Dark Factory Initiative
Material transport at night, inspection cells, packaging lines: unattended hours exist in European plants today. How far they spread depends less on the robots than on process stability, master data and product design.
Global industrial robot installations have more than doubled in ten years. Cheaper vision sensors, better grippers and models that learn a task rather than following a fixed program now reach work that resisted automation: order picking in intralogistics, visual inspection, packaging and sub-assembly. Horváth works with manufacturers to find where unattended operation pays first, and what plan-ning, data and product design have to deliver before it does.
Asia sets the hardware cost curve, and that question is largely settled. Whether a plant's processes, data and product design can run without human improvisation decides who profits.
growth in global robot installations within 10 years
Behind the number
Capacity doubled, productivity did not.
Most of that installed base still sits behind fences, serves one purpose and needs an engineer for every change. Engineering
hours are the scarce resource, not robots.
of new robot deployments take place in Asia
What follows
The learning curve sits where the deployments are. European plants can buy identical hardware, though not the integration experience that comes with it.
of global installations in China alone
Sourcing implication
Chinese volume increasingly defines automation standards, interfaces and spare-part ecosystems. Sourcing turns into a dependency question for the board.
expected growth of China's humanoid output in 2026
Business case impact
Suppliers are building capacity ahead of proven demand. Unit prices will fall faster than most business cases assume, so calculate with the 2029 price rather than today's quote.
Horváth Research tracks 218 humanoid robotics companies worldwide. More than half have their headquarters in China, one in five in the United States – and only around one in ten in all of Europe combined.
The industry forming around autonomous production is young: three out of four players started in 2020 or later, and nearly four in five run on venture capital. Companies are taking their positions now.
Supplier risk
Three in four of your future suppliers have never been through a downturn. Screen for survivability, not demo quality.
Contract implication
Roadmaps follow funding rounds, not your ramp-up plan. Insist on source-code escrow and data exit clauses.
Our expectation
Two different bets: many cheap builders versus few deeply funded ones. Consolidation is the base case for 2027/28.
Europe builds few of these machines and buys many. Access to the technology, the data and the ecosystems forming around them becomes a board question rather than a procurement one
Europe is under-represented as a builder and over-represented as a buyer. That is not automatically a weakness – but it means Europe's value will be created in application, integration and process know-how, not in the machine itself. Whoever treats humanoids as a procurement topic will import someone else's productivity advantage at full price.
Source: Horváth Research 2026, based on 218 tracked humanoid robotics companies. Deal sizes: PitchBook Q1 2026.
Most companies ask when humanoid robots will be good enough. The technology will arrive on schedule. A factory designed to run without human improvisation will not arrive with it.
The all-or-nothing dark factory belongs in press photos. Autonomy spreads through a plant in fragments: night shifts, material transport, inspection, packaging. That makes lights-out a sequencing question – which process moves next, and what has to hold steady before it runs alone.
A humanoid exists to cope with a world built for humans. Every part a robot struggles to grip goes back to a design decision someone made years ago. The cheapest path to autonomous production usually runs through R&D and design-for-automation rather than the robot budget.
The setter becomes a process owner, maintenance shifts from repair to prediction, and new roles appear for robot fleet management and master data ownership. A lights-out plan built around headcount reduction stalls early. One built around process stability delivers the savings as well.
Every operating hour of a deployed robot produces training data, and that advantage compounds. A company that starts three years from now competes against systems which have already learned the task.
With one player in ten and rising cost pressure, Europe will not out-build Asia on hardware. The hard part of autonomous production sits elsewhere: in the interplay of planning, quality, maintenance and supply chain. European plants already coordinate exactly that across sites, and claiming the ground deliberately matters more than building the machines.
The argument for waiting holds up on paper: units get cheaper and more capable every year. The buyer who waits also skips the learning curve and enters a market where competitors have absorbed the integration cost already. Early movers pay for competence, not for hardware.
Our position in one paragraph:
Operating models decide the dark factory, not robotics labs. Autonomy pays where the product design supports it, where processes hold without correction, and where a machine can trust the planning data. The companies that win the next decade of manufacturing will not own the most robots. Their factories will simply stop needing anyone to rescue them.
The question is no longer whether autonomous production will reshape manufacturing, but which companies will move first to capture its full potential.
The same technology arrives in every sector and lands differently in each one.
Our industry has been supplying the components of the dark factory to everyone else – while assembling its own products by hand.
Industrial goods and high tech sit on both sides of this market. We build the drives, sensors, controls and optics that make autonomous production possible, and we are among the last to apply them to our own high-mix, low-volume assembly. The reasons are familiar: variant complexity, small batch sizes, engineering-to-order. Every argument for why automation does not pay in our own plants is an argument a competitor is currently disproving.
Humanoids and adaptive AI change the economics of variance. Classical automation needed a business case per variant, while a system that learns a task rather than following a fixed program pushes the break-even point down. For machinery and electronics manufacturers, automation turns from a volume question into a capability question, and that brings mid-sized specialists into play alongside the large OEMs.
Where we would start: Rank your product portfolio by design-for-automation readiness before evaluating a single robot. The bottleneck is almost never the machine. Statements reflect the personal assessment of the respective partner. Wenzel and Brauchle as supplied; the Sauter statement is a draft pending approval.
Every workaround your shopfloor performs today becomes tomorrow's stoppage. Stan-dardization remains key.
Operations leaders consistently underestimate the importance of informal correction: the setter who knows which machine drifts, the planner who quietly re-sequences the shift, the supervisor who finds the missing part. Autonomy is therefore the most honest audit of operational maturity: it converts every data gap and every unstable process directly into downtime.
The foundation of the dark factory remains codified expert knowledge, embedded in standards. Master data quality, realistic routings, predictive maintenance and a scheduling logic a machine can trust are the prerequisites. Those investments pay for themselves long before the first humanoid arrives, and plants that run to standard often sit closer to it than they expected.
The first measure we would take: Measure the share of production hours that already run without human intervention. That single number tells you how far you really are.
Automotive OEMs and suppliers do not have another ten years for basic digitalization. Competitors' plants are creating the cost gap per vehicle today.
Cost position, volume volatility and competitors that scale faster and price lower press on the European automotive industry at the same time. Autonomous production is not optional in that setting: it is one of the few levers that works on all three pressures at once. Lights-out capacity cuts cost per unit and makes capacity elastic, which counts in a market where volume forecasts have stopped being reliable.
Automotive sets the hardest conditions for this technology: safety-critical parts, full traceability, model cycles that run for years, and highly optimized brownfield plants that nobody can simply rebuild. The sector will not go dark in the press-shop sense. Unattended operation will start in logistics, intralogistics, quality inspection and sub-assembly, and suppliers will feel the competitive consequences fastest.
Our recommendation: Identify the dark factory potential across your own production network and your supplier ecosystem. Transform the whole value chain rather than single sites.
The pattern behind the four signals
Volume, commercial model, component depth and data all moved within the same twelve months. Industries look like this at their beginning, while the ground rules are still open
The study tracks 218 companies worldwide and records location, funding, maturity and technical readiness. It also names the profit pools we expect to form, and the ones we expect to disappoint.
