Executive brief
CORYS is one of the strangest industrial assets in the Grenoble basin: a software house of a few hundred people whose products sit inside the licensing envelope of nuclear power stations, metro signalling systems and refineries across four continents. Its statutory identity is unambiguous: CORYS SAS, SIREN 413 851 924, registered at 44 rue des Berges in Grenoble. Almost everything a market analyst would want next is contradictory or absent. The company register dates its creation to 1997; the company's own communication describes more than thirty-five years of simulation, which points to the late 1980s. It states its own shareholders, Framatome 50%, EDF 25%, IFP Training 25%, with a share capital of €4,666,361, but no effective date for that split. Its headcount is published three times as 291, as a 200–249 bracket, and as over 380. No audited revenue figure exists in any openable source, and we refuse to print the estimate that circulates in its place. What that record does support is a precise reading of the mechanism: this is not a software business with industrial customers, it is a regulated engineering business whose deliverable happens to be code.
I. The mechanism: simulation as a licensing artefact, not a product
The instinct when reading a simulation company is to treat it as software with a niche. That reading fails here, and the failure is instructive. A full-scope nuclear plant simulator is not sold into a procurement department the way a CAD seat is. It exists because the operating regime of the plant requires operators to be trained and periodically requalified against a replica of their own control room, and because the fidelity of that replica is itself an object of scrutiny. The customer is therefore not buying features. The customer is buying a defensible correspondence between a mathematical model and a physical installation that a regulator, an insurer and an operating utility all have standing to interrogate.
That single structural fact explains the company's shape better than any market forecast. It explains why the shareholder register published by Framatome, the reactor vendor at 50%, the operator EDF at 25%, and the refining-and-petrochemical training arm IFP Training at 25%, reads less like a cap table and more like a list of the parties who need the models to be right. It explains why the product names are platforms rather than applications: ALICES Plus for nuclear, INDISS PLUS for process industries, and a rail line built around Ultra Light Simulators and interfaces to OCTYS-class CBTC signalling. And it explains the durability of the customer list, because replacing a qualified simulator means re-establishing that correspondence from scratch.
It also explains the opacity. A company whose output is embedded in someone else's safety case has structural reasons not to publish, and its owners have structural reasons not to break it out. The absence of numbers here is not sloppiness. It is a consequence of the business model, which is exactly why an analyst should refuse to fill the gap with estimates.
50 · 25 · 25
the published shareholder split, Framatome, EDF, IFP Training, for a company with share capital of €4,666,361
The company publishes the percentages and the capital, but no effective date for the split, so no history is inferred from it here.
II. What the registers actually say
The statutory spine is openable and consistent across the French company-information services. CORYS is a société par actions simplifiée, SIREN 413 851 924, establishment SIRET 413 851 924 00022, VAT FR16413851924, APE 6202A (computer consultancy activities), registered with the Grenoble commercial court at 44 rue des Berges, 38000 Grenoble.
| Attribute | Value | Where it is published |
|---|---|---|
| Legal form | SAS | French company registers (Annuaire des Entreprises, Société.com, Pappers) |
| SIREN / SIRET | 413 851 924 / 413 851 924 00022 | Same |
| Registered address | 44 rue des Berges, 38000 Grenoble | Same |
| Statutory creation date | 1 September 1997 or 24 September 1997 | Register services disagree by 23 days |
| Narrative founding | Late 1980s ("more than 35 years") | Company and encyclopaedic sources |
| Shareholders | Framatome 50% · EDF 25% · IFP Training 25% | The company’s own about page; corroborated by a tertiary encyclopaedia |
| Share capital | €4,666,361 | The company’s legal notice |
| Headcount | 291 · 200–249 · "over 380" | Three publishers, three answers |
| Revenue | Not published | No openable audited or statutory figure |
The two creation dates are a small discrepancy with a large lesson. Neither is wrong in its own frame: a company can be constituted on one date and registered on another, and a legal entity created in 1997 can be the vehicle of an activity that started earlier: as a division, a joint venture or a laboratory programme. What no publisher does is state which frame it is using. So we print 1997 for the legal person and the late-1980s narrative for the activity, and we decline to write a founding year.
III. The headcount spread, and why averaging it would be an error
Three numbers circulate. A professional network shows 291 associated profiles. A register service publishes a workforce bracket of 200–249 for the 2022 reference year. The company's own site states that it employs over 380 engineers and scientists worldwide. Those are not three attempts at one quantity; they are three quantities. The network figure counts self-declared affiliations, including alumni-adjacent noise and excluding anyone without a profile. The register bracket counts declared salaried employment at a legal entity for a past fiscal year. The recruitment claim counts a professional population and is not bounded to the French entity.
An average of 291, 224 and 380 would be a number with no referent: no date, no entity boundary, no counting rule. It would also propagate: divided into a revenue estimate it would manufacture a productivity figure, and compared across years it would manufacture growth. We print the spread and treat the workforce as being in the low-to-mid hundreds, with the explicit finding that no publisher states its counting basis.
Three headcounts, three counting rules
| Figure | What it counts | Reference date | Publisher |
|---|---|---|---|
| 291 profiles | Self-declared affiliations on a professional network | Consulted 2026 | |
| 200–249 bracket | Declared salaried employment at the legal entity | 2022 reference year | Pappers |
| over 380 | Engineers and scientists worldwide, recruitment claim | Undated | CORYS |
Sources 3 Pappers · 5 LinkedIn Economic Graph · 8 CORYS
IV. Revenue: an absence of measurement, not a range
There is no audited or statutory revenue figure for CORYS in any source we could open. A widely visible business-network profile displays a revenue estimate, labelled by the platform itself as an estimate rather than a filing. We do not print it, and the reason matters beyond this company. An estimate that sits at roughly one million dollars for an organisation of two to three hundred engineers is not a plausible figure for an engineering business of that size; publishing it, even hedged, would give a fabricated order of magnitude the authority of a citation. The honest finding is that the company's turnover is unmeasured in public. Readers who need it should look for it where it would legally exist, in the consolidated accounts of the shareholders, and note that none of them appears to break out this subsidiary.
V. The project record: nine dated deliveries in nine months
Paradoxically, the most verifiable material about this company is not financial but operational. Simulator work is announced, dated and described in technical detail, because the customers are public operators and licensed utilities. Between February and September 2026 the company published the following, and read together it is a better disclosure than most annual reports offer.
| Date | Customer | What was delivered |
|---|---|---|
| 27 Feb 2026 | N2G2V consortium (UMN, Grenoble INP-UGA, Université Grenoble Alpes) | Cloud C EPR simulator selected as a teaching tool under a France 2030 call, in a consortium targeting a rise from 130 to 218 five-year nuclear graduates a year by 2028 |
| Mar 2026 | EDF: COCASE fleet renovation | Factory acceptance for the PP4 (1300 MW) VD4 digital-control-room stage, after completed 900 MW (CP1, CP2) and 1300 MW (P4) stages; delivery to DIPDE Marseille in April and to the Cattenom full-scope simulator in July 2026 |
| 20 Mar 2026 | Epic Games | Certified Unreal Engine Service Partner after three years of R&D from early 2023 by a team of more than 15 engineers |
| Apr 2026 | EDF, with REEL | Fuel-handling simulator using real plant PLCs, installed beside operating teams at the Flamanville 3 EPR site, replicating reactor building, fuel building and transfer facilities |
| 23 Apr 2026 | Berliner Verkehrsbetriebe (BVG) | Complete tram training suite: eight replica desk simulators with full cabs plus Ultra-Light Simulators |
| 6 May 2026 | Framatome: Ugine site | INDISS PLUS dynamic simulator for a Vacuum Arc Remelting furnace modelling a full zirconium ingot melt, phased nominal-then-fault-scenarios |
| 17 Jun 2026 | GTS: Elizabeth line, for Transport for London | Twelve Class 345 Aventra Ultra-Light Simulators replicating CBTC, ETCS Level 2 and AWS/TPWS across 35 miles of route, in classrooms of six around one instructor station |
| 22 Jun 2026 | Pacific National, Australia | Sixteen Ultra-Light Simulators across Greta (NSW) and Sarina (QLD), including a geo-specific reproduction of the Hunter Valley heavy-haul corridor |
| 6 Jul 2026 | RATP Dev: Caen | Site acceptance 22–26 June, under five months after kick-off: ten ultra-light simulators, one instructor station, one scenario-preparation station, ahead of a July 2027 Normandy service takeover |
| 13 Jul 2026 | Direct Rail Services, United Kingdom | Class 68 ETCS simulators commissioned, one replica plus eight Ultra-Light Simulators, with speed attributed to factory validation before shipment |
Four mechanisms are visible in that table, and none of them is a marketing claim.
Mandated change creates the backlog. Six of the ten items exist because a rule changed, not because a customer wanted new software: ETCS rollout in the United Kingdom, a ten-year inspection of a 1300 MW digital control room (the PP4 VD4 stage accepted in March 2026 above), fuel-handling qualification at an EPR. Regulated renewal is a demand curve that recessions do not flatten.
Delivery risk is moved into the factory. Twice in three months the company put its schedule performance in the headline: a factory acceptance test on schedule, then a site acceptance under five months from kick-off. In a trade where the customer cannot start training until acceptance passes, punctuality is the product feature.
The image pipeline became a moat in 2023. The Epic Games certification is dated and quantified: three years, more than fifteen engineers, a rebuilt image-generation system. That is the clearest capital-allocation signal the company has ever published, and it explains why the InnoTrans demonstrator leads with eye tracking and AI rather than with fidelity claims.
The skills shortage is now a customer. The N2G2V item is the strategic tell. A simulator vendor being written into a national programme to lift nuclear graduate output from 130 to 218 a year is no longer selling only to operators; it is selling to the training system that supplies them: in its own city.
VI. The company contradicts itself on its own scale, and that is worth printing
The sharpest contradictions in this file are not between rival publishers. They are between pages of the company's own website, consulted on the same day.
| Claim | One page says | Another page says |
|---|---|---|
| Simulators in service | Transport: "approximately 2000 simulators in-service" | Transport, same page: "1,500+ simulators in service worldwide"; Process: "more than 2600 simulators" |
| Country footprint | Transport: present in more than 30 countries | Process: simulators delivered in more than 60 countries |
| Years of experience | About us: more than 35 years | Power: more than 30 years; Transport: 35 years |
None of these is necessarily false. A fleet count for rail is not a fleet count for process; countries where simulators were delivered is not the same set as countries with a presence; and 2,000 in-service against 1,500+ worldwide may be counting rail differently from the whole book. But no page states its scope, which means none of the numbers is usable as published. We print them as they stand and adopt none. The one fleet figure that does carry a scope is the British one, over 200 driving simulators already in service across the UK network, published in the Elizabeth line announcement, which proves that scoped counting is possible whenever the company chooses to do it. The nuclear line separately states that 80 engineers and experts work on power-plant simulators: the only quantified sub-population in the entire record, and a useful floor: it is inconsistent with the smallest headcount publication once the rail and process lines are added. A staff gathering reported on 23 June 2026 as “almost 200 members of staff” at the Château de Sassenage is a fifth data point of a fifth kind, attendance, not employment, and is printed as such.
Two verifiable quality signals do survive: the company publishes ISO 9001, ISO 14001 and ISO 27001 certifications, the third of which matters commercially because customer plant data is itself sensitive; and it states the application of Framatome's ethical code, which corroborates the shareholder relationship from a second direction.
Read as a portfolio, this is a hedge that most software companies cannot construct: three regulated verticals whose investment cycles are driven by different forces, nuclear licensing and new build, urban transit capital plans, and refining turnarounds, with the same modelling and validation core underneath. That is the strategic argument for the asset, and it does not depend on a single financial figure.
VII. What we refuse to publish
Naming the gaps is part of the analysis, not an apology for it.
- Any revenue, EBITDA, backlog or contract value. None is published; the circulating estimate is a platform artefact and is excluded.
- Any single founding year. The 1997 statutory dates and the late-1980s narrative describe different objects.
- Any effective date for the 50/25/25 split. The company publishes the percentages, not the date. A tertiary encyclopaedia narrates a 1997 Tractebel and TechnicAtome entry, a 2009 Tractebel exit, an Areva 66% / EDF 34% stage and a February 2015 renaming; we report that as a circulating narrative, not as a cap-table history, because no statutory ownership document was opened.
- Any headcount as one number, and therefore any growth rate, attrition rate or revenue per employee.
- Any site list beyond the four the company publishes. The company names Grenoble plus CORYS Inc. (Jacksonville and Houston), CORYS India Simulation Systems in Pune and CORYS Simulation Middle East in Nicosia; a third-party listing adding China is not corroborated and is not printed as a location.
- Any salary band. No published listing carries one.
- Any market share. No openable source measures it.
- Any single fleet or country count. The company publishes 2000, 1,500+ and 2600 simulators, and both 30+ and 60+ countries, without scope; we adopt none of them.
- Any total headcount inferred from the 80 nuclear engineers. It is a floor for one line, not a company total.
VIII. The conclusion an investor or an engineer should draw
The temptation with a company this quiet is to conclude that there is nothing to see. The opposite is true: the silence is the finding. CORYS is a small Grenoble software organisation that has been made structurally necessary to three regulated industries, owned by the counterparties who most need its models to be defensible, and therefore insulated from the pricing pressure that governs ordinary enterprise software. Its risk is not competitive displacement; it is dependence on the capital cycles of nuclear, transit and refining, and on shareholders who could reorganise it without a press release. Anyone reasoning about this asset should reason about those cycles and that governance, not about a revenue number that nobody has published.