We'd rather own fewer racks than oversell more of them.
SiliconCore is an early-stage compute infrastructure startup based in Kathmandu, Nepal. We build GPU cloud and bare-metal infrastructure for teams training and serving AI in production.
From one rack to a fleet
First rack online
Eight nodes, one region, and a lot of cable management. The first external training run finished without a single dropped checkpoint.
InfiniBand fabric deployed
Moved off commodity networking entirely. NVLink inside the node, NDR InfiniBand between nodes — the fabric a real training run needs.
Bare metal, three regions
Added dedicated bare-metal nodes alongside GPU instances, and brought US West and EU Central online.
12,000+ GPUs provisioned
Crossed six live regions. Still a small team — every node added gets the same hands-on attention the first eight did.
Principles we build with
Hardware you can actually see
No reselling capacity we don't operate. If a node is listed, we know exactly which rack it's in.
Small team, direct answers
A hardware issue gets routed to someone who understands racks and fabric, not a script reading from a template.
Built from Kathmandu, for anywhere
Our team sits in Nepal. Our GPUs sit in data centers across six regions — proximity to us was never the point.
Want to see the fleet for yourself?
We're happy to walk through the exact hardware behind an instance before you commit to a run.
Get in touch