A train can be ordered, built and delivered, and still not carry a single passenger. Before it runs, a safety authority has to authorise it. Before it can run fast and close to the train in front, the line has to be fitted with signalling that can talk to it. And before either matters, the track under it has to be renewed often enough to stay safe. None of these three layers makes the headlines that a large train order does. All three decide whether a railway runs. In May 2025, France’s transport regulator reported that signalling on the core network was still 27.4 years old on average.
The first piece of this chapter followed the money and found it in two places: the makers' long order books and the public schedules that renew a network. This piece asks a different question. Which parts of the railway cannot be replaced quickly, and who holds them? The answer is less about famous brands than about slow layers: signalling that takes decades to roll out, approval that every vehicle must pass, and the people who design, test and renew the system.
A railway can wait for new trains; it cannot run without the people who signal, approve and renew it, and those three layers are slow to replace and short of hands, which is why a candidate who can work in any of them holds a skill the sector cannot buy off the shelf.
This is the second piece of the rail and urban mobility chapter of the CareerOn Industry Atlas. It reads eight documents, each opened and its passage checked before a word was written: a special report of the European Court of Auditors, the latest report of the EU Agency for Railways on safety and interoperability, the annual activity report of the French rail safety authority, an opinion of the French transport regulator on the network's performance contract, a sector skills study, the rail sector contract signed with the French state, and the annual reports of Hitachi and VINCI. Three further documents we wanted, the network manager's own safety report, Colas's annual report and the universal registration document of its parent Bouygues, refused our download and are not used.
Three layers that cannot be rushed
It helps to picture a railway as a stack. At the top are the trains everyone sees. Underneath them sit three layers that are easy to forget. The first is control: the signalling and train protection that tell a driver, or a computer, how far and how fast the train may go. The second is permission: the authorisations without which a vehicle, a line or an operator may not carry passengers. The third is condition: the track, ballast and overhead wires that wear out every day and must be renewed on a cycle.
What these layers share is time. A train fleet can be replaced over a decade. A signalling system is replaced over a generation, one section of line at a time, while trains keep running on it. An authorisation cannot be bought; it has to be earned, file by file, test by test. And track renewal is limited by the hours at night when a line can be closed. Money can speed each of them up only so far. What limits them is skilled people and the calendar.
That is the lens of this piece. It does not rank companies or claim that any single supplier is irreplaceable. None of the eight documents counts suppliers or measures market shares, so this piece does neither. It shows instead where the documents say the slow layers are, how large they are and who is needed to move them.
Signalling is the slowest layer to replace
Europe decided long ago that its railways should share one signalling and train control system, the European Rail Traffic Management System, known as ERTMS. Its train protection part is called ETCS; its radio part is GSM-R. The idea is simple: a train fitted once should be able to cross borders without changing equipment at every frontier. The execution has been anything but simple.
In 2017 the European Court of Auditors put a price on the ambition. The overall cost of deployment, it wrote, "could be up to 80 billion euro for the core network corridors or 190 billion euro for the comprehensive network". That is an estimate, made for a whole continent, and this piece prints it as one. Its point is scale: signalling is not a component, it is a programme measured in decades and in tens of billions.
Signalling is the slowest layer to replace
| What the document prints | Measure | Publisher, date |
|---|---|---|
| ETCS on 15 % of the EU core corridors | Share equipped, end of 2024 | EU Agency for Railways, June 2026 |
| About 22,000 km of lines equipped with ETCS | Lines equipped across Europe, end of 2025 | EU Agency for Railways, June 2026 |
| Signalling equipment aged 27.4 years on average, down from 28 | French core network, 2024 | Autorité de régulation des transports, May 2025 |
| Up to 80 billion euro for the core corridors, 190 billion euro for the whole network | Estimated cost of full deployment | European Court of Auditors, 2017 |
Sources 1 European Court of Auditors · 6 European Union · 7 Autorité de régulation des transports (ART)
The EU Agency for Railways measures how far the programme has come. In its report published in June 2026, it records that by the end of 2025 around 22,000 km of lines were equipped with ETCS. It also records a harder number: by the end of 2024, ETCS covered 15 % of the core corridors, against 57 % for the GSM-R radio. The two figures use different reference dates, and the chart keeps them apart. Read together, they say the same thing. The radio layer is largely in place; the train protection layer, the part that actually controls movement, is still at the beginning.
France shows the other side of the same problem. The system that ERTMS is meant to replace is ageing. In its opinion of 6 May 2025 on the network's performance contract, the French transport regulator notes that the average age of signalling equipment on the core network went from 28 years to 27.4 years in 2024, better than the contract's objective. The direction is right. The level is still close to three decades, which means much of the signalling a young engineer will work on was installed before they were born, and much of it will be replaced during their career.
Makers know how scarce this capability is. Hitachi's integrated report for 2024 states that on May 31, 2024 it completed the acquisition of the GTS signalling business, which it describes as a way to "acquire market-leading signaling and rail control technology, and add 9,000 highly skilled colleagues". That is a company describing its own strategy, and it proves nothing about market shares. What it shows is revealing enough: a large industrial group chose to buy signalling know-how and the people who hold it rather than build them from scratch.
Every train passes an approval gate
The second slow layer is permission. A new or modified vehicle, a new line and a new operator all need an authorisation before they may carry passengers. In France, the authority that issues most of them is the Établissement public de sécurité ferroviaire, the EPSF. Across the EU, the Agency for Railways authorises vehicles that run in several countries.
Every train passes an approval gate, and it is busier
| What the document prints | Measure | Publisher, date |
|---|---|---|
| 138 opinions and decisions, against 115 in 2023 (+20 %) | Authorisations issued in France, 2024 | EPSF, July 2025 |
| About 2,000 applications and nearly 21,000 vehicles authorised | Vehicle authorisations handled by the EU agency, 2025 | EU Agency for Railways, June 2026 |
Sources 2 Établissement public de sécurité ferroviaire (EPSF) · 6 European Union
Both gates are getting busier. The EPSF's activity report for 2024 records a sharp rise in the number of authorisations issued in 2024, with 138 opinions and decisions against 115 in 2023, an increase of 20 %. The EU agency, for its part, reports that the vehicle authorisations it handled rose in 2025, with around 2,000 applications and nearly 21,000 vehicles authorised. The two counts measure different procedures, so they are never added. Each on its own points the same way: more files, more tests, more decisions.
Behind each decision stands work that few candidates picture when they think of a railway career. Someone has to assemble the evidence that a train is safe: the design analysis, the type tests, the integration tests with the signalling, the demonstration that every requirement has been met. Someone else has to assess that evidence independently. Someone at the authority has to decide. Each of those roles needs people who understand both the engineering and the rules, and that combination is rare.
There is one claim we did not make. It is common to hear that approvals in rail take too long. Neither document measures how long an authorisation takes, so this piece does not say whether they are slow. It says only what the documents print: the gate is universal, and the queue in front of it is growing.
The shortage is people, on the track and at the drawing board
We began this research expecting to find that the scarcest input was the crews who renew track at night. The documents tell a broader story.
The shortage is people, on the track and at the drawing board
| What the document prints | Measure | Publisher, date |
|---|---|---|
| 7,430 to 10,120 recruitments over 2024-2030, design and engineering first | Projection, rolling stock manufacturing only | EDEC ferroviaire, April 2025 |
| About 70,000 recruits over 2024-2030, 40,000 of them by 2027 | Target the sector set itself | Rail sector contract, June 2024 |
| Contracts to renew 675 km of lines and 800 km of track and ballast; close to 1,750 km revamped to date | One works contractor’s own account (ETF) | VINCI, March 2025 |
Sources 3 Observatoire des compétences industries (EDEC ferroviaire) · 4 Conseil national de l’industrie, Comité stratégique de filière ferroviaire · 8 VINCI SA
The renewal work is real and large. VINCI's universal registration document for 2024 describes its specialist subsidiary ETF, which "carries out renovation and maintenance operations on France's entire rail network and has revamped close to 1,750 km of track to date". In 2024, ETF won two contracts to renew 675 km of lines and 800 km of track and ballast. That is one contractor describing its own order book, and it proves nothing about the whole market. It does show the kind of volume that has to be delivered, mostly at night, by crews and engineers who know how to close a line safely and hand it back on time.
But the sector's own numbers point further. The skills study published in April 2025 under the rail industry's employment and skills programme projects that rolling stock manufacturing alone will need between 7,430 and 10,120 recruitments over 2024 to 2030, or between 1,050 and 1,450 a year. And in every scenario, it says, the largest need is in the design and engineering families of jobs, ahead of production. The rail sector contract signed in June 2024 goes further still, with a target to recruit about 70,000 people across the sector over 2024 to 2030, 40,000 of them before the contract ends in 2027.
Those are a projection and a target, and the chart labels them as such. A target is what the sector wants, not what it has achieved. Taken together with the renewal volumes, they reframe the question. The railway is not short of one trade. It is short of people across the slow layers: engineers who design and integrate signalling, specialists who test and certify, and crews and managers who renew the network.
What the documents do not say
A careful reader should know where the evidence stops. None of the eight documents counts the suppliers of signalling or measures their shares, so this piece does not say that the market is concentrated, even though that is often assumed. None measures how long an authorisation takes, so we say nothing about delays. The skills projection covers rolling stock manufacturing only; it does not cover track works or operations, and we do not stretch it to them.
Three documents that would have deepened the picture could not be used. The network manager's own safety report, Colas's annual report and Bouygues's registration document all refused our download, so nothing they say is reported here. Where a company appears, it appears through its own filing, and we say so each time.
What this means for your career
The lesson of the slow layers is that scarcity lasts. A skill in a fast layer can be replaced by the next product cycle. A skill in signalling, approval or renewal stays valuable for as long as the programme runs, and these programmes run for decades.
If you are drawn to permission, learn to build and read a safety case. The work sits between a finished train and its authorisation to carry passengers, and it asks for engineering judgement and patience with rules in equal measure. The CareerOn rail quality and certification engineer simulation rehearses exactly that: assembling evidence, handling a non-conformity and defending a file.
If you are drawn to control, learn how a system is tested and brought into service. Signalling and train protection only matter once they work on a real line with real trains. The metro testing and commissioning coordinator simulation rehearses the sequence of tests, the interfaces between teams and the decisions that let a new system go live.
If you are drawn to condition, learn how a fleet or a network is kept available. Renewal and maintenance are daily disciplines in which planning, safety and reliability decide the result. The tramway fleet maintenance engineer simulation rehearses that work: reading failure data, planning interventions and keeping vehicles in service.
A last piece of advice, drawn from the documents themselves. In an interview, show that you understand why your future employer's work is slow. A candidate who can explain why signalling takes decades, why every vehicle needs an authorisation and why renewal is limited by the night sounds like someone who already works in the railway.
How we read these documents
Every figure in this piece appears in a passage of one of the eight documents, and each passage was checked against the document itself before it was quoted. Estimates, projections and targets are labelled as such wherever they appear. Figures that measure different things, or the same thing at different dates, are shown side by side and never added or averaged. Company documents are read as what they are: a company's own account of its work. The full list of documents, with their dates and what each cannot prove, is in the sources below.