Why Anthropic’s ultra-dirty deal shouldn’t surprise you at all – Ketan Joshi

Skip to content

This is a quasi-epilogue to my recent essay linking infrasound fears from wind power to the same around data centres. I found through my career in the wind industry that weird wind farm health beliefs were mostly a marker of development done wrong. Irate communities were exploited by bad-faith disinformation groups. Data centres see something similar, but the scale is different. Impatience, bloat and corruption are resulting in growth that is frequently harmful.

What we know is that most apples are bad apples. The fundamental drivers of this change are panic and impatience, and that is what is behind the worst impacts of data centre growth, including a material risk to decarbonisation and climate action.

Anthropic are a frustrating company because they’re pretty clearly bad actors who nevertheless seem to draw a huge amount of support from well-intentioned progressives, centrists, lefties and generally the broad world of professional knowledge workers.

Their chatbot, Claude, is seen as the sensible ‘serious’ option, as competitors like OpenAI’s ChatGPT and Google’s Gemini fall behind.

A few months back, the #QuitGPT campaign kicked off. It piggybacked off public outrage around OpenAI’s seeming willingness to work with the US government Department of War, and Anthropic’s seeming hesitancy. That distinction was wildly exaggerated, with Anthropic more than happy to supply its services for acts of extreme violence, as long as that violence was directed at foreigners and not Americans. That did not stop the QuitGPT campaign from frequently and openly urging people not just to quit GPT, but to join Claude, and specifically urge people to take up paid subscriptions.

This was specifically led by historian Rutger Bregman, if not a part of the ‘effective altruist’ movement at least a fan of it. At a few points it got really, really egregious.

The heroes at Anthropic have been struggling to provide enough computational power to meet the demands of an ultra-inefficient text generation program, resulting in usage restrictions even for users on the top paid tiers (probably at least partly driven by this surge in users).

So, in their heroic fashion, they decided to do a deal with one of the most well known ultra-polluting fossil-fuelled data centres in the US:

This is, amusingly, the first time Anthropic has ever disclosed anything even close to meaningful information about how much energy their service consumes, and which data centres that service is performed at.

It’s important information. If you are one of the many paid Claude subscribers who signed up at Rutger Bregman and QuitGPT’s behest, you can can be confident that the dark-background green-font vibe-coded website no one besides you is going to use is being written in a facility that is globally notorious for choking a black community in Memphis.

It is not just air pollution. The entire fleet of existing and planned "Colossus" data centres will emit so much greenhouse gas pollution that, by my reckoning, they’ll likely undo the entire emissions saved by Tesla’s global fleet of electric vehicles. That estimate is already out of date, by the way: a few days ago, it was revealed that xAI has added a whopping 500 megawatts of additional fossil gas turbines to Colossus 2 project, which will increase Colossus 2’s total climate footprint by 65% (from 3.1 to about 5.1 MTCO2-e). These, like most of the others at Colossus 2, are unpermitted, meaning they’re being operated illegally. It’s this second stage that’ll host xAI’s Grok chatbot, meaning this is where all the non-consensual abusive imagery X users love so deeply will be fabricated.

NEW: xAI appears to have added more than 500 MW (!) of unpermitted gas turbines to its Colossus 2 campus in a matter of weeks, according to internal emails between the company and regulators<br>— molly taft (@mollytaft.com) 2026-05-13T00:11:03.102Z

It is amusing to me to note that every Substack blogger putting together posts about how the water issues are "fake" assisted by Claude output will now be specifically doing so using a data centre that faked a promise for a water recycling plant:

"Musk pledged two years ago to build a state-of-the-art water recycling plant in Memphis to guarantee that his xAI servers wouldn’t deplete the city’s groundwater""Now that Musk’s first data center is up and running, construction on the recycling facility has come to an abrupt halt"<br>— Ketan Joshi (@ketanjoshi.co) 2026-05-08T13:38:01.345Z

It was already clear a while ago that Anthropic mostly do not care about climate, environment or air pollution in any material way, even if they care a little bit more about bad PR than Elon Musk does.

In a July 2025 whitepaper, Anthropic explicitly urged the US government to set a target for a minimum amount of new fossil gas generation, to power their energy hunger. These "Notional targets for natural gas...