Engineering
An engineering company, not a purchasing desk
Isengard Computing was founded in 2014 on the view that a host that does not write its own software ends up living with someone else’s. Sixty people, forty-one of them engineers, and an R&D team whose code runs in production.
- Founded
- 2014
- People
- 60
- Engineers
- 41
- Data centres
- 4
- In R&D
- 17
- Functions subcontracted
- 0
Timeline
Twelve years, four sites, three pieces of software
Each step corresponds to a component we chose to build rather than buy. That is what explains the shape of the company today.
2014
Founding and first site
Gaël Marchand and Inès Kerdraon start the company with seven engineers and a 200 m² suite in the north of France. The initial goal: bare metal delivered in under five minutes, at a time when two days was the standard.
2016
In-house orchestrator
First version of the orchestrator. Provisioning a bare server drops from 22 minutes to 3. That code has been in production ever since, rewritten twice.
2017
Second site and backbone
ORT-2 opens in Île-de-France and the private backbone goes in. Inter-site replication becomes viable, which is what makes a 30-minute recovery objective achievable.
2019
DDoS mitigation of our own
After a 1.1 Tbit/s attack that was absorbed but badly handled, the R&D team builds its own detection and scrubbing chain. Time to neutralisation drops below 30 seconds.
2020
Sovereign hypervisor
Our hypervisor, built in house, goes into production. That is the point at which sovereignty stops being a clause and becomes a technical property.
2023
Fourth site and GPU compute
ORT-4 opens in the south for recovery and immutable archives. In parallel, the first liquid-cooled GPU nodes enter service at ORT-3.
2026
Sixty people, nothing subcontracted
NOC, SOC, network engineering, operations and R&D are all in house. No on-call duty, no monitoring and no development is handed to a third party.
Culture
Four engineering rules
These are not values on a corridor wall: they are the criteria that settle our technical arguments.
- Strength
- We size for the worst case, not the average one. Capacity that is never used is the ordinary cost of infrastructure that does not fall over.
- Security
- Every layer is designed assuming the one before it has given way. A mechanism that has not been tested in real conditions is treated as absent.
- Performance
- We publish percentiles. An average hides precisely what bothers your users, and we decline to communicate on that basis.
- Mastery
- What we do not control, we build or we do without. That is our R&D team’s reason for existing, and its only prioritisation criterion.
R&D
Seventeen people, three components in production
Our R&D team does not build prototypes: it maintains code that carries our clients’ production. Three components, rewritten rather than bought.
Hypervisor
Virtualisation kernel and control plane. Seven engineers. In production since 2020, across the whole cloud fleet.
Orchestrator and API
Provisioning, lifecycle, quotas, Terraform provider. Six engineers. This is the component that holds the 112-second delivery time.
Mitigation chain
Flow-based detection, diversion, scrubbing. Four engineers. Handles roughly 1,800 events a month.
Leadership
Four leads, four remits
Gaël Marchand
Co-founder · Chief executive
Network architecture and carrier relations. Still writes the routing policies.
Inès Kerdraon
Co-founder · Chief technology officer
Runs R&D and decides what the company builds rather than buys.
Paul Vasseur
Head of SOC
Detection, incident response and the mitigation chain. Runs the six-monthly failover test.
Ludivine Arnoux
Head of operations
NOC, capacity and fleet lifecycle across the four sites. Owns the two-hour replacement time.
How the sixty are split: 17 in R&D, 14 in operations, 10 in the NOC, 8 in the SOC, 6 in pre-sales and client engineering, 5 in administration.
Your technical counterparts are named in the contract and introduced at the kick-off meeting.
The sizing is done by the people who run it
Pre-sales is handled by six engineers who also run production. They commit to what they know they can hold, and say so when a request falls outside our remit.