This observation is not an impression. It is a repeated finding, across several years, across several vocational training centres, across different cohorts. Engineering classes: close to 50/50 gender balance. Developer training: the same. Systems & Networking: two women out of twenty, consistently. Sometimes zero.

The consistency of this figure across multiple years and multiple institutions rules out coincidence. It is not an atypical cohort. It is a pattern. And a pattern that repeats under the same conditions points to structural causes, not random ones.

The contrast with other technical disciplines is striking and is itself informative. If the issue were "women do not go into technical fields", the deficit should be visible across all technical training. It is not. It is concentrated on infrastructure, systems and networking — and on the physical terrain associated with it: the datacentre.

The first spontaneous explanation one hears — and is tempted to offer — is the physical one. The datacentre is a hot, noisy, physically demanding environment. A single intervention in a server room covers roughly 7,500 steps. Racks are heavy. Cables resist. There is something in this work that resembles a building site more than an office.

This explanation has an apparent logic. But it collapses against a simple, documented counter-example.

When construction machinery could not reach the villages of the Moroccan Atlas, it was women who carried the concrete blocks on their backs to build their village mosques. Physical resilience is not a genetically masculine trait. It is contextual, cultural, motivational. It is deployed where circumstances call for it.

If the physical explanation does not hold, three structural mechanisms — each distinct, cumulative in their effects — can explain the observed imbalance. None of these mechanisms implies malicious intent on the part of individuals. They are system-level effects.

In "Anatomy of the Loss", training was identified as the central lever for reconquering infrastructure competence — against the dissolution of hard limits and the cognitive lock-in of the sector. The same logic applies here: if the gender imbalance is decided at the training stage, then training is where the levers for correction are found.

The shortage of women among SysAdmins and SREs is not decided at the point of hiring. It is decided at the point of orientation, and then consolidated during training. A young woman who does not walk into a Systems & Networking classroom will never have the opportunity to demonstrate that she can excel in this field. Discrimination through image precedes discrimination through competence.

The false perception of physical difficulty, described in section 2, acts as an invisible filter at the entrance to training. It is never written down. It is never stated explicitly. It is conveyed by the image of the room, by the instructor's language, by the composition of the cohort — and by the absence, within that cohort, of women who could have demonstrated through their presence that this perception is false.

The systems and network infrastructure sector suffers from a shortage of qualified profiles that is documented, persistent and structural. Positions remain unfilled. Teams are chronically understaffed. The load per SRE engineer increases. On-call schedules grow denser. Burnout sets in.

In this context, effectively excluding half the working population from the talent pool is not a neutral choice. It is a choice that actively worsens an already existing shortage. A recruiter who does not think to recruit women, a training centre that does not think to attract women, a sector that does not think to retain them: each link in this chain halves the available talent pool. The market suffers directly.

There is also a quality argument. Diversity of profiles within a technical team produces diversity of approaches to problems. A datacentre seen through different perspectives is a datacentre whose blind spots are more often identified. Diversity is not a symbolic objective — it is a measurable operational advantage in terms of system resilience and diagnostic quality.

The levers that follow are not grand statements of intent. They are precise actions, achievable within an ordinary training centre, without exceptional budget, by instructors who decide to think about this problem rather than not thinking about it.

Infrastructure is the foundation of our digital ecosystem. There is no reason for it to be a private club. Not for ideological reasons. For reasons of competence, talent pool, and the long-term survival of a sector that needs every available talent.