Is AI really going to meet 20% of our 2030 climate goals?

I read, just now, in a post on Fortune by Google's head of sustainability Kate Brandt and global chair of consultancy BCG Rich Lesser a pretty incredible statistic:

"Research shows ?that by scaling currently proven applications and technology, AI could mitigate 5 to 10% of global greenhouse gas emissions by 2030–the equivalent of the total annual emissions of the European Union"

Brandt and Lesser highlight that COP28 featured a wave of new AI-themed panels, events and programs. "For the first time, AI was highlighted at COP28 as one of the key potential solutions to tackle climate change". It's true: it has replaced crypto, blockchain, web3, NFTs and the metaverse as the new COP buzzword.

Reducing total global greenhouse gas emissions by 10% in the next seven years would be incredible. That is a full fifth of what total planetary emissions need to reduce by, to at least somewhat align with the Paris Climate Agreement's goals. As you can guess, I was curious about the research behind the number.

It gets cited semi-frequently. This article , for instance. A paper published recently in the journal Energy Nexus cites it. And here .

In November this year, Google published a blog post entitled "Accelerating climate action with AI", authored by Brandt and VP of Engineering Yossi Matias.

The top line: "AI has the potential to mitigate 5-10% of global greenhouse gas emissions according to our new report with Boston Consulting Group".

Click through to the report , and you get to this section:

The Microsoft / PWc study linked by BCG at least does the classic consultancy trick of embedding bad assumptions and clumsy modelling in a fancy-looking set of outputs that mimics scientific research. The Capgemini survey literally just asks people what they think the reductions might be.

But what about the BCG study? Well - okay, another link. It's a weird nest of citations, but sure, let's have a look. Click through and you get to a 2021 blog post - not a study, report, or research of any kind - from BCG, which, as far as I can see, comes up with the 5% to 10% figure in this way:

I swear I stared at this for ten full minutes trying to see where I'd missed the hyperlink. But no: it truly seems the origin of this claim is BCG vaguely intuited the percentage of reductions from their dealings with clients, and then literally just applied to the entire planet's greenhouse gas emissions.

That has mutated over two years into a "study by BCG", and most recently into just "research", which almost sounds like it was a scientific paper and not an extremely clumsy calculation in an old blog post.

The Google-commissioned BCG report does serve up some useful case studies in which AI, as a software tool, has helped a few climate projects, including the very interesting ClimateTrace project I shared recently.

But big tech comes burdened with a toxic habit for exaggeration and hype. Sure, climate is an area in which we need to move fast, but not if it comes at the cost of breaking things.

And one thing we can clearly see being broken is the ability to shut down fossil fuels on power grids while power demand is skyrocketing thanks to massive new data centres - new renewables end up simply filling in that new demand rather than cutting down into coal and gas output.

Texas, with its record-breaking renewable growth yet unchanged emissions, is a nice example :

As is Ireland :

Not all new demand here is data centres, and not all data centre demand is running AI stuff. But it is the centrepiece, and the most clearly advertised growth engine, in companies running hard to both promote it as a climate solution but also market it as a product.

In five years, Google's total power consumption has doubled. In four, Microsoft's will double too:

And as Benedict Evans points out in his recent Slush presentation , investment in energy hungry data centres is only just getting started:

It's already very clear that the old "renewable energy certificate" schemes will facilitate worse emissions rather than mitigate them, when it comes to new sources of electrical demand - a carbon offsetting scheme for electrons.

Improvements in data centre efficiency seem to be plateauing , and while schemes like 24/7 matching of carbon free energy can help ensure new demand is powered by new clean sources, it doesn't necessarily help push pre-existing fossil fuels out of the grid. Don't forget the deeper context: our job isn't to run hard in the same spot, it's to get rid of fossil fuels. Power grids with rising demand that's met by new clean power but fossil fuels that remain generating the same amount lead to climate disaster.

Neither efficiency nor clean-power matching are bad: but are either or both enough to counter the effects of a massive new volume of terawatt hours of demand? It really doesn't seem like it.

There is already a shadow of suspicion around big tech companies, and almost always with good reason. Overinflated promises and bad business practices are helping to confirm these suspicions. A bit of honesty goes a long way.