Dark Factory
Last reviewed
May 10, 2026
Sources
17 citations
Review status
Source-backed
Revision
v2 · 2,492 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 10, 2026
Sources
17 citations
Review status
Source-backed
Revision
v2 · 2,492 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: artificial intelligence terms
A dark factory, also called a lights-out factory, is a manufacturing or logistics facility that runs production with little or no human presence on the floor. The label refers to the idea that machines do not need lighting, climate control, or shift breaks the way people do, so a fully automated plant can in principle operate in the dark, around the clock. In practice, very few sites are completely unmanned, but the term is widely used for facilities where robots, artificial intelligence systems, and the industrial internet of things carry out the bulk of production while a small crew handles maintenance, supervision, and quality assurance from a control room.[1][2]
Dark factories are often discussed as a flagship outcome of Industry 4.0. They sit at the intersection of robotics, machine learning, computer vision, and 5G networking. They are most common in semiconductor packaging, electronics assembly, and robot building, where parts are small, tolerances are tight, and the production environment can be tightly controlled. As of 2025, well-known examples include FANUC's robot plant in Oshino, Japan, Philips's electric razor factory in Drachten, Netherlands, Xiaomi's Changping smart factory in Beijing, and a growing list of Foxconn sites in China and India.[1][3][4][5]
"Lights-out manufacturing" describes a methodology in which a factory's main production processes run without on-site human labor. "Dark factory" is a more dramatic synonym; the two are usually used interchangeably, although some writers reserve "dark" for plants that literally turn off the lights to save energy. A looser label, smart factory, covers any heavily digitized plant, whether or not humans are still on the line.[1][2]
A fully automated factory is not necessarily a dark factory. Many highly automated plants still keep workers nearby for setup, inspection, and exception handling. The defining feature of a dark factory is that humans are not required during normal production. They might step in for maintenance, retooling, or audits, but the line keeps moving when they are not there.[2]
The imaginative roots of the dark factory go back to mid-twentieth-century science fiction. Philip K. Dick's 1955 novelette Autofac, first published in Galaxy Science Fiction, describes self-replicating robotic factories that keep producing goods after a devastating war, ignoring whether the surviving humans still want them. It is widely cited as one of the earliest portrayals of fully autonomous, self-sustaining production.[6]
The industrial concept took shape in the 1980s. Fujitsu Fanuc opened a robotics plant at Yamanashi, Japan in January 1981 in which robots and computer numerical control machine tools produced parts for other robots with limited human oversight. That plant is generally treated as the first practical lights-out facility.[1][2] Around the same time, General Motors chairman Roger B. Smith pushed an aggressive automation strategy in the United States. Smith spoke openly of "lights-out" car factories where the only employees would supervise robots, and during the 1980s GM spent on the order of $90 billion on automation, including a 1981 joint venture with Fujitsu Fanuc called GMF Robotics. The program was famously troubled: robots painted each other instead of cars, welded doors shut, and were ripped out of several plants soon after installation. Industry historians often cite the GM experience as evidence that automation alone, without process and management changes, does not deliver lights-out production.[7] After the 1985 Plaza Accord pushed the yen up and squeezed Japanese exporters, robotics offered a way to keep production at home while controlling cost, and Japanese firms in electronics and machinery became the early laboratory for unmanned shifts.[2]
A modern dark factory is less a single technology than a stack. The main layers are summarized below.
| Layer | Role | Examples |
|---|---|---|
| Industrial robots and cobots | Pick, place, weld, screw, assemble, palletize | FANUC arms, ABB, KUKA, Adept Technology |
| Autonomous mobile robots (AMRs) | Move parts and finished goods between cells | AGV and AMR fleets coordinated by warehouse software |
| Machine vision and AI inspection | Detect defects, read labels, guide grippers | Deep learning vision used by Philips, Xiaomi, Foxconn |
| Industrial internet of things | Connect machines and feed sensor data to control software | Networked PLCs, edge gateways |
| 5G and time-sensitive networking | Low-latency wireless links between machines and AMRs | Private 5G campuses in large Chinese plants |
| Digital twin | Virtual replica of the line for simulation and remote monitoring | Siemens Xcelerator, Nvidia Omniverse |
| Manufacturing operations management software | Schedules jobs, tracks work in progress, raises alarms | Siemens Opcenter, Xiaomi Hyper Intelligent Manufacturing Platform |
| Predictive maintenance | Flags likely failures before they happen | Vibration and thermal monitoring with ML models |
| Cybersecurity stack | Segments operational technology from IT, watches for intrusions | OT firewalls, zero-trust architectures |
Most commercial sites keep some lighting on for cameras and visiting engineers, so "darkness" is partly metaphorical. Where lights are off, optical systems use infrared illumination or task lights that come on only when needed.[1][2]
The table below summarizes the best-documented examples. Numbers are taken from company statements and reporting at the dates indicated; output and headcount figures change over time.
| Facility | Location | Operator | Products | Notable details |
|---|---|---|---|---|
| FANUC robot plant | Oshino, Yamanashi, Japan | FANUC | Industrial robots, robot parts | Lights-out since 2001; about 50 robots assembled per 24 hour shift; line can run unsupervised for up to 30 days; HQ campus produces about 6,000 robots per month overall.[1][3] |
| Drachten razor factory | Drachten, Netherlands | Philips | Electric razors | About 128 Adept Technology robots and 9 quality assurance staff; produces roughly 15 million razors per year; line can run unmanned for stretches of up to 30 days.[1][8] |
| Changping smart factory | Beijing, China | Xiaomi | Foldable smartphones (MIX Fold 4, MIX Flip) | About 80,000 m²; cost about 2.4 billion yuan ($330 million); 11 production lines; designed for 10 million phones per year; CEO Lei Jun has said key processes are 100% automated, with one device finished roughly every three seconds; runs on the in-house Xiaomi Hyper Intelligent Manufacturing Platform (also called Pengpai).[4][9] |
| Yizhuang pilot factory | Beijing, China | Xiaomi | Mix Fold smartphones | Earlier pilot site, capacity around 1 million units per year; preceded the Changping plant.[9] |
| Foxconn Guanlan plant | Shenzhen, China | Foxconn | Smartphone metal covers and components | Reported to run lights-out, with cloud-based AI controlling about 150 different cutting tools.[5] |
| Foxconn Kunshan plant | Kunshan, China | Foxconn | Electronics assembly | Reported in 2016 to have replaced about 60,000 workers with robots; part of a wider plan to automate 30% of operations by 2025.[5][10] |
| ASE Group test and packaging sites | Taiwan | ASE Group | Semiconductor packaging and test | Operates dozens of lights-out cleanrooms for chip packaging, where contamination control already favors very low human presence.[1] |
| Chinese textile dark mill | China | Multiple operators | Woven textiles | A reported facility runs about 5,000 looms 24/7 with AI orchestration and almost no on-floor staff.[10] |
Tesla is sometimes listed alongside these examples, but its history is more cautionary. During the Model 3 ramp in 2017 and 2018, Tesla pursued very aggressive automation at its Fremont plant and Gigafactory Nevada, with conveyor networks meant to reach a near lights-out state. Production hit roughly 2,000 cars per week while the company lost about $100 million per week, and CEO Elon Musk later admitted that "excessive automation" had been a mistake and that "humans are underrated." Tesla now runs a hybrid model with around 75% automation on the line and people involved in wiring harnesses and final assembly.[11]
Dark factories are still a small share of total manufacturing capacity, but the surrounding base is growing fast. The International Federation of Robotics 2025 World Robotics report counted about 4.66 million industrial robots in operation worldwide in 2024, a 9% rise on the prior year, with 542,000 new units installed during 2024, more than double the figure from a decade earlier.[12] Robot density has climbed accordingly: South Korea leads with 1,220 robots per 10,000 manufacturing employees, followed by Singapore, China, Germany, and Japan. Western Europe averaged 267 per 10,000, North America 204, and Asia 131.[12]
China has driven much of the recent growth. Industry reporting indicates that the country surpassed 2 million operational factory robots in 2024 and accounts for over half of new global installations, supported by the Made in China 2025 policy and roughly $1.4 billion in state-backed robotics R&D in 2023.[10]
The World Economic Forum's Global Lighthouse Network, launched in 2018, recognizes manufacturing sites that use Fourth Industrial Revolution technologies at scale. By the October 2024 cohort the network had grown to 172 sites, with 19 new Lighthouses and three Sustainability Lighthouses spread across ten countries including China, Germany, India, Singapore, Sweden, Türkiye, and Vietnam. Not every Lighthouse is a dark factory, but many of the most heavily automated sites in the world appear on the list, and AI is now the most common qualifying technology cited.[13]
The table below summarizes the practical differences.
| Property | Traditional automated factory | Dark factory |
|---|---|---|
| Human presence | Operators on every shift | A few staff for maintenance and supervision, often off-site |
| Lighting and HVAC | Full | Reduced or shut off in production zones |
| Operating hours | Limited by labor shifts | 24/7, including nights and weekends |
| Decision making | Mostly human, with PLCs | AI and software, with humans on exceptions |
| Layout | Designed for human reach and ergonomics | Designed for robot reach, dense cells, narrow aisles |
| Flexibility | Easy to retool by retraining people | Retooling requires reprogramming and recalibration |
| Failure mode | Worker spots and fixes the issue | If the system cannot self-recover, the line halts |
Many real plants sit somewhere between the two columns. "Lights-sparse" is a useful informal term for sites where some shifts run dark and others are staffed.[2]
Reported benefits include continuous output, since machines do not take breaks; lower energy use in lighting and climate control once human comfort is no longer required; consistent quality, because robots repeat motions to tighter tolerances than people; safer conditions, because dangerous, repetitive, or contaminated tasks no longer require human exposure; and a smaller building footprint, because aisles, locker rooms, and ergonomic workstations can be cut. In semiconductor packaging, the absence of people also helps maintain cleanroom standards because humans are the largest source of particulate contamination.[1][2][14]
Dark factories also have well-documented downsides. Capital cost is the first hurdle: Xiaomi's Changping plant cost about $330 million before it produced a single phone.[4][9] Brittleness is a second concern, since a single equipment failure can stop the whole line when no one is on hand to clear a jam; operators mitigate this with redundant cells and predictive maintenance.[2][14] Robots also handle limited flexibility poorly. Rebuilding a line for a new product can take weeks of programming and testing, which favors high-volume, slowly changing products such as smartphones, razors, and chips.[1][2] A networked plant is a large cybersecurity attack surface, and a successful intrusion in operational technology can stop production or damage equipment.[14] The work shifts from line operation to robotics, controls, data engineering, and AI, and many regions report a shortage of qualified technicians who can run these systems.[2]
Job displacement is the most visible social cost. Highly automated plants employ a small fraction of the headcount of their conventional equivalents. Foxconn's Kunshan reduction of about 60,000 jobs, and the broader 30 million drop in Chinese manufacturing employment since the early 2010s, point to the scale of the shift.[5][10] The World Economic Forum's Future of Jobs Report 2023 projected a net change of about 14 million jobs by 2027 from technology and other forces, with automation a major driver.[15]
Because labor cost matters less when humans are not on the line, some manufacturers have used full automation to repatriate production to high-wage countries or to keep it close to demand, which can shorten supply chains. The savings tend to flow to capital rather than labor, which raises distributional questions for governments that have historically taxed wages.[2][14] The United Nations Industrial Development Organization has warned of a "global governance deficit" in industrial automation and called for international cooperation on safety, workforce, and environmental standards.[16] AI researcher Max Tegmark has likened the broader race in autonomous systems to earlier arms races, although his concerns extend well beyond manufacturing.[17]
Fully unmanned plants will probably remain the exception through the end of the 2020s; Gartner has projected that around 60% of manufacturers will adopt some form of lights-out production by 2026, but mostly as nights and weekends rather than as continuous full automation.[2] The most likely path is hybrid: people on the day shift handling setup, exception handling, and improvement, with AI and robots holding the line on nights, weekends, and high-volume runs. As foundation models and vision-language-action models move from research into industrial deployment, the share of tasks that can be automated without painstaking custom engineering is likely to grow, which would lower the entry cost for smaller manufacturers.