This is a legacy post, aimed at our members and discussing information that was widely available at the time. We leave posts that have been superseded in place as a historical record.
For the latest information on AI and the environment, we recommend the United Nations University’s Environmental Cost of AI Energy Use (June 2026)
Michael Webb
In our AI and the environment blog series, we’ve covered taking a responsible approach, provided an update on the current landscape, and looked at putting the numbers into perspective. In this final blog of the series, we’re highlighting a few trends we’ve seen emerging lately. These range from greener energy solutions for AI infrastructure, to evolving policies, regulations and exciting technical innovations and collaborations. While we can’t cover everything, we’ve selected a handful we think you’ll find particularly interesting.
Cleaner energy and manufacturing
Recently, there has been a growing push to use renewable energy to meet the rising demands of AI. This is a positive development and could lead to a more sustainable future. As data centres require 24/7 power, it’s possible that the most sustainable option could be a combination of renewables and nuclear, but this is up for debate.
Last year (2024), Microsoft admitted their 2030 net zero goals may not be met due to their AI strategy. Since then, they announced they were building their first data centre with wood in an attempt to slash carbon emissions. Reducing the amount of steel and concrete, which are well known sources of carbon emissions, in the manufacturing process could be a great way forward. Although this was published in late October (2024), and no further updates appear to be publicly available, it still offers an interesting glimpse into what the future may look like.
Towards the end of last year (December 2024), The Tony Blair Institute for Global Change published a paper on how the UK can power AI and lead the clean-energy transition by revitalising nuclear energy. The paper mentions how Google, Amazon, Microsoft and Oracle have all committed to investing in new nuclear projects to resolve the ongoing energy problem. In the UK specifically, an agreement was reached last year (February 2024) to install four small modular reactors (SMRs) in North Tees, so it’s possible that nuclear could play a role in our national energy demand going forward.
More recently, in February 2025, Apple announced they will spend more than $500 billion in the U.S. over the next four years. This will go towards both manufacturing and data centres in the US to support Apple Intelligence. Their press release claims teams have designed the servers to be ‘incredibly energy efficient’ and are set to reduce the energy demands of Apple data centres, which already run on 100% renewable energy.
Technology
In January this year (2025), Microsoft released an AI-powered sustainability roadmap for a net zero future, which outlined how AI is transforming climate action, energy efficiency, and environmental protection. The report highlights AI’s potential to accelerate net zero targets, through its implementation in renewable energy projects, biodiversity monitoring, smart data centres, battery efficiency and global climate risk assessment.
In February (2025), Google released a study on the full lifecycle emissions of its AI accelerator chips. Over two generations, they made their chips three times better at using energy while producing fewer carbon emissions. They also used a new way to measure their emissions called ‘Compute Carbon Intensity’ which helps to track and reduce the environmental impact of AI chips. Soon after, in April (2025), Google announced Ironwood, the first Google TPU for the age of inference. This AI hardware was designed specifically for AI inference and the chip has more than double energy efficiency compared to Trillium TPU, which was announced in May 2024.
Reports
The National Engineering Policy Centre report, released in February 2025, calls for mandatory reporting and stricter sustainability requirements to ensure AI growth aligns with the UK’s climate goals. They proposed five foundational steps to begin the process towards environmentally sustainable AI. These focus on expanding environmental reporting, addressing information gaps, setting sustainability requirements for data centres, reconsidering data management practices, and increasing government investment in sustainable AI. This aligns with the growing call for transparent data from companies. Greater data availability leads to better insights, which will enable policymakers to make informed, actionable decisions and hopefully drive consumer and provider behaviour.
In April (2025), the International Energy Agency released a special report on Energy and AI, which is noted as being the ‘most comprehensive, data-driven global analysis to date on the growing connections between energy and AI’. The report covers projections for how much energy AI could consume over the next decade, along with analysis around the uptake of AI and what this could mean for future energy security, emissions, innovation and affordability. Interestingly, as pointed out by Boris Gamazaychikov (Head of AI Sustainability at Salesforce), the word ‘uncertain’ is mentioned 61 times in the report, drawing attention to the ongoing challenges in this area. The report suggests that data centres are reversing the long standing trend of flat or declining electricity demand in advanced economies, which was previously driven by efficiency improvements. It also highlights how AI is becoming a powerful tool in accelerating innovation in energy technologies, like batteries and solar photovoltaics. That being said, it states that the energy sector hasn’t yet harnessed AI’s full potential, partly due to a shortage of skilled professionals and other challenges. For more details, we’d recommend reading the full report.
Initiatives
Collaboration is crucial to innovation and is something we really champion in the AI team and within Jisc more widely. At the AI Action Summit in Paris (February 2025), Hugging Face, Salesforce, Cohere and CMU launched an AI energy score rating system, which helps AI developers and users to evaluate, identify and compare energy consumption of AI models. You can look specifically at uses, including image generation and text generation, which is a great way of increasing awareness and perhaps offers a standardised framework that could be adopted going forward. This project is still ongoing and has successfully tested hundreds of open source models. The project has since released a call to action in April (2025) calling for customers of closed AI models to apply pressure, offering examples of tender questions and procurement contract requirements.
The Coalition for Sustainable AI was also announced at the AI Action Summit, which ‘aims at building a global community of stakeholders willing to contribute to initiatives for aligning AI development with global sustainability goals and fostering responsible AI that supports the environmental policies.’ This was initiated by France, in collaboration with the UN Environment Programme and the International Telecommunications Union with support of international organisations.
More recently, Hugging Face released an interactive demo that estimates the energy consumption of different models across different tasks and modalities. It displays real GPU energy consumption (in Wh or Joules) for each message you send and adds real-world equivalents like phone battery percentage, seconds of microwave use and LED light hours. It’s really interesting to see how much energy our AI conversations consume and in future, could be something we see across all models.
Rebound effects
We wanted to end with a note on potential rebound effects. We mentioned Jevons Paradox in a previous blog post, and this is certainly becoming increasingly relevant. When technology becomes more efficient, it often leads to increased use, which can offset or even reverse the intended resource savings. This is important when it comes to AI, where efficiency gains may encourage greater adoption and therefore demand.
Building more efficient data centres is certainly a step in the right direction, but as AI becomes increasingly embedded into our daily lives, and governments shape the future of our energy demand around AI’s needs, we need to address the potential rebound effects. If rising demand is left unchecked, efficiency gains alone may not be enough. It’s important to stay mindful of the scale of this growth and reflect on what it may mean for long term sustainability.
As awareness of AI’s environmental impact continues to grow, we are hopeful about the direction of progress, and excited to see what the next innovations may be.
Find out more by visiting our Artificial Intelligence page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
Join our AI in Education communities to stay up to date and engage with other members.
Get in touch with the team directly at AI@jisc.ac.uk