Why we run compute the way we do
"Sustainable" gets used loosely in this industry, usually to paper over an operation that isn't. We'd rather make a smaller, more honest argument than a bigger one we can't back up. Here's the actual case for how we run compute: no offsets, no certifications, no round numbers.
Hardware lifecycle extension
Every machine running here is secondhand. GPUs and ASICs that would otherwise be sitting in a drawer, sold for scrap, or landfilled get bought, repaired, tuned, and put back into service instead. This matters more than it sounds like it should: manufacturing a new GPU carries an embodied carbon cost (the mining, refining, fabrication, and shipping that go into it before it's ever plugged in) that no amount of efficient operation afterward can recover. The greenest server is the one that already exists. Buying used and extending its working life is a straightforward way to avoid adding to that cost.
Heat as a product, not a waste stream
Compute converts electricity to heat at essentially 100% efficiency. That's just thermodynamics, not a design choice. Most operations treat that heat as something to get rid of: exhaust it, vent it, spend more energy cooling it away. In a Minnesota winter, we don't have to. Heat that a machine produces while computing displaces heat we'd otherwise buy from a furnace. A machine that warms a room and runs a workload at the same time is doing two jobs with one input, for roughly half the year.
Right-sizing over maxing out
Hardware here is tuned to a power target, not run flat out because it can be. Undervolting and autotuning each card or unit means giving up a small fraction of raw output in exchange for a large fraction of power draw, a trade most operations don't bother making because it takes more setup and more attention than just running everything at stock settings. We think the trade is worth it. It costs us some peak performance and buys back a meaningful amount of headroom on every machine, every day it runs.
Small, distributed, local
There's no dedicated facility behind any of this, no new construction, and no new grid interconnect. It runs on existing space, existing power, and existing buildings, scaled to what they can actually support. That keeps the operation small by necessity, not as a marketing choice, and it's part of why the heat-reuse argument above works at all: the machines are close to where the heat is useful, not in a warehouse three states away.
What we're not claiming
We're not carbon neutral. We're not net zero. We don't run on 100% renewable power, and we haven't had any of this audited or certified by a third party. We don't have a tonnage or percentage figure to hand you, because we haven't measured one and we're not going to make one up. What we do have is a straightforward argument: reuse hardware instead of replacing it, and use the heat instead of wasting it. That argument doesn't need inflating, and inflating it is the one thing that would undermine it.