week, found that when AI is adopted at similar rates across the fossil fuel and renewable energy sectors, the net effect is a rise in emissions of 0.47–1.8 gigatonnes of CO2 annually. That’s equivalent to Mexico’s annual emissions at the low end, and to Russia’s – the world’s fourth-largest emitter – at the high end. It is also 3.3 to 13.3 times higher than the emissions currently generated by powering AI data centres.

We spoke with Alpine about the risks of overlooking this side of the AI climate story and what can be done to shift the focus.

Q: Why has the climate conversation focused so heavily on data-centre power use when your modelling suggests that’s the smaller part of the AI emissions story?

A: It’s been quite unfortunate that it has been framed that way and that it has stuck so much because that framing is wholly incomplete, very misleading and is leading to very dangerous consequences.

It’s in the fossil fuel industry and the technology companies’ favour to frame the equation in this way because it leaves out any responsibility and accountability of the tech’s use by fossil fuel companies, which is a large part of their business. They’re some of their largest customers and they have teams of engineers and sales folks who are dedicated to the fossil fuel industry.

Simply comparing the power needed to run the technology and its [clean energy] applications is also kind of apples to oranges. On the one hand, you have real-world actual emissions and, on the other, hypothetical future avoidance of emissions as a result of potential future use cases for renewables.

What we are saying is that we need to look at both sides of the ledger for AI applications, renewables versus fossil fuels, and then also add the emissions generated by running data centers on top of it.

Q: How do AI applications help fossil fuel companies in a way that drives up emissions?

A: It’s everything from finding more oil and gas underground by processing hundreds of terabytes of seismic and well data that would otherwise have to be done manually. These AI models can process this data extremely quickly and create high-resolution images of what is underground. It helps companies pinpoint the oil and gas reserves that are most likely to be commercially recoverable.

Fossil fuel companies can identify and develop fossil fuel deposits with a lot more certainty, allowing them to move forward with projects that would otherwise have been too risky or too slow to pursue. AI makes them viable.

We’ve seen that rig counts [number of active drilling rigs] have dropped dramatically, so they need fewer resources to get out even more fossil fuels. Their costs are decreasing, while their production is increasing.

Q: How deep do these relationships run between Big Tech and fossil fuel companies? How do they compare with equivalent relationships with renewable energy companies?

A: I have to caveat that I have not worked for Microsoft for about two years. But what we saw at the time was that the fossil fuel-dedicated teams were much larger in terms of the number of employees, the size of the contracts, and the long-standing relationships.

This is not new. Microsoft has worked with the fossil fuel industry for many years and has deep partnerships, starting with the humble machine-learning going back many years. AI is just the latest wave of technology being applied in this way.

UN asks AI companies to reveal full environmental impacts

There are also relationships between the tech companies and renewables companies [and] battery storage developers. There are definitely sustainability-related applications of the technology.

One of the recommendations that we had given the company [Microsoft] was to shift the ratio of engineering resources from fossil to low and no-carbon energy sectors within the company. When they came out with their principles for engagement with the fossil fuel industry in 2023, they committed to shifting engineering resources. But then we did not see any actual change in business practices.

Q: Tech companies are now quietly scaling back some of their climate commitments, but there was a point, not long ago, when they wanted to be seen as climate leaders. Was there ever a genuine commitment to do that, or was it just an image they were projecting?

A: It depends on how you evaluate a company for its climate impact. If all we are looking is its own operational emissions, then in that case, Microsoft was and, still is to some extent, a climate leader.

But if we evaluate a company based on what it is producing, then I would say it’s a very different story. Back in 2019, ExxonMobil said it was able to produce an extra 50,000 barrels [of oil] per day purely thanks to Microsoft technology. There was also another public and quantified deal with Chevron.

We calculated that those emissions alone from just two deals among dozens were 300% of Microsoft’s entire operational emissions, including data centres. So, how do you want to evaluate your company?

If you look at other sectors and, say, evaluate a weapons manufacturer on its violence footprint, you don’t just look at their supply chain and the violence within it to create the weapons. You look at the real-world impact of the weapons they’re manufacturing. Yet we completely left technology companies off the hook.

Q: You make some recommendations as well in the paper. They include the idea of putting some supply-side constraints on this AI-enabled productivity for fossil fuel companies. What would that look like in practice?

A: Ultimately, our goal would be to have disclosure and governance measures that limit AI’s role in increasing fossil fuel productivity. The first thing would be a recognition of “enabled emissions” even as a measurable category because, at the moment, they are not included in any emissions disclosure or accountability frameworks.

Then we should require transparency around these fossil fuel contracts and constrain some of these specific mechanisms that the research identifies.

We are not trying to have a blanket ban on AI or even a blanket ban on AI use in the fossil fuel industry. There are some great applications, like methane leak detection, for example. But we just want to align applications with climate science and ensure that any contracts that move forward have been evaluated against a 1.5C future.

AI governance debate silent on risks to nature, campaigners warn

The easy thing would be for companies to voluntarily put guardrails on how their tech can be used, which is not new. There just currently are none for climate. But we do think that… policy is what needs to be implemented.

We also think that if we can change the market structure and incentives, then this kind of restriction will follow. If we look at ESG investing and how sustainable investing is defined, if we include what these companies are doing into that evaluation, then that can move capital flows.

Q: What do you think are the most promising avenues where you can shift the AI narrative and drive the change you are seeking to achieve?

A: We are now building off the study and there are various governance frameworks that we are attempting to incorporate this sort of evaluation into like the Greenhouse Gas Protocol or the Science Based Targets initiative (SBTi)

Luckily, we have seen some very promising drafts for the future of those frameworks that do include evaluations and disclosures of this work, which is really exciting.

The vote that stopped a data center: US communities query resource-hungry AI

We also need to look at companies for impacts in order to evaluate their sustainability metrics, and there could be potential greenwashing concerns that we could address on the legal side of things.

And then [there are] different policy workstreams. In the EU, we were quite hopeful about the AI Act,and the various use cases that were classified as high risk and would go through additional scrutiny. Unfortunately, with the Omnibus passing [in July], that opportunity is a little restrained.

But now with the Cloud and AI Development Act (CADA) coming out with various European frameworks around evaluating tech’s impacts, we hope to inform those discussions with this research.

The post “Dangerous consequences” – how AI’s climate framing lets Big Tech off the hook appeared first on Climate Home News.