Someone asked me the other day about whether the current demand shock around RAM could be fixed by recycling. Their reasoning was sound enough from a layperson’s perspective: most households have a couple of old computers and a stack of old smart phones just gathering dust, and if you add up the gigs of RAM in these devices, you’re probably looking at hundreds per home. Can’t all that be stripped from the original devices and plugged into data centres? Unfortunately, things aren’t that simple, because not all RAM is created equally. High-speed RAM like DDR5 is what matters for serious AI workloads, and older consumer RAM isn’t really usable for cutting-edge applications. And I suspect the economics of aggregating scattered consumer devices would be brutal compared to buying decommissioned server equipment in bulk. In any case, the data centre industry is already recycling hardware components. It’s just that the recycling is happening in the other direction—that is, components that used to be at the cutting edge are being reassigned to less demanding tasks. Microsoft’s circular centres are hitting 90% component reuse. Google has resold 44 million hardware components into secondary markets since 2015. Hewlett Packard Enterprise processes about 3 million units annually with roughly 90% reuse rates. There’s clearly a model for harvesting and redeploying compute hardware. I guess the general principle, then, is that recycling of IT components really has to follow a path where the component’s new life is a less demanding one. Still, the question is worth asking. Is there a business—or even a policy angle—around better recycling and redeployment of recent consumer hardware? Could incentivizing device recycling help feed supply and reduce bottlenecks on innovation somewhere else in the ecosystem?
I would say some that previously retired servers are making a come back as "AI Servers" right now where you can put a couple of GPUs and run some local AI models pretty effectively. Check out the Lenovo P620 or HP Z6 G4 as examples, basically older machines that can access multiple GPUs and run Linux with older memory and CPU...
Great question Scott. You could always strip the materials down to their component metals and have them reprocessed into new things. Urban Mining can be more lucrative than old-fashioned mining because you already know the yields of the critical minerals in the products you're harvesting it from. Mine Canada’s E-Waste for Metals - Macleans.ca
Interesting insights and good question Scott. I wonder if there's a real opportunity for innovation — whether policy or business practice — in investing toward efficiency and optimization of the hardware platforms that already exist, extending their useful life in the applications they were built for, rather than chasing more raw supply to feed ever-growing demand. It might smooth the curve out a little, and those same innovations may trickle down to previous generations of gear and legacy platforms.