Inspired by a bridge in the sky, built for the rest of us.
Before MemWaxing had a name, it was Membridge. Before that, just a problem worth solving.
Every galaxy has its own gravity. So does every machine's memory.
43,000 light-years from here, two galaxies sit close enough that one's gravity strips stars from the other, pulling them into a continuous bridge of light across the void between them. Astronomers call it the Magellanic Bridge. It's not a metaphor chosen after the fact. It's the whole idea, found already sitting in the sky.
Memory works the same way once you stop treating each machine as its own separate system. Pulled across the gap, formed into one continuous thing. The first name for this project was Membridge, before Magellanic Conversion, before MemWaxing. The idea never changed. Only the name got easier to say out loud.
A data center running at partial GPU utilization isn't inefficiency. It's the standard shape of the industry today.
DDR3, DDR4, DDR5, GDDR6. Technically incompatible by design. A machine with idle RAM can't natively help a GPU running low on memory three feet away, even on the same network. Solving that gap has simply never been the industry's priority when selling more hardware has always been the easier answer.
MemWaxing pools what's already there instead. Idle memory on machines that are already running, already drawing power, becomes usable capacity rather than unused overhead. For a single homelab that's a few extra gigabytes of headroom. At data center scale, it's a genuinely different way to think about utilization before reaching for more racks.
"The infrastructure to do this already exists. The question has always been whether it's being used."
Infinity Ember Systems, on why the fabric matters as much as the fixA 2014 iMac answered.
32GB of DDR3 that had no business contributing to modern AI inference sent back a live health check anyway, alongside two other machines on the mesh. A binary cross-compiled from a completely different operating system, running on hardware one package manager quietly classifies as barely supported. It answered.
Getting there wasn't clean. Five separate theories about why a 32B model kept failing to load were tested and ruled out one at a time with hard data, not guesses, before the sixth turned out to be the real answer. Every dead end got written down. That's the actual engineering behind the pitch.
Hardware that already exists is hardware that doesn't need replacing.
MemWaxing is aiming to reduce energy consumption and carbon footprint by extending the useful life of hardware that already exists, at scale, instead of demanding more of it be manufactured. Existing hardware, doing more before it gets replaced.