Northbit started with a single Dell R620 in a cabinet at a colocation facility in 2018. We now own 312 servers across six regions, run by eleven engineers who all picked up a soldering iron before they picked up a keyboard. None of this is resold from anyone else's cloud.
Lukas Maier rents a quarter-rack at a colo facility in Stockholm and buys a refurbished Dell R620. The first customer is a friend with a Postgres database that needs a real home. The R620 is still in the fleet, now running as a backup node in eu-north-1.
The fleet reaches twelve servers and the first hire — Maja Lindqvist — joins to run the NOC. The first incident-response runbook is committed to the internal wiki. The phrase "median nine-minute reply" becomes an internal metric. We have been measuring it ever since.
The first VPS class launches alongside bare-metal. Custom hypervisor stack on KVM, billing-per-second, snapshot system. Account onboarding crosses one thousand. We move into a small warehouse in Halton Lane that still serves as the head office.
First GPU SKU lands — eight L40S boxes at launch, doubled by year-end. The H100 class follows in 2024 after a quiet year of engineering to handle the thermals. The fleet now spans bare-metal, VPS, and GPU compute, with a unified billing and provisioning layer underneath.
Today. Three hundred and twelve servers across six regions, eleven engineers on staff, four mechanics-of-silicon on the workshop floor. Independent and profitable. We have done one pricing change in seven years and it went down. The original R620 is still in the rack and we will keep running it for as long as it boots.
When you call the NOC line, the person who picks up will have racked, cabled, or rebooted one of our servers in the last week. There is no tier-one filter, no offshore call centre, no chatbot. Just engineers with root access.
Bought the R620. Wrote the first version of the orchestration layer in 2018, still maintains it. Picks up the phone Monday mornings and on a rotation overnight.
Joined as employee #1 in 2019. Wrote half the incident-response runbooks. Will tell you, gently, that BGP problems are usually DNS problems. Owns the nine-minute reply metric.
Came over from a hardware-design role at a Brazilian HPC lab in 2023 to bring up our first GPU class. The reason H100 thermals work the way they do. Lives ten minutes from the warehouse.