In June 2025, ArcelorMittal walked away from 1.3 billion euros of German public money rather than build the hydrogen-based plants that money was meant to fund. Nobody had forced the decision. The company simply concluded that the economics did not work at the electricity prices it was looking at. For anyone trying to understand how heavy industry is being rewired, that refusal is more instructive than any of the announcements that preceded it.
Key takeaways
- 542,000 industrial robots were installed worldwide in 2024, more than double the figure of a decade earlier (IFR).
- Western Europe reached 267 robots per 10,000 manufacturing employees in 2024, against 204 in North America.
- ArcelorMittal cancelled two German hydrogen steel projects in June 2025 and gave up the subsidies attached.
- Sweden’s Stegra is still building, with first customer deliveries now expected in 2027.
Three versions of the same story, and what the record shows
The phrase “old industry meets new tech” gets applied to situations that have almost nothing in common. Separating them makes the picture legible.
| The claim | What the record shows | Status in 2026 |
|---|---|---|
| Factories are filling up with robots | Global operational stock reached 4.66 million units in 2024, up 9% in a year | Happening, unevenly by region |
| Steelmaking is switching to hydrogen | One large greenfield plant under construction, several announced projects cancelled | Contested, energy-price dependent |
| Software is replacing industrial know-how | Deployments still require the process engineers who understand the plant | Overstated |
The control layer has already landed
Automation is the part of the story where the data is unambiguous. The International Federation of Robotics counted 542,000 industrial robots installed in 2024, the fourth consecutive year above half a million units, taking the worldwide operational stock to 4.66 million. Density figures tell you where that capital went: 267 robots per 10,000 manufacturing employees in Western Europe, 204 in North America, 131 across Asia as a whole, with the Republic of Korea far ahead of everyone at 1,220 and Germany at 449.
Two things in those numbers matter more than the headline. First, Asia absorbed 74% of new installations in 2024, so the centre of gravity of industrial automation is not where the reporting on it is written. Second, density is a ratio, and a rising ratio can reflect either more robots or fewer employees. In the European case it has mostly reflected more robots against a broadly flat manufacturing headcount, which is a different phenomenon from the one usually described as job replacement.
The process layer keeps stalling
Changing how a material is made is a different order of problem. Direct reduced iron using hydrogen instead of coking coal is technically demonstrated, and the economics remain brutal. ArcelorMittal made that explicit in June 2025 when it abandoned direct reduction projects at Bremen and Eisenhüttenstadt and relinquished a subsidy package worth 1.3 billion euros, citing high electricity prices, weak steel demand and hydrogen costs that did not clear. The company’s European management said plainly that even with the support, viability was not assured.
Sweden’s Stegra, the company formerly called H2 Green Steel, is the counter-example still standing. Its Boden site combines around 700 MW of electrolyser capacity with direct reduction and steelmaking on one plot, funded to roughly 6.5 billion euros and designed for about 2.5 million tonnes of steel a year. Production has been targeted for 2026, with first customer deliveries now expected in 2027. That timeline has moved, which is normal for a first-of-a-kind plant and worth stating rather than smoothing over.
The variable separating the two cases is not ambition. It is the price of clean electricity at the meter, which northern Sweden has and the Ruhr does not. Anyone weighing whether industry can decarbonise without going broke is really asking a question about power markets wearing the costume of a question about steel.
Where the silicon actually sits
The “steel to silicon” framing invites a misreading, as though chips were replacing metal. What has happened is narrower and more useful: heavy plants have acquired a nervous system. Vibration and thermal sensors on rotating equipment, controllers close enough to the machine to act inside a production cycle, and models trained on a specific plant’s history rather than on a generic asset class.
The gains from that layer are real and unspectacular. Fewer unplanned stoppages, tighter quality windows, less scrap. They accumulate quietly and they do not photograph well, which is part of why the coverage prefers the hydrogen story. In our reading, the plants making steady progress are almost always the ones that instrumented an existing line properly before commissioning anything described as transformational.
Four checks before you believe a transformation claim
- Is it a retrofit or a greenfield? Retrofits carry the constraints of the existing plant and rarely deliver the modelled figure.
- Where does the electricity come from, and at what price? For anything electrochemical, this single input decides the outcome.
- Has the subsidy been drawn down or merely awarded? The German case shows an awarded package can be handed back.
- Is there a shipped tonne or a signed offtake? Announced capacity and delivered capacity differ by years in this sector.
Questions this usually raises
Does automation destroy industrial employment? The evidence is more mixed than either camp allows. Robot density has risen in Europe against a fairly stable manufacturing headcount, while the composition of the work has shifted towards maintenance, programming and quality roles. What we would avoid is treating a density statistic as a headcount forecast.
Is green steel dead after the cancellations? No, but the delivery date has moved right and the geography has narrowed. Projects sited where clean power is cheap and abundant are proceeding. Projects that assumed cheap hydrogen would arrive on schedule have mostly been paused or dropped.
Why do these plants take so long? An integrated steelworks is a chemical process, an energy asset and a logistics hub at once, each with its own permitting path. A first-of-a-kind facility has no reference plant to copy, so every subsystem is being commissioned for the first time by people who have not done it before.
What should a manufacturer outside steel take from this? That the sequencing matters. Instrumentation and control pay back on a timescale a finance director recognises. Process change depends on inputs you do not control, so it needs a different kind of business case and a different tolerance for delay.
The rewiring is genuine. It is just proceeding at two speeds, and the mistake most often made is quoting the pace of the fast layer while describing the ambitions of the slow one.
Curious where the computing actually happens?
The control layer described here depends on processing close to the machine rather than in a distant data centre.
Sources: International Federation of Robotics, World Robotics 2025, installation and density figures for 2024; ArcelorMittal statement on the cancellation of its Bremen and Eisenhuttenstadt direct reduction projects, June 2025, and associated German federal and state funding of 1.3 billion euros; Stegra company disclosures on the Boden plant, including capacity, funding and delivery timing. Robot density figures are ratios per 10,000 manufacturing employees and are not employment forecasts. Updated August 2026.
