Executive brief
Most career writing about defence photonics describes the sector and then guesses at the jobs. This dossier does the opposite: it reads only what Exosens publishes — its products, its dated announcements and its own openable job postings — and derives the competences from there. Four families come out of that record, and none of them is what a general engineering CV expects. First, vacuum process and photocathode work, because the group's defensible asset is a tube whose fourth generation is marketed at a Figure of Merit of 1800 minimum and whose plants are Brive-la-Gaillarde and Roden. Second, detection-physics engineering across five acquired specialisms — infrared, hyperspectral, ultraviolet, ion and electron, radiation — bought as six companies in Belgium, Canada, Germany, Israel, the United Kingdom and France between December 2022 and November 2024. Third, electronics design close to the sensor, evidenced directly by a posting for an FPGA design engineer at the Palaiseau cooled-infrared unit. Fourth, the industrial-scale-up functions — IT infrastructure, quality, programme management — that a group adding a first US plant and a €140m innovation financing has to staff, evidenced by a posting for a cloud and infrastructure engineer reporting to a global IT infrastructure manager. What we will not print: any salary band, because no posting we opened states one; any per-site headcount, because none is published anywhere; and any claim that a named skill is scarce in the local labour market, because no openable source measures that.
I. Why the competence map has to be read from the plant, not from the sector
Sector-level reasoning about defence would predict a certain kind of employer: systems engineers, integration managers, cleared programme staff, a lot of documentation. That prediction is not wrong about primes. It is wrong about Exosens, and understanding why is the whole value of this dossier for anyone deciding where to spend five years of a career.
The group's economic position rests on a component that other people's programmes are qualified around. An image-intensifier tube is a vacuum device: a photocathode converts scarce photons into electrons, a microchannel plate multiplies them, a phosphor screen returns an image to the eye, and the whole assembly must hold its performance over a service life inside equipment carried by soldiers. Its performance is expressed as a single figure of merit, and the company's own leaflet for its 4G tube advertises FOM 1800 minimum, while its product page claims the technology is "today the benchmark in all major European Land Forces programs."
What that means for work is specific. In a business whose moat is a qualified physical process, the decisive competence is not architectural taste. It is the ability to hold a process inside limits, to know why it drifts, and to prove that it has not. That is a laboratory-and-line discipline, and it is learned in the plant. The two plants named in the record are Brive-la-Gaillarde — the company states it has been in the town since 1937 and on the current site since 1962 — and Roden in the north of the Netherlands. A third, Exosens Scientific USA at Sturbridge, Massachusetts, sits behind the March 2025 announcement of a €20m two-year capacity investment in Europe and the United States, presented in the international release as the group's first US night-vision production investment.
II. Family one: vacuum, photocathodes and the discipline of a qualified process
We do not have a published Exosens job description for a photocathode engineer, and we will not invent one. What the record does support is the shape of the demand. The company sells a tube whose performance is guaranteed as a minimum, not as a typical value; it sells it into land-forces programmes; and it has committed capital to producing more of it in two continents. Each of those three facts imposes something on the people doing the work.
A guaranteed minimum turns statistics into an engineering obligation: yield, screening and the physics of the tail of the distribution stop being quality-department topics and become design topics. Programme customers turn change control into a first-class activity: a process improvement that raises the mean but perturbs the qualified configuration can be worth less than nothing. And a capacity expansion turns tacit know-how into a transfer problem: whatever is in the heads and hands of the people in Brive and Roden has to be made explicit enough to reproduce elsewhere. The competence that spans all three is the one this industry rarely advertises — the ability to write down why a process works, in terms another site can execute and a customer can accept.
III. Family two: detection physics, acquired five ways
Between December 2022 and November 2024 the group bought Xenics in Leuven (infrared, described by the acquirer as "over 65 people"), Telops in Quebec (hyperspectral and infrared imaging), ProxiVision in Bensheim (ultraviolet detection), El-Mul in Rehovot (ion and electron detection), Centronic in Croydon (radiation detection) and Noxant in Palaiseau (high-performance cooled infrared cameras). Read as a labour-market fact rather than as a corporate-development timeline, that list says something unusual: the group now contains five distinct detection physics, in six countries, under one commercial roof.
For an engineer, the practical consequence is that the internal mobility surface is physics-shaped rather than product-shaped. Someone who understands noise, dark current, detector cooling and calibration can move between an infrared camera line, a hyperspectral imager and a radiation detector without leaving the group, because those objects differ in wavelength and readout, not in method. That is a real career argument, and it is derived here from the acquisition record alone. It is not evidenced by any published mobility data, and we say so rather than dressing the inference as a measurement.
The same list carries a caution worth printing. Six acquisitions in twenty-three months, none of them with a disclosed price, means a group in the middle of integration. Integration is where reporting lines, tooling and process standards are still being decided — energising for people who like writing the rules, expensive for people who need them already written.
IV. Family three: electronics next to the sensor — the one demand a posting states outright
Here the record speaks in the company's own recruitment voice, and it is worth quoting the shape of it precisely. Exosens' careers portal serves individually openable postings, not a JavaScript shell: an FPGA design engineer position, dated on the page to 6 August 2026, France, explicitly for the team of Noxant, based at Palaiseau (91). That single posting closes an inference that would otherwise be speculative: the group hires digital-design engineers into the cooled-infrared unit it acquired in November 2024, at that unit's own site.
Why FPGA work sits at the centre of a detection business is not obvious to outsiders, so it is worth stating. A detector's output becomes an image through readout timing, non-uniformity correction, integration control and latency management — work that lands in programmable logic because it has to be deterministic and close to the sensor. The competence family that follows is therefore: digital design and verification, sensor readout chains, and the calibration mathematics that turns raw counts into a radiometrically meaningful image. No programming language, toolchain or framework is named here beyond what the posting itself names, because inventing a stack is exactly the failure mode this dossier exists to avoid.
V. Family four: the functions a scale-up has to staff, and the evidence for them
A second openable posting — a Cloud & Infrastructure Engineer (Azure), dated on the page to 5 June 2026, France, "reporting to the Global IT Infrastructure Manager" — is more informative than its title suggests. The existence of a global IT infrastructure function, hiring in France, tells you that the six acquisitions are being consolidated onto shared foundations rather than run as a federation of independent shops. For a candidate, that is the difference between joining an established platform and joining a migration.
Two further scale-up demands are implied by dated company events rather than by postings, and are presented as reasoning, not as vacancies: the €140m European Investment Bank financing announced 25 June 2026 to support innovation in the European defence and security industry, which is a research-programme management obligation as much as a balance-sheet event; and the December 2025 order of 100,000 MIKRON binoculars carrying 200,000 tubes for about €500m, which is an industrial planning obligation of a different order from anything in the group's prior history. We note, as the companion dossier does at length, that three publishers name three different counterparties for that order, and we adjudicate none of them.
VI. The ladder: what separates a good candidate from a decisive one
The following ladder is an argument from the burden of proof that qualified components impose on their makers. It is not a published framework, and no certification is named.
- Measure honestly. Produce a number with its uncertainty and its method attached. In a business selling a guaranteed minimum, a measurement without an error budget is not evidence.
- Bound the process. State the limits inside which a result holds — temperature, vacuum, materials batch, operator step — in writing, so that a deviation is detectable rather than debatable.
- Own the qualification. Carry a change through the customer's acceptance logic, not merely through the internal lab. This is the rung where engineers in programme-qualified businesses either become indispensable or stay junior for a decade.
- Transfer it. Make the know-how reproducible at another site. This rung is what a capacity expansion in two continents actually buys, and it is the rarest of the four.
The transferability of these four rungs beyond photonics — into semiconductor qualification, medical devices, nuclear instrumentation and metrology — is stated here as reasoning from the shared discipline of proving a process to a demanding acceptor. It is explicitly not evidenced by any published mobility data.
VII. What the record does not publish, and what we therefore refuse to say
No salary figure appears in this dossier, because neither posting we opened states one and no aggregator publishes a bracket we could attribute to the company; we do not substitute a regional average. No group headcount is asserted: the only figure in circulation, 1,800, comes from a single regional trade parenthesis dated 3 July 2026 that we could not re-open, and no Exosens document we retrieved states it. No per-site headcount split exists in any source, so nothing here calls Brive, Roden, Sturbridge or Palaiseau large or small. No standing tube production volume is published — the 200,000 tubes in the record belong to one contract, not to a capacity statement. No named apprenticeship or graduate programme with intake numbers is published on the company's site; the only education-adjacent evidence we found is participation by the Brive team in the third edition of a school-outreach day organised with the IUT GEII de Brive on 26 March 2026, reported by the university's Brive campus, which states that 157 pupils from named local lycées and collèges attended. That is an outreach fact, not a hiring pipeline, and it is printed as such.
VIII. How to approach this employer
The honest advice from this record is narrow and therefore useful. If you are drawn to defence because of the systems, this is probably the wrong company: the value here is upstream of systems, in a vacuum process and in five detection physics. If you are drawn to the physics, it is one of a small number of places in Europe where the process itself — not the integration around it — is the product, and where a first US line and an EIB-financed innovation programme mean the process is being extended rather than merely maintained. Ask about the qualification chain, ask which site owns which process, and ask what integration stage the acquired units are in. Those three questions are answerable from the inside and unanswerable from the outside, which is exactly why they are the ones to ask.