5 June 26
On Green AI
Contrary to the normal progression of things, the computer that I built a year ago in May has increased in value by about $1150. This is because of the extraordinary demand being placed on supplies of memory and storage components of all sorts by the buildout of AI data centers — most spectacularly in my case, the 64 GB of DDR5 RAM which I purchased in May of 2025 for $165 is now priced at $915. With my workstation, I am continuing to play around with local AI models. The quality of these local models has increased dramatically over the past year, which is leading many developers to explore how they can be used in preference to the huge models from providers in the cloud such as Anthropic, OpenAI, and Google.
One could make the argument that running AI models locally is more sustainable than invoking the resources of huge server farms every time one calls upon AI. But are there other threads that might constitute “green AI”? Sharon Stein has a good piece about this topic on resilience.org. She identifies three visions for green AI. The first of these technical greening. This means running AI with fewer resources, for instance by using smaller models that demand less energy and materials. The second of these is ecological intervention. This refers to developing AI applications that help us understand and ameliorate environmental change, for instance by optimizing agricultural resource use. The third vision is relational reorientation. This focuses on questioning the assumptions around nature, intelligence, and relationships that are built into contemporary AI. A lot of this vision comes from Indigenous perspectives: one interesting research program in this vein is called Abundant Intelligences (also see an open access paper titled Abundant intelligences: placing AI within Indigenous knowledge frameworks).
On a more amusing note, here is a literally hand-cranked solution to the cost and energy use of contemporary AI systems.
Previous: Buglets On The March Next: