Skip to Sidebar Skip to Content
Adopt · sector guide · ISIC C · checked 16 Nov 2026

AI in manufacturing: what's new, what works and how to start

AI on the factory floor and in design: new tools, who is using them at what scale, official figures in the US and EU, and the rules.
22.9% of US manufacturing businesses said they used AI in the latest US Census Bureau survey, against 24.6% of all businesses. 17.3% of EU enterprises in manufacturing used AI in 2025, by Eurostat. This guide tracks 25 developments since January 2025 and 15 named manufacturers using AI.

Latest in manufacturing AI

Every development →
Research New

LG's CLOiD robot runs GR00T-based manufacturing tasks in a real factory

NVIDIA and LG Electronics demonstrated LG's CLOiD robot doing manipulation and manufacturing tasks with NVIDIA Isaac GR00T in a real manufacturing environment. The work linked real-world data collection, Cosmos-based data augmentation, GR00T training and simulation-based evaluation. source

Benchmark New

G1 Factory benchmark tests humanoid robot policies on industrial tasks

NVIDIA published G1 Factory Benchmark: 900 human-teleoperated demonstrations, recorded in simulation, across nine industrial locomanipulation tasks for a Unitree G1 humanoid. It released a fine-tuned GR00T N1.7 policy as a baseline. source

Product New

Hitachi and FANUC partner to bring physical AI into their factories

Hitachi and FANUC formed a partnership to jointly validate and deploy physical AI that combines Hitachi's AI models with FANUC's industrial robots. Hitachi's factories in the Ibaraki region will be the first test sites, used as 'Customer Zero'. source

Product New

AI Welding Agent turns component drawings into robot welding programs

FANUC announced an AI Welding Agent, built with Google's Gemini Enterprise, that reads component drawings, sets welding parameters and generates and runs robot arc-welding programs without manual teaching. FANUC said shipments were scheduled to start by the end of December 2026. source

How many are using it

US businesses using AI22.9%

of US manufacturing businesses used AI in the fortnight 21 Sep 2026 to 4 Oct 2026 (US Census Bureau, BTOS). All businesses: 24.6%. ±1.4 points.

EU enterprises using AI17.3%

of EU enterprises with 10 or more people in manufacturing used AI in 2025 (Eurostat). All covered activities: 19.9%. Over time: 2021: 7%, 2023: 7%, 2024: 11%, 2025: 17%.

What EU firms in this sector use AI for, 2025
Marketing or sales5.2%
ICT security3.6%
Production processes3.5%
Research and innovation3.0%
Accounting and finance2.9%
Logistics1.1%

Percent of all enterprises in the activity. Source: Eurostat, isoc_eb_ain2, read 14 Nov 2026.

More official figures

From statistics offices, central banks, regulators and governments. Each measures something a little different, so they are not directly comparable.
PublisherWhat it measuresFigureSource
Australian Bureau of StatisticsAU · 2024-25 (financial year)Characteristics of Australian Business, 2024-25 financial yearManufacturing 9% (all businesses 12%)9%source
Federal Reserve Bank of New YorkUS: New York–Northern New Jersey region · 2025-08Are Businesses Scaling Back Hiring Due to AI? (Liberty Street Economics, from the Regional Business Surveys)26 percent of manufacturers used AI, up from 16 percent a year earlier26%source
Statistics CanadaCA · 2025-Q2Analysis on artificial intelligence use by businesses in Canada, second quarter of 202511.3% in manufacturing (second-highest industry after finance and insurance, 13.2%)11.3%source
Statistics CanadaCA · 2025-Q2Analysis on artificial intelligence use by businesses in Canada, second quarter of 202516.2% in Q2 2025, up from 11.8% in Q2 2024 (all businesses: 17.9%)16.2%source
Office for National Statistics (UK)GB · 2023Management practices and the adoption of technology and artificial intelligence in UK firms: 20235% of firms in the manufacturing sector (services: 9%)5%source
Office for National Statistics (UK)GB · 2023Management practices and the adoption of technology and artificial intelligence in UK firms: 2023manufacture of textiles (11%), the highest AI adoption among manufacturing sub-sectors11%source

Start from your role

Look up any job →
Pick your role. Each task gets our call, computed from how much a mistake could hurt and how easily a person can check the output (how calls are made), and the evidence rating for it.

Production managers and supervisors

Try with checks

Pressure-test a plan

It is worth trying only with the checks on this page. Part of the work in 7 of the 7 jobs we map for this role.

Try with checks

Notes into a report

It is worth trying only with the checks on this page. Part of the work in 7 of the 7 jobs we map for this role.

Try with checks

Analyse a spreadsheet

It is worth trying only with the checks on this page. Part of the work in 6 of the 7 jobs we map for this role.

Machine operators and assemblers

Try with checks

Summarise a long document without losing what matters

It is worth trying only with the checks on this page. Part of the work in 12 of the 34 jobs we map for this role.

Quality inspectors

Try with checks

Notes into a report

It is worth trying only with the checks on this page. Part of the work in 1 of the 1 jobs we map for this role.

Try with checks

Pressure-test a plan

It is worth trying only with the checks on this page. Part of the work in 1 of the 1 jobs we map for this role.

Industrial engineers

Try with checks

Research a question

It is worth trying only with the checks on this page. Part of the work in 9 of the 9 jobs we map for this role.

Try with checks

Plan and test code

It is worth trying only with the checks on this page. Part of the work in 7 of the 9 jobs we map for this role.

Try with checks

Pressure-test a plan

It is worth trying only with the checks on this page. Part of the work in 7 of the 9 jobs we map for this role.

Try with checks

Fix a software problem

It is worth trying only with the checks on this page. Part of the work in 4 of the 9 jobs we map for this role.

Maintenance technicians

Try with checks

Notes into a report

It is worth trying only with the checks on this page. Part of the work in 1 of the 2 jobs we map for this role.

Try with checks

Analyse a spreadsheet

It is worth trying only with the checks on this page. Part of the work in 1 of the 2 jobs we map for this role.

What's new: key developments

New models, products, rules, benchmarks and research since January 2025, newest first. Each is from the company’s, regulator’s or authors’ own publication and was checked against it (checked 2026-11-14). A product or model listed here is not a finding that it works; the evidence section is where results are rated. Pick a kind, or a dot in the chart at the top, to go straight to it.
  1. Nov 2026 · 1 development

  2. NVIDIA and LG ElectronicsResearchNew LG's CLOiD robot runs GR00T-based manufacturing tasks in a real factoryNVIDIA and LG Electronics demonstrated LG's CLOiD robot doing manipulation and manufacturing tasks with NVIDIA Isaac GR00T in a real manufacturing environment. The work linked real-world data collection, Cosmos-based data augmentation, GR00T training and simulation-based evaluation. source
  3. Oct 2026 · 1 development

  4. NVIDIABenchmarkNew G1 Factory benchmark tests humanoid robot policies on industrial tasksNVIDIA published G1 Factory Benchmark: 900 human-teleoperated demonstrations, recorded in simulation, across nine industrial locomanipulation tasks for a Unitree G1 humanoid. It released a fine-tuned GR00T N1.7 policy as a baseline. source
  5. Sep 2026 · 2 developments

  6. Hitachi and FANUCProductNew Hitachi and FANUC partner to bring physical AI into their factoriesHitachi and FANUC formed a partnership to jointly validate and deploy physical AI that combines Hitachi's AI models with FANUC's industrial robots. Hitachi's factories in the Ibaraki region will be the first test sites, used as 'Customer Zero'. source
  7. FANUC, with GoogleProductNew AI Welding Agent turns component drawings into robot welding programsFANUC announced an AI Welding Agent, built with Google's Gemini Enterprise, that reads component drawings, sets welding parameters and generates and runs robot arc-welding programs without manual teaching. FANUC said shipments were scheduled to start by the end of December 2026. source
  8. Jul 2026 · 1 development

  9. European Union (European Commission, Council, Parliament)RulesNew EU delays AI Act high-risk rules for AI in machinery to 2028The EU's AI Omnibus entered into force, moving AI Act high-risk rules for AI in products such as machinery to 2 August 2028. The amended Article 6 also says AI used only for quality control, performance optimisation or automation is not a safety component unless its failure would endanger health and safety. In force 2 Aug 2028. source
  10. Jun 2026 · 1 development

  11. Health and Safety Executive (HSE), UKRulesNew UK safety regulator says existing law covers AI at work and in machineryBritain's Health and Safety Executive published a policy statement saying health and safety law applies to AI whatever the technology. Employers must risk-assess uses of AI that affect workplace safety, and HSE regulates AI in workplace machinery as a market surveillance authority. source
  12. Mar 2026 · 1 development

  13. StockmarkModelNew Open Japanese model for reading manufacturing documents and drawingsJapanese company Stockmark released Stockmark-DocReasoner, an open-weights 32B vision-language model tuned to read and reason over Japanese manufacturing and business documents, including engineering drawings and technical reports. The work was supported by Japan's GENIAC programme. source
  14. Jan 2026 · 3 developments

  15. Ministry of Industry and Information Technology (MIIT) and seven other Chinese government bodiesRulesNew China issues 'AI + Manufacturing' implementation opinionsChina's industry ministry and seven other national bodies issued implementation opinions for an 'AI + Manufacturing' special action. The plan covers industry-specific large models, industrial AI agents, industrial datasets, adoption by manufacturers and AI safety governance. source
  16. Siemens and NVIDIAProductNew Siemens and NVIDIA widen partnership on AI-driven factoriesSiemens and NVIDIA expanded their partnership to bring NVIDIA accelerated computing, AI physics models and Omniverse into Siemens' design, simulation and automation software. Siemens said Foxconn, HD Hyundai, KION Group and PepsiCo are evaluating some of the capabilities. source
  17. Boston Dynamics (Hyundai Motor Group)ProductNew Production Atlas humanoid built, with 2026 units going to Hyundai and Google DeepMindBoston Dynamics unveiled the product version of its Atlas humanoid at CES and said it would begin production at its Boston headquarters immediately. All 2026 units are committed to Hyundai's Robotics Metaplant Application Center and Google DeepMind, and a new partnership will put DeepMind foundation models into Atlas. source
  18. Nov 2025 · 2 developments

  19. Figure AI, at BMW Group Plant SpartanburgProductNew Humanoid robots' 11 months on a BMW line: results publishedFigure reported results of an 11-month deployment of its Figure 02 humanoid robots at BMW's plant in Spartanburg, South Carolina: 10-hour weekday shifts, more than 90,000 parts loaded and 1,250+ hours of runtime. Figure then began retiring Figure 02 in favour of Figure 03. source
  20. Deutsche Telekom, with NVIDIAProductNew Industrial AI cloud for European manufacturers launched in MunichDeutsche Telekom announced an Industrial AI Cloud in Munich, built with NVIDIA, where German and European manufacturers can run simulation, digital-twin and AI workloads on their own data. The release said operations would start in the first quarter of 2026. source
  21. Oct 2025 · 2 developments

  22. Samsung ElectronicsProductNew Samsung and NVIDIA plan an 'AI Megafactory' for chips, phones and robotsSamsung announced plans with NVIDIA to embed AI across its whole manufacturing flow, including NVIDIA Omniverse digital twins of its chip fabs. It reported a 20x gain in computational lithography performance for optical proximity correction using NVIDIA cuLitho. source
  23. European CommissionRulesNew EU Apply AI Strategy makes manufacturing a flagship sectorThe European Commission adopted its Apply AI Strategy, with flagship measures for ten industrial sectors including manufacturing, engineering and construction. For manufacturing it plans support for frontier AI models adapted to the sector, industrial data pooling, funding to move AI from the lab into factories, and a stronger testing facility to validate and certify AI-driven machinery. source
  24. Sep 2025 · 1 development

  25. Google DeepMindModelNew Gemini Robotics 1.5 plans multi-step tasks and transfers skills across robotsGoogle DeepMind released Gemini Robotics 1.5, a vision-language-action model that reasons before acting and learns across robot types, and Gemini Robotics-ER 1.5 for planning and tool use. The ER model is open to developers through the Gemini API; the action model is limited to selected partners. source
  26. Aug 2025 · 1 development

  27. State Council of the People's Republic of ChinaRulesNew China's 'AI+' action plan calls for AI across industrial design and productionChina's State Council issued opinions on carrying out its 'AI+' action. A section on industry calls for AI in design, pilot testing, production, service and operations, faster progress in industrial software, and more smart manufacturing equipment. source
  28. Jul 2025 · 1 development

  29. European Commission, EMA and PIC/SRulesNew Draft GMP Annex 22 sets the first rules for AI in medicines manufacturingThe European Commission and PIC/S opened a consultation on a new Good Manufacturing Practice Annex 22 on AI, drafted by EMA and PIC/S inspectors, alongside a revised Annex 11 on computerised systems. It covers selection, training, validation and ongoing monitoring of AI models used to make medicines and active substances. source
  30. Jun 2025 · 2 developments

  31. Google DeepMindModelNew Gemini Robotics On-Device runs robot control on the robot itselfGoogle DeepMind released Gemini Robotics On-Device, a vision-language-action model optimised to run locally on the robot. An SDK lets developers in a trusted tester programme adapt it to new tasks with 50 to 100 demonstrations. source
  32. IBM ResearchBenchmarkNew AssetOpsBench tests AI agents on industrial maintenance tasksIBM Research released AssetOpsBench, an open benchmark for AI agents doing industrial asset operations and maintenance, with four domain agents, 140+ human-written queries and a simulated sensor environment. The latest paper version reports 250+ users and 500+ agents submitted. source
  33. May 2025 · 1 development

  34. SiemensProductNew Siemens adds AI agents behind its Industrial CopilotsSiemens announced industrial AI agents that run across its Industrial Copilot products, with an orchestrator handing tasks to specialised agents for design, planning, engineering, operations and maintenance. It also announced a planned agent marketplace on Siemens Xcelerator. source
  35. Mar 2025 · 4 developments

  36. MicrosoftProductNew Factory Operations and Factory Safety agents reach public preview in Copilot StudioAhead of Hannover Messe, Microsoft put its Factory Operations Agent into public preview in Copilot Studio, so operators and managers can query machine and production data in plain language. It also previewed a Factory Safety Agent that answers workers' questions on health and safety rules. source
  37. NVIDIAModelNew GR00T N1, an open foundation model for humanoid robotsNVIDIA released Isaac GR00T N1, an open, customisable foundation model for humanoid robots with a fast action system and a slower reasoning system. Developers can post-train it with real or synthetic data for their own robot and task. source
  38. Google DeepMindModelNew Gemini Robotics models bring Gemini 2.0 to robot controlGoogle DeepMind introduced Gemini Robotics, a vision-language-action model built on Gemini 2.0 that outputs robot actions, and Gemini Robotics-ER for spatial reasoning. It partnered with Apptronik on humanoid robots and opened the ER model to trusted testers. source
  39. Hon Hai Research Institute (Foxconn)ModelNew Foxconn builds FoxBrain, an in-house language model for manufacturingFoxconn's research institute launched FoxBrain, a 70B-parameter Traditional Chinese model based on Meta's Llama 3.1, trained in about four weeks on 120 NVIDIA H100 GPUs. It was built for internal use and is meant to support manufacturing, supply chain management and decision-making. source
  40. Jan 2025 · 1 development

  41. NVIDIAModelNew Cosmos open world models generate synthetic training data for robotsNVIDIA released Cosmos, a family of world foundation models that generate physics-based synthetic video to train and test robots and autonomous vehicles. The first models came out under NVIDIA's open model licence on Hugging Face and the NGC catalog. source
Followed over time in The AI Record: US Chip Export Controls on China.

AI models built for manufacturing, open and closed

3 models built or tuned for manufacturing; 3 have open weights anyone can download and run. Scores are the makers’ own, on the tests they chose, so they are not comparable across rows.

ModelWeightsWhat it is forAvailability and statusPublished scores
G1-Factory-GR00T-1.7NVIDIA · 19 Oct 2026Open weightsNVIDIA Open Model License (not an OSI open-source licence)Performing and evaluating nine factory locomanipulation tasks with a Unitree G1 humanoid in simulation; a baseline and starting point for further fine-tuning or reinforcement learning.3.1B (3,144,016,000 parameters) · head-camera image, robot joint state and an English instruction in; whole-body robot actions outopen download (Hugging Face)None publishedsource
Stockmark-DocReasoner-Qwen2.5-VL-32BStockmark · 19 Mar 2026Open weightsApache-2.0Reading and reasoning over Japanese manufacturing and business documents, such as technical documentation, engineering design drawings and experimental reports.32B · text, images (documents, charts, tables, diagrams, engineering drawings)open download (Hugging Face)None publishedsource
FoxBrain (Llama 3.1 FoxBrain 70B)Hon Hai Research Institute (Foxconn) · 10 Mar 2025Open weightsLlama 3.1 Community License plus Hon Hai's own usage agreement; gated, academic and research use only, no commercial useBuilt for Foxconn's internal applications; Foxconn says it will support manufacturing, supply chain management and decision-making, and its smart manufacturing platform.70B · text (Traditional Chinese and English)gated download on Hugging Face (FoxconnAI/Llama_3.1-FoxBrain-70B and -V1.2) for approved academic and research usersNone publishedsource

Who's building manufacturing AI

Company trackers →

AI-native manufacturing companies

Companies whose main product is AI for manufacturing, with figures each published itself (checked 2026-11-14). Not a ranking.
US · founded 2011

Augury

Sensors and AI that monitor industrial machines and processes to predict failures and plan maintenance in factories.

  • $75 million (round led by Lightrock) (19 Feb 2025) source
  • Over 40 countries (19 Feb 2025) source
  • Over 500 million hours of machine data analysed (19 Feb 2025) source
CH · founded 2019

Neural Concept

AI platform that learns from CAD and simulation data to predict how product designs will perform, used by engineering teams in cars, aerospace and electronics.

  • $100 million Series C (18 Dec 2025) source
GB

PhysicsX

AI models trained on physics simulation to speed up engineering design and manufacturing decisions in sectors such as automotive, aerospace and foundries.

US · founded 2015

Instrumental

Camera and AI software that finds defects and failure causes on electronics assembly lines.

  • $50M Series C (16 Feb 2022) source
US

Figure AI

Builds general-purpose humanoid robots and the Helix AI model that controls them; its Figure 02 robots ran on a BMW car-plant line for 11 months.

  • More than $1 billion in committed capital (Series C) (16 Sep 2025) source
  • Valued at $39 billion post-money (16 Sep 2025) source
US · founded 2016

Apptronik

Builds the Apollo humanoid robot for factory and warehouse work such as moving parts to assembly lines.

DE

NEURA Robotics

Makes cognitive industrial robots, cobots and humanoids with built-in sensing and AI for tasks such as welding, machine tending and quality inspection.

CN

UBTECH Robotics

Chinese maker of humanoid robots, including the Walker S series used in car, electronics and semiconductor factories; listed in Hong Kong.

  • Revenue total revenue exceeded RMB 2 billion in 2025 (13 Jul 2026) source
  • 1,079 full-size humanoid robots sold in 2025 (13 Jul 2026) source
CN

Mech-Mind Robotics

3D cameras and AI vision software that let industrial robots pick, place, assemble and measure parts.

IN

Ripik.ai

Computer vision AI agents that monitor processes and equipment in steel, cement, aluminium and power plants.

JP

ARUM

AI software (the ARUMCODE series) that automatically estimates machining costs and writes NC programs for CNC cutting of metal parts.

What the big AI companies are doing in manufacturing

Who's using it

Named manufacturers rolling out AI, at the scale they state, from their own or their vendor’s announcement (checked 2026-11-14). 15 organisations in 8 countries; 8 say the tool is in use now, the rest are rolling out, planned or piloting it. An organisation’s reported result is its own claim, not a rated finding.

Which vendor, for which use

Each number is how many of the named organisations below use that vendor for that job. Tap one to see them. Among the deployments we checked, not market share.
VendorQuality inspectionPredictive maintenanceDesign and engineeringProduction planningRobots and automationOtherAll
Augury
Google Cloud
Microsoft
Accenture
Apptronik
Datagon AI
Figure AI
Siemens
Built in-house

The organisations

OrganisationUse and vendorStage and scaleWhat
Emirates Global Aluminium (EGA)AE · 3 Jul 2026Otherin-house (with a Microsoft cloud platform)In usemore than 80 Industry 4.0 use casesEGA runs more than 80 Industry 4.0 applications in its aluminium operations, from AI vision that checks carbon anode quality in real time to tools that predict commodity prices.Reported by the organisation: Some $100 million in financial impact from its Industry 4.0 use cases.source · announced by the organisation
Grupo BimboMX · 29 Jun 2026OtherMicrosoft (Microsoft 365 Copilot, Copilot Studio)In usebetween 50% and 60% of auditors already actively using the agents (global internal audit team of more than 100 auditors)Grupo Bimbo, the bakery manufacturer, gave its global internal audit team two AI agents that pull audit standards and templates from its document store and help plan audits.Reported by the organisation: About 20% less time to complete the planning phase of audits.source · announced by the vendor
BMW Group (Plant Spartanburg)US · 25 Jun 2026Robots and automationFigure AIPilotthe Figure 02 robot supported the production of more than 30,000 BMW X3 vehicles over ten monthsA Figure 02 humanoid robot inserted sheet-metal parts for welding in the Spartanburg body shop for about ten months in 2025. BMW is now starting the successor, Figure 03, on a sequencing task in logistics.source · announced by the organisation
UnileverUK · 16 Jun 2026OtherAccentureRolling outdigital twins already in use at sites including Raeford (US), Poznan (PL), Gandhidham (IN) and Cu Chi (VN); more than 40 new twins planned over 18 monthsUnilever uses AI-enabled digital twins of production lines at several factories to predict flow problems and tune ingredient dosing. With Accenture it is building more than 40 more.Reported by the organisation: Poznan (mayonnaise): minor stoppages down by up to 20% and waste down by nearly 30%. Cu Chi (detergent): 1–2% savings in premium ingredients.source · announced by the vendor
Tata SteelIndia · 22 Apr 2026OtherGoogle Cloud (Gemini, Google Cloud AI stack)In usea fleet of over 300 specialized AI agents in just nine monthsTata Steel built more than 300 AI agents across its operations on Google Cloud, including agents that check equipment health for maintenance and agents that sort customer complaints from photos. Employees who are not data scientists can build agents on an internal low-code platform.Reported by the organisation: Complaint-handling agents reduced average turnaround time by 50%.source · announced by the vendor
GE Appliances (a Haier company)US · 22 Apr 2026Production planningGoogle Cloud (Gemini Enterprise)In usemore than 800 AI agents across its manufacturing, logistics, and supply chain operationsGE Appliances put Gemini into its Brilliant Factory production data platform. Agents summarise shift data and let staff query production data across lines, shifts and plants.Reported by the organisation: A Supplier Collaboration Agent that automated order-status inquiries with more than 600 suppliers led to a 25% reduction in backorders.source · announced by the vendor
Fortune Brands InnovationsUS · 7 May 2025Predictive maintenanceAuguryRolling outdeployed on two-thirds of the 1,000 machines targeted across 16 sites in the U.S., UK, and MexicoFortune Brands Innovations, which makes home products such as Fiberon decking, is rolling out Augury's AI machine-health monitoring across 16 plants after a 40-machine pilot.Reported by the organisation: The Fiberon pilot achieved 2.5x return on investment in eight months.source · announced by the vendor
BMW Group (Plant Regensburg)DE · 28 Apr 2025Quality inspectionDatagon AIPilotan individual inspection catalogue for each specific customer vehicle; around 1,400 vehicles roll off the line every workdayBMW's Regensburg plant piloted an AI tool, GenAI4Q, that builds a tailored list of quality checks for each car in final assembly. It was developed with the Munich start-up Datagon AI.source · announced by the organisation
Toyota Motor CorporationJP · 19 Nov 2024Design and engineeringMicrosoft (Azure OpenAI Service)In usesome 800 engineers who work on powertrains ... have had access to O-BeyaToyota built 'O-Beya', a set of generative AI agents grounded in its own design reports and regulations, for its powertrain development department. Engineers can ask several expert agents, such as a vibration agent, at once.source · announced by the vendor
thyssenkrupp Automation EngineeringDE · 24 Oct 2024Design and engineeringSiemens (Industrial Copilot, on Microsoft Azure OpenAI Service)In useOver 100 companies, including Schaeffler and thyssenkrupp Automation Engineering, are currently using the Siemens Industrial Copilotthyssenkrupp Automation Engineering uses Siemens' generative AI assistant to help engineers build automation projects for machines such as battery quality-check lines. It was the first customer to plan a global rollout.source · announced by the vendor
PepsiCo (Frito-Lay)US · 3 Jul 2024Predictive maintenanceAuguryIn use36 Frito-Lay sites in the U.S. and Canada and two more in the U.K.Frito-Lay uses Augury's machine health sensors and AI to spot equipment faults before breakdowns. It started as a pilot in four factories.Reported by the organisation: Augury says it has helped mitigate 900 unplanned downtime events, avoiding an estimated 4,500 hours of unplanned downtime.source · announced by the vendor
BoschDE · 22 Apr 2024Quality inspectionin-housePilottwo Bosch plants in GermanyBosch started projects in two German plants where generative AI makes synthetic images of good and faulty parts to train optical inspection models.source · announced by the organisation
Mercedes-BenzDE · 15 Mar 2024Robots and automationApptronikPilotpilot of Apollo humanoid robots in Mercedes-Benz manufacturing facilitiesMercedes-Benz signed a commercial agreement to pilot Apptronik's Apollo humanoid robot, starting with moving kits of parts to the assembly line and checking components.source · announced by the vendor
LenovoCN · 6 Dec 2023Production planningin-house (Advanced Production Scheduling)In useOver 75% of the process is now automated, and the new solution has an adoption rate of almost 95%Lenovo built its own AI production scheduling system, using deep reinforcement learning, to plan PC production lines with many custom orders. Planning a schedule went from two hours to two minutes.Reported by the organisation: 24% more production line capacity, 19% higher production volumes, a 3.5x increase in on-time deliveries, and schedule planning cut from two hours to two minutes.source · announced by the organisation
Audi (Neckarsulm plant)DE · 30 Jun 2023Quality inspectionin-house (WPS-Analytics)Rolling outaround 1.5 million spot welds on 300 vehicles each shiftAudi uses AI to check spot welds in the Neckarsulm body shop, replacing random ultrasound sampling. Audi said it was installing the infrastructure at Audi Brussels and planned it for Ingolstadt and Volkswagen's Emden plant.source · announced by the organisation

AI in manufacturing by country

AI by Country →
Where AI in manufacturing is being used and regulated: firms using ai, deployments, rules. This is activity, not a full count: only the EU measures AI use by sector and country. Click a country to open its tracker.

Leading countries

CountryFirms using AINamed deploymentsRules on this page
Belgium40%06
Denmark39%06
Luxembourg38%06
Sweden33%06
Austria33%06
Netherlands29%06
Norway26%00
Germany24%56
Slovenia23%06
Estonia23%06
Firms using AI: Eurostat, enterprises with 10 or more people. Deployments and rules: the checked lists on this page; EU rules count for every member state.

Official registers you can search

Official lists of AI systems or AI activity in manufacturing, kept by governments and regulators. They are the best place to check what is actually in use.

What the studies show so far

All rated use cases →

Studies of AI in manufacturing

3 studies with a comparison group or a before-and-after measure, done in manufacturing. Grades: A, a well-run randomised trial in real work; B, a strong natural experiment or a well-run trial on realistic tasks; C, a trial with clear limits; D, before-and-after, surveys or company case studies. How we grade.
  • Procter & Gamble staff developing new product ideas spent 12-16% less time with AI, in a randomised trial.
    Dell'Acqua et al. (2026) · grade B, randomised trial in live work · task: Analysis and consulting with AI
  • At a small manufacturer, staff using an AI assistant that suggests fixes from past quality problems took 19 minutes less per problem than before, in a five-workstation pilot run by the tool's developers.
    Caglar (2026) · grade D, before and after · task: Finding information with AI
  • At appliance maker Beko, pull requests took 42% longer to close after an AI code-review bot was added, in a before-and-after comparison.
    Cihan et al. (2024) · grade D, before and after · task: AI in code review
3 more studies found but not counted
  • Trapero et al. (2026), grade C: Not work: volunteers of unstated occupation in a 10-minute lab simulation of an assembly line, with different tasks in the two conditions (section 4, rule 1).
  • Hekman (2026), grade D: Not work: student groups in a machine design course, compared by how much AI they chose to use (section 4, rule 1).
  • Gopalakrishnan (2026), grade D: No comparison within the study (section 4, rule 3): the 'conventional' figures come from other papers, and the group sizes do not match the results reported.

Ratings that include this sector

Where results were positive

  • Requirements and design with AI, Bluemelhuber and Junker (2026): −30.4% Time on simple tasks, GenAI vs manual (mean 184.7 s vs 265.3 s).
  • Requirements and design with AI, Bluemelhuber and Junker (2026): −31% Time on medium tasks, GenAI vs manual (246.9 s vs 357.8 s).
  • Requirements and design with AI, Bluemelhuber and Junker (2026): −25.8% Time on complex tasks, GenAI vs manual (318.2 s vs 428.6 s).
  • Data analysis with AI, Ipeirotis and Zheng (2025): −31% Completion time per query, LLM interface vs manual SQL (Snowflake).
  • Data analysis with AI, Lu et al. (2025): −57% Mean minutes per question (4.0 vs 9.3; % computed from reported means).
  • Analysis and consulting with AI, WorkLab (2025): −51.8% Time to determine the root cause of a simulated risk event (loan approval delays), Copilot group v control without Copilot.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): +12.2% Tasks completed, 18 consulting tasks inside the AI frontier.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): −25.1% Time to reach the final question (first 17 of 18 tasks inside the frontier); -22.5% GPT + overview, -27.6% GPT only, about -25.1% on average.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): +29.9% Graded quality of solutions inside the frontier.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): +0.37 SD Expert-rated solution quality, individual with AI vs individual without AI.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): +0.39 SD Expert-rated solution quality, two-person team with AI vs individual without AI.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): −16.4% Time spent on the solution vs individuals without AI.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): +0.46 SD Positive emotions (excitement, energy, enthusiasm), individual with AI vs control.
Watch-outs: where results were worse
  • Requirements and design with AI, Broccia et al. (2026): Group that did the first session WITH ChatGPT: Improved about 6-7% between sessions vs about 12% for the group that started without ChatGPT; the authors read this as slower skill learning
  • Data analysis with AI, Fitzpatrick et al. (2025): −12% Accuracy of answers to six questions on insurer reporting templates, 40 values in all, graded against ground truth (32.3 Gen AI vs 36.5 control out of 40).
  • Data analysis with AI, Fitzpatrick et al. (2025): Participants scoring under 100% on the baseline assessment (N=18): With Gen AI, the correct-answer rate fell from 79.63% on the assessment to 65.83% on the experiment task
  • Data analysis with AI, Fitzpatrick et al. (2025): Participants scoring 100% on the baseline assessment (N=49): With Gen AI, the correct-answer rate fell from 100% to 84.48%; the baseline task was too easy to separate people
  • Data analysis with AI, Brodeur et al. (2026): −0.95 major errors detected per team Major coding errors detected per team, AI-assisted vs human-only (0.74 vs 1.70).
  • Data analysis with AI, Brodeur et al. (2026): Teams where ChatGPT led the work (minimal human oversight): Reproduced 56.8 points fewer papers than human-only teams (37%) and took about 119% longer.
  • Analysis and consulting with AI, Dell'Acqua et al. (2026): −19 points Correct solutions on a complex managerial task chosen to be outside the frontier.
  • AI in code review, Maddila et al. (2025): +5.5% TimeInReview: time a diff spent in review when AI patches were shown to reviewers (Expt. 1).
  • AI in code review, Cihan et al. (2024): +42% Average pull request closure time, before vs after the bot (5h52m to 8h20m; % computed).
  • AI in code review, Cihan et al. (2024): Project #1: Closure time rose from 2h48m to 4h38m (p < 0.001)
  • AI in code review, Cihan et al. (2024): Project #3: closure time rose from 20h22m to 30h51m (p < 0.001)
Rules and changes with a date already set, soonest first.
  1. European Union (European Commission, Council, Parliament)EU delays AI Act high-risk rules for AI in machinery to 2028The EU's AI Omnibus entered into force, moving AI Act high-risk rules for AI in products such as machinery to 2 August 2028. The amended Article 6 also says AI used only for quality control, performance optimisation or automation is not a safety component unless its failure would endanger health and safety. source

Rules to check first

AI in machinery (EU)

In the EU, AI that is a safety component of machinery, or is itself the product, counts as high-risk under the AI Act; for products in Annex I, such as machinery, these rules apply from 2 August 2028. AI used only for quality control, performance optimisation or automation is not a safety component, unless its failure would endanger health and safety. European Union →

Managing factory workers

Annex III of the EU AI Act has no point for manufacturing itself, but point 4(b) makes AI that allocates tasks by individual behaviour, or monitors and evaluates workers' performance, high-risk. Those duties apply from 2 December 2027. European Union →

Workplace risk assessment

In Great Britain, existing health and safety law covers AI. Any use of AI that affects workplace safety, such as AI-driven robots or machine controls, needs a risk assessment and controls, and HSE checks AI in workplace machinery as a market surveillance authority. United Kingdom (Great Britain) →

Robot safety (US)

OSHA has no standard written for robots, so AI-driven robots and machines fall under its general machinery and hazard rules, such as those on amputation hazards. Check OSHA's robotics hazard guidance before putting an AI-controlled robot near people. United States →

Medicines manufacturing (GMP)

Drug makers using AI in production must already validate it as a computerised system under GMP. The draft EU and PIC/S Annex 22 adds rules on intended use, performance metrics, test data and ongoing monitoring of AI models; it was still a draft in November 2026. European Union and PIC/S member authorities →

A starting checklist, checked against each source (2026-11-14), not legal advice. Laws differ by country: AI laws in your country; EU AI Act: what applies when.
  1. Start with one taskPick one task with a clear owner and a measure you already track, not a company-wide launch.
  2. Measure against a baselineRecord time, quality and errors before AI and after, for users and a comparison group.
  3. Keep a person checkingSomeone accountable reviews what the AI produces before it reaches a customer or a decision.
  4. Check the rules and dataConfirm what data may go into the tool, the vendor's terms, and the rules for your sector below.
  5. Train the people using itShort, task-specific training; free courses are in our learning directory.
Practice, not graded evidence: steps drawn from how the rated studies were run. For the full 90-day plan, see the AI rollout kit.

Learn for free

All free courses →

Questions people ask

How many manufacturers use AI?

22.9% of US manufacturing businesses said they used AI in the latest US Census Bureau survey, against 24.6% of all businesses. 17.3% of EU enterprises in manufacturing used AI in 2025, by Eurostat. Surveys count organisations, not individual workers.

What do manufacturers use AI for?

Among the 15 named deployments on this page, the most common uses are other (4), quality inspection (3), robots and automation (2). That is what organisations have announced, not market share. See who →

Are there AI models built for manufacturing?

Yes. This page lists 3, of which 3 have open weights anyone can download. Compare them →

Does AI help people write requirements and designs?

Not enough evidence yet to say whether AI helps people write requirements and designs. Evidence strength 1 of 5, from 3 counted studies. The evidence →

Does AI help people analyse data more accurately?

Mixed results: studies of AI for data analysis disagree on whether it helps. Evidence strength 2 of 5, from 7 counted studies. The evidence →

Does AI help with consulting and analysis work?

AI for consulting and analysis tasks may work for tasks that sit inside what the AI does well. Evidence strength 2 of 5, from 3 counted studies. The evidence →

Does AI make code review faster or better?

Not enough evidence yet to say whether AI makes code review faster or better. Evidence strength 1 of 5, from 4 counted studies. The evidence →

Which rules apply?

Start with ai in machinery (eu): In the EU, AI that is a safety component of machinery, or is itself the product, counts as high-risk under the AI Act; for products in Annex I, such as machinery, these rules apply from 2 August 2028. AI used only for quality control, performance optimisation or automation is not a safety component, unless its failure would endanger health and safety. All the rules to check →

How this page is made

This guide covers ISIC Rev. 5 section C (all 14 sector guides). Developments, models, deployments, rules and company figures come from each organisation’s own publication, a regulator or a statistics office, and each was checked against its source by a second, independent pass; items that failed were removed. Adoption figures are the US Census Bureau’s and Eurostat’s own estimates. Use cases are rated from published studies by fixed rules; a product listed here is not a finding that it works. Editorial owner: AI Adoption editors. Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work, and some companies on this page make them. No company has any say in this page. Spot a mistake? Corrections.

Copied with its source. Want the next number like this in your inbox?Get the evidence