Infrastructure

Three core sites. Twelve locations. One network.

Our own routing, our own addressing plan, and enough redundancy to lose a site without losing the service. Here is what sits behind the status page.

Core sites

Where the metal lives.

Compute and storage run in three facilities we operate ourselves. Each has independent power, cooling and transit paths before a customer workload is allowed through the door.

London

LON1
Core site

Primary core, split across two halls in separate buildings. The original rack from 2017 is still here, now surrounded by rather better hardware.

Power
N+1, A+B feeds
Cooling
N+1
Transit
2 × 10G
Peering
LINX + 3 private
Floor space
340 m²

Amsterdam

AMS2
Core site

European peering hub with direct routes to London and Frankfurt. Two new racks came online here in October 2026.

Power
N+1, A+B feeds
Cooling
N+1
Transit
2 × 10G
Peering
AMS-IX, NL-IX
Floor space
210 m²

Frankfurt

FRA1
Core site

EU core for data-residency workloads, with separate power paths and a strictly separated management network.

Power
N+1, A+B feeds
Cooling
N+1
Transit
2 × 10G
Peering
DE-CIX
Floor space
260 m²

Network

We own the interesting parts.

We announce our own prefixes, filter with RPKI and keep a live map of every circuit. Traffic takes the path with the best latency, not the cheapest one.

DDoS scrubbing is always on, not an add-on. Volumetric traffic is redirected to a scrubbing cluster and analysed; legitimate flows come back through a clean path. We publish a monthly summary of what we saw.

  • Own ASN and provider-independent addressing
  • RPKI-validated routing with strict prefix filters
  • Always-on DDoS scrubbing at every core site
  • Dual-stack IPv4 and IPv6 on every service
  • Documented BGP communities for customers

Live route summary

Median latency, sampled every 60 seconds

  • LON1 → AMS27 msdirect
  • LON1 → FRA111 msdirect
  • AMS2 → FRA16 msdirect
  • MAN1 → LON15 msedge
  • DUB1 → LON19 msedge

Routing changes are logged and reviewed. Customer-visible changes are announced on the status page.

Locations

Twelve locations, not twelve data centres.

Three core sites host compute and storage. Nine edge PoPs extend private network attachment and regional ingress, without pretending to be full facilities.

Location Code Route from London Role
Manchester, United KingdomMAN15 msNorthern ingress
Dublin, IrelandDUB19 msEU west ingress
Paris, FrancePAR18 msPeering
Amsterdam, NetherlandsAMS27 msCore site
Frankfurt, GermanyFRA111 msCore site
Zurich, SwitzerlandZRH115 msDR capacity
Milan, ItalyMXP119 msSouthern peering
Madrid, SpainMAD124 msIberian ingress
Copenhagen, DenmarkCPH117 msNordic peering
Stockholm, SwedenARN120 msNordic ingress
Vienna, AustriaVIE121 msCentral Europe ingress
London, United KingdomLON1—Core site

Latencies are median round-trip times from LON1, measured in September 2026. Edge PoPs are network attachment points; compute and storage live in the core sites.

The physical layer

Controls you can walk up to.

Cloud pages tend to skip the doors, cables and diesel. We think those are the interesting parts.

Access & physical

Biometric entry, mantraps and 24/7 staff. Cabinets are individually locked, CCTV is retained for 90 days, and every visit is logged.

Power & cooling

N+1 UPS and generators with 48 hours of fuel. Hot-aisle containment, raised floors and a design PUE of 1.25 across the core estate.

Environment

Renewable electricity across all core sites, refrigerant leak monitoring and an annual waste and water review. Boring, measurable, audited.

Change management

Maintenance with a way back.

Customer-impacting work happens Tuesday to Thursday, between 01:00 and 04:00 UTC. We publish the window at least five working days ahead and keep the previous configuration on the shelf until the new one has survived a full business day.

Emergency security work follows a shorter path, but not a quieter one: the status page still gets an entry, and the incident review still gets written.

  • Peer-reviewed change plan and rollback steps
  • Pre-change snapshot of every affected configuration
  • On-call engineer and network engineer both present
  • Customer notice by email and status page
  • Post-change review within two working days

Next planned windows

London core switch firmware

Expected impact: brief failover, no customer action required.

22 Nov 2026
01:00–04:00 UTC

Storage cluster upgrade

Expected impact: none; redundancy reduced during the window.

3 Dec 2026
01:00–02:30 UTC

Full history and updates: System status.

Want the network diagram?

We will send the version with real latencies and the mistakes left in. Useful for architecture reviews, due diligence, or deciding whether we are a fit.