We design and build computing hardware.
Compute is being built at two scales. There is the very large — campuses measured in hundreds of megawatts, gated on transmission capacity and years of permitting — and there is the very small, which mostly means someone’s laptop.
The interesting range is the one in between: tens to low hundreds of kilowatts, inside a building that already has the power, close to the data and the people who need it. Large enough to serve an organisation. Small enough to install rather than develop.
That range is under-served in Europe in particular, and it is where we build.
The three lines share a discipline rather than a chassis. Hold the power and thermal budget. Measure what the system actually does rather than what the datasheet predicts. Ship the software that makes the hardware useful on the day it arrives.
That discipline comes from embedded work, where there is nowhere to hide. When the heap is seven hundred and sixty-four kilobytes and the frame deadline is fixed, you size arenas from instrumented measurement or the product fails in the field. Applying the same habit to a rack is unusual only because racks have historically had enough headroom to make guessing survivable.
A protocol company and a hardware company, kept apart on purpose.
Tenzro is a network for inference, training and model distribution. It is a protocol, and protocol companies that start selling boxes tend to do both things badly — supply, certification, repair logistics and unit economics all break before the network effects arrive.
So the two are separate. Machine Core Systems builds and sells the hardware and supports it. Tenzro is the network our systems join. The integration is deep because the boundary is clear.