Top universities are using AI, but quiet on the fine print

For a while, I had a running joke. Every time an organisation or person I loved announced they were pivoting to using AI, I would post about it like the loss of battlements in a war. “Margaret Atwood has fallen”. “ProPublica has fallen”. “DuckDuckGo has fallen”, etc. At some point I stopped because friends adopting AI became so common the joke stopped working.

It has been crushing to see so many fall prey to the hype and hollow promise of generative AI, but one particularly consequential loss has been in higher education. It has been an ugly few years, with executives and higher-ups eagerly cramming generative AI down students’ throats despite the well-documented, serious effects of mandated cheating software. 

What interests me most is the combination of universities’ self-image as incubators and world-leading labs of climate action and innovation, paired with the obscene energy bloat and fossil fuel hunger of generative AI. It’s something a little worse than mere hypocrisy, more in line with betrayal and treachery.

Related Article Block Placeholder Article ID: 1268083

The University of New South Wales (UNSW), for instance, describes itself as a “global leader and record-breaker in solar PV engineering”. Climate is one of UNSW’s “grand challenges”, it houses a world-leading climate change research centre, and it brags about its sustainability rankings every year. UNSW Press was savvy enough to publish my 2020 book, Windfall, so they must surely know a thing or two about climate change.

Yet this time last year, UNSW expanded a 2024 pilot with OpenAI into a full enterprise agreement, buying 10,000 ChatGPT Edu licences for all its permanent and fixed-term staff. When UNSW made this announcement, it was striking a deal with OpenAI while the AI company was already planning some of the world’s largest fossil gas power stations to run generative AI. Pre-empting the might grate with some staff, UNSW added:

“UNSW recognises the importance of addressing the environmental impacts of emerging technologies and AI use. In response to feedback from our community, the university will purchase carbon offsets verified under a reputable standard to mitigate the emissions associated with the ChatGPT Edu agreement. As with other purchased goods and services, carbon emissions associated with the rollout will be included in UNSW’s emission reduction targets and annual reporting.”

This is really significant, at a global scale. While there has been plenty of community pushback to data centres and much focus on government regulation, there hasn’t been enough scrutiny at the level of purportedly clean higher education institutions doing dirty deals with methane-slurping chatslop peddlers. 

Generative AI is extremely unusual software. Performing basic, day-to-day tasks is far more energy-intensive than other software systems like email, spreadsheets or video calling. Querying a database representing human knowledge and then probabilistically reproducing a slice of it using repetitive language-creating loops is brute-force content production, and it burns through energy and hardware like Bitcoin mining. This is why we’re seeing city- and country-sized fossil gas plants being proposed to power generative AI, whereas the previous decades of digital work required a more reasonable level of data centre growth. 

Related Article Block Placeholder Article ID: 1266086

Weird, then, that UNSW’s latest sustainability report does not seem to actually include the carbon emissions of ChatGPT’s integration, as promised in September last year. The calendar-year 2025 report, updated only recently, neither notes nor specifies the inclusion of generative AI in the section where disclosures of third-party services are included (“scope 3”). 

I emailed UNSW to ask if its new AI use was included but aggregated, or not included at all, and didn’t receive a reply. 

NOISE, UNSW’s independent student publication, reported that scrutiny of the program’s environmental impacts is only growing. Aside from the total lack of disclosures, the use of carbon offsets to wrongly claim those emissions are “neutralised” is a key area of concern among staff and students. 

We don’t know what UNSW’s AI emissions are yet, but it’s shocking how heavily UNSW relies on various types of offsets, and how that masks the fact its actual emissions have barely budged at all in recent years: 

“Offsets” here are mostly large-scale generation certificates, or LGCs, from clean energy generators. These have been criticised as being of low quality, just as traditional offsets have. As I’ve covered here before, the big fight in corporate accountability right now surrounds how electricity specifically is offset, rather than the controversial “traditional” carbon offsets you’ve likely heard about (where any emissions reduction can be neutralised by people claiming to have not cut down trees, among other methods).

I don’t know if this ratio will last for long, as we could see a swing back towards cheaper, dodgier traditional carbon offsets. A friend just sent me this amazing new offering from Greenfleet, an Australian carbon offset company: 

Peruse the website for these special AI-dedicated carbon offsets and find all the tropes now the norm in the offsetting industry: namely a feeble reminder, frequently stated but rarely obeyed, to always reduce your emissions first. We still don’t know if or how UNSW is reporting AI pollution — but it sure won’t be “reducing first and offsetting later” when it comes to its AI climate impacts. 

There are a few of these AI offset offerings popping up now around the world, each offering moral absolution for the decision to switch from lean and efficient work to bloated, fossil-fuelled reproduction of other people’s work. 

One recent JWS Research poll found the top concerns Australians have for data centres include environmental impact, power prices and water use. I’ve seen this result reproduced consistently in different regions around the world. It is no surprise the institutions most eagerly doing dirty deals with US tech companies are engaging in the same wrongdoing as those companies: hiding critical data, and when forced to disclose, using dodgy offsets to mask a failure to act. I have still not found a single company or institution, anywhere in the world, that discloses a total emissions estimate of its adoption of generative AI. 

I am pretty sure we’re past peak hype for AI and data centres now, but that doesn’t make things easier. This is where the messy process of actually holding institutions and corporations to account matters most, and UNSW should face real pressure to at least do the bare minimum and come clean about the climate impact of its shift to digital bloat.