2026-11-03 · own notes + six research passes
AI in manufacturing: a value-chain map of 142 startups
One factory, drawn as layers from the sensor out to the people, then out to the companies on either side. The four companies we studied are placed first; 138 other startups under about 200 people follow in colour by region.
The value map
placement
colour = where the company is from
studied up closeEuropeIsraelChinaJapanKorea / TaiwanUS / CanadaIndiaother
marker = how it touches the machine
writes to the machinereads sensorsreads MES / ERPdocuments only
the plant, company X rows ordered by distance from the machine; the lowest row touches the sensor, the top row touches only people
L5Knowledge & peopleSOPs, work instructions, audit, training, RAG over documents, connected worker17 primary +6 also here
L4Planningscheduling, machine allocation, inventory, MRP, ERP ordering, procurement19 primary +21 also here
CirconomitakenoALGO ARTIS (株式会社ALGO ARTIS)Aolin Technology (傲林科技)ArchletArkestroFlexcitonFlowlityFulcrumImpactive AI (임팩티브AI)InterAgent (塔台智能網絡)KeelvarOplitparetosSkydisc (株式会社スカイディスク)SupplyETTactoYuanmu Intelligence (圆木智能)ZentioSmart Steel TechnologiesArch SystemsSmart Craft (株式会社Smart Craft)DiracChAICofactrCosmo TechDideroEnmovilKavida.aiKodiak HubKREATIZELevaDataLingMao SCM (领猫SCM)LyticaNulogyPactumPartsimonySourceDaySyncFabThroughPut Inc
L3Line & plant monitoring, causal analysisOEE, root cause, downtime and causal reports a human reads13 primary +35 also here
EthonArch SystemsBraincubeENLYZEFabriqFactbirdFourJaw Manufacturing AnalyticsGuidewheelMIPU Predictive HubRaven.aiShuyi Link (数益工联)Smart Craft (株式会社Smart Craft)TEEPTRAKCyberInsight (天泽智云)ElmodisFalkonryGroundup.aiMOVUSNanoprecise Sci CorpBasetwo AIFero LabsImubitIntelecyMogulinker / Mushroom IoT (蘑菇物联)Oden Technologiesplus10Profet AI (杰倫智能科技)ElementaryInstrumentalRipik.aiSixSenseAolin Technology (傲林科技)FulcrumZentioAugmentirAzumutaDatchEdmund AILeucineSquintNulogyFactryFlow SoftwareGoodLinker (谷林運算)Quartic.aiUlala Lab (울랄라랩)United Manufacturing HubZerynth
L1Machine configuration & controlparameter setting, recipe optimisation, closed-loop control20 primary +5 also here
GigatonAISing (株式会社エイシング)Aixtal (アイクリスタル株式会社)All Sense Technology / Quanying (全应科技)AMESA (formerly Composabl)Basetwo AIDeep.MetaFero LabsImubitIntelecyIntellegensINTERX (인터엑스)KelvinLimitless Labs (formerly LimitlessCNC)Mogulinker / Mushroom IoT (蘑菇物联)Oden Technologiesplus10Profet AI (杰倫智能科技)Smart Steel TechnologiesTVARITEthonPolysenseRipik.aiBraincubeQuartic.ai
L0Machine & sensorraw signals, sensor correction, soft sensors, edge anomaly detection15 primary +15 also here
Amiral TechnologiesCyberInsight (天泽智云)ElmodisFalkonryFeelit TechnologiesGroundup.aiIPercept TechnologyMOVUSNanoprecise Sci CorpNeuron SoundwareOnePredict (원프레딕트)RotomateSamoticsTimeseer.AIViking AnalyticsGigatonAISing (株式会社エイシング)All Sense Technology / Quanying (全应科技)Basetwo AIIntelecyMogulinker / Mushroom IoT (蘑菇物联)Profet AI (杰倫智能科技)FourJaw Manufacturing AnalyticsGuidewheelMIPU Predictive HubcraftworksIdein (Idein株式会社)IOTech SystemsNeuCloud (寄云科技)Ulala Lab (울랄라랩)
underneath every row where the data lives
LXInfrastructure & datafactory edge and on-prem data centre, cloud, UNS and historian, MLOps for plants22 primary +7 also here
BarbaraCirrus Link SolutionscraftworksCybusFactryFlow SoftwareGoodLinker (谷林運算)HighByteHiveMQi-flowIdein (Idein株式会社)IOTech SystemsLitmusN3uronNeuCloud (寄云科技)Prosys OPCQuartic.aiRhize (formerly Libre)Ulala Lab (울랄라랩)United Manufacturing HubZEDEDAZerynthCyberInsight (天泽智云)Timeseer.AIAixtal (アイクリスタル株式会社)KelvinOden TechnologiesENLYZEFactbird
between companies the layer most tools do not reach
Upstream: company X−1suppliers, raw material, tier-2
purchase orders, call-offs, forecasts →
L6Supply networkdemand and capacity signals between companies, 1 to 3 year horizon19 primary +6 also here
Downstream: company X+1OEM, distributor, end demand
← demand signal, capacity reservation, 1 to 3 years out
The four we studied, in order of distance from the machine
Gigaton → Ethon → Circonomit → Operations1: from writing kiln setpoints to a search box over every document a worker reads.
L1 · L0
Gigatonwrites to the machine · closed loop
Control software for cement kilns that forecasts the process, tests setpoint moves on an ML model of the kiln and writes fuel, oxygen and kiln-speed setpoints to the plant's control system.
L3 · L2 · L1
Ethonreads sensors · recommends, human acts
A factory analytics platform that links production data in a process knowledge graph and runs causal root-cause analysis; engineers read the result and act on it.
L4
Circonomitreads MES / ERP · recommends, human acts
A planning tool in which a mid-sized manufacturer models its production decisions, simulates scenarios and optimises order sequence, capacity and stock with mathematical solvers.
L5
Operations1reads MES / ERP · human-facing only
Digital work instructions, checklists, audits and training for the shop floor, with an assistant that answers workers from uploaded manuals with page citations.
Layer × who closes the loop
Each dot is one company at its primary layer. The vertical position says who acts on the output: the machine itself at the top, a human reading a report at the bottom. The studied four are labelled. Hover for names, click to jump to the row.
How to read it
The further a product sits from the machine, the less it needs a plant integration and the more it competes with general software. L0 and L1 need an OT connection and a champion in maintenance or process engineering. L3 and L4 read what the MES and ERP already hold. L5 is bought by quality, HR and operations and could in principle run on any document store.
Two of the four working placements moved once the notes were reread. Gigaton forecasts the kiln 15 to 60 minutes ahead and writes setpoints to the control system, with soft sensors and sensor-fault flagging as inputs rather than the product, so it sits at L1 with L0 behind it. Circonomit models order sequence, make-or-buy and stock on a six to eight week horizon from ERP exports; the only multi-site case in the notes is one company's German and US plants, and nothing models a supplier or a customer. So the supply-network column stays empty of the four.
The right-hand column of the map is the gap this map is built to show: almost every startup models one plant. Very few tools carry a demand signal from company X+1 into company X's capacity plan the way Nvidia and TSMC do by contract with their suppliers. The "Signal between companies" tab works through where that could sit.
Counting primary placements, the crowded rows are infrastructure, the supply network and planning, and the between-companies row is crowded with sourcing and supplier-risk tools rather than with anything that moves a demand signal into a capacity plan. The thin cell is closed loop at L1 from a small team, because a plant will not hand a start-up write access to a machine before years of reference customers. Gigaton is in that cell, which is the strongest and the riskiest position of the four: the documented case so far set targets on an existing ABB controller rather than driving the kiln directly.
142 companies · click a layer in the map or the left column to filter
Companies by layer
layer
region
machine link
loop
search
| company | layers | machine link · loop | people | what it does · what was verified | read |
|---|---|---|---|---|---|
| Gigaton studied | L1 L0 | writes to the machine | 32 LinkedIn 11-50 range, checked 2026-11-03 (uk.linkedin.com/company/gigaton-technologies); 32 is the number of named people on gigaton.co/about before 'Show more' as read 2026-10-26, a lower bound, not a headcount | Self-learning control software for cement kilns that forecasts the process, simulates setpoint moves on an ML twin and writes fuel, oxygen and kiln-speed setpoints to the plant's DCS or APC. | gigaton.co gigaton.co gigaton.co |
| Ethon studied | L3 L2 L1 | reads sensors | 50 LinkedIn 51-200 range, checked 2026-11-03 (linkedin.com/company/ethonai); 'about 50 people' per a Siemens page cited in our 2026-10-08 report; an older unverified snippet said about 31 | A factory analytics platform that unifies production data in a process knowledge graph and runs causal root-cause analysis, setpoint recommendations and camera inspection, delivered by forward-deployed engineers. | ethon.com ethon.com ethon.com |
| Circonomit studied | L4 | reads MES / ERP | ? LinkedIn 2-10 range, checked 2026-11-03 (de.linkedin.com/company/circonomit); a June 2024 article said six employees; seven openings were live in Sept 2026; no current integer found | A web app in which a mid-sized manufacturer models its production decisions as connected blocks, simulates scenarios and optimises order sequence, capacity and stock with MILP and CP-SAT solvers. | circonomit.com circonomit.com circonomit.com |
| Operations1 studied | L5 | reads MES / ERP | 54 LinkedIn 51-200 range, checked 2026-11-03 (linkedin.com/company/operations1); 54 members shown on that page when read 2026-09-24; careers page says '50+' | Connected-worker SaaS that replaces paper on the shop floor with digital work instructions, checklists, audits and training, now with a RAG assistant that answers workers from uploaded manuals with page citations. | operations1.com operations1.com operations1.com |
| Amiral Technologies | L0 | reads sensors | 16 LinkedIn 11-50 range, 16 members listed, checked 2026-11-03 | DiagFit, software that learns normal behaviour from equipment sensor time series and flags anomalies and fault diagnoses for predictive maintenance. | amiraltechnologies.com linkedin.com |
| CyberInsight (天泽智云) | L0 L3 LX | reads sensors | 120 IoT ONE / Asia Growth Partners supplier profile band 51-200 (https://asiagrowthpartners.com/supplier/cyberinsight/compare/v3471), read 2026-11-03; a search snippet from the same profile family gave 122, not re-opened | Edge intelligence hardware (iEC, iGW gateways) and the EdgePro platform for equipment condition monitoring, anomaly detection, health management and remaining-life prediction on industrial machines. | pitchhub.36kr.com asiagrowthpartners.com |
| Elmodis | L0 L3 | reads sensors | 14 LinkedIn 11-50 range, 14 members listed, checked 2026-11-03 | Hardware and software that monitors industrial electric machines by sampling motor current and voltage at high frequency and deriving performance and health KPIs with ML. | elmodis.com linkedin.com |
| Falkonry | L0 L3 | reads sensors | 30 LinkedIn 11-50 band (midpoint used), checked 2026-11-03 | Time Series AI platform that ingests PLC and sensor data for multivariate anomaly detection, pattern recognition and root-cause analysis at the edge and in the cloud. | falkonry.com falkonry.com linkedin.com |
| Feelit Technologies | L0 | reads sensors | 32 LinkedIn 11-50 range, 32 members listed, checked 2026-11-03 | Printed nanotech sticker sensors with a wireless edge device and analytics that watch pipes, valves, steam traps, pumps and heat exchangers for leaks, thinning and deformation. | feelit.tech linkedin.com |
| Groundup.ai | L0 L3 | reads sensors | 30 LinkedIn 11-50 band (midpoint used), checked 2026-11-03 | IoT sound and vibration sensors with an AI platform (GINA) for anomaly detection and root-cause analysis on industrial machines, sold as cognitive maintenance. | groundup.ai thesaasnews.com linkedin.com |
| IPercept Technology | L0 | reads sensors | ? not found (LinkedIn page not located; press gives no figure) | AI health monitoring for CNC machine tools from a retrofit motion sensor, detecting wear and misalignment before breakdowns. | tech.eu vestbee.com |
| MOVUS | L0 L3 | reads sensors | 30 LinkedIn 11-50 band (midpoint used), checked 2026-11-03; older press figure 16 FTE plus 13 interns in July 2018 | FitMachine magnetically mounted vibration/temperature sensors and the PlantOS platform that turn machine data into prioritised maintenance recommendations. | movus.com.au startupdaily.net linkedin.com |
| Nanoprecise Sci Corp | L0 L3 | reads sensors | 127 press: https://betakit.com/nanoprecise-closes-series-c-to-predict-industrial-equipment-failures-with-ai (127 staff, March 2025, planning +50); LinkedIn 51-200 band, checked 2026-11-03 | Wireless vibration/acoustic/temperature sensors plus AI software that predicts failures of rotating industrial equipment and schedules maintenance. | betakit.com linkedin.com |
| Neuron Soundware | L0 L2 | reads sensors | 15 LinkedIn 11-50 range, 15 members listed, checked 2026-11-03 | nEdge sensor boxes plus nGuard software that detect machine faults and end-of-line product defects from sound and vibration with edge AI. | neuronsw.com linkedin.com cc.cz |
| OnePredict (원프레딕트) | L0 | reads sensors | 92 THE VC profile (thevc.kr/onepredict): 92 employees as of 2026-09, pension-registration based; read 2026-11-03 | GuardiOne predictive-diagnostics products (Turbo, Substation, Motor, Cyclone) that read equipment sensor signals to predict the timing and cause of failures in rotating machinery, transformers and motors. | onepredict.ai news.mtn.co.kr thevc.kr |
| Rotomate | L0 | reads sensors | 10 press: https://www.tamradar.com/funding-rounds/rotomate-pre-seed-2-1m (team grew from 2 to 10) | Condition-monitoring AI that reads existing vibration sensors plus CMMS and ERP context and returns prioritised maintenance decisions instead of raw alerts. | rotomate.ai tamradar.com |
| Samotics | L0 | reads sensors | 100 company about page says approximately 100 specialists; LinkedIn 51-200 range, checked 2026-11-03 | SAM4: condition and energy monitoring of motor-driven equipment from current and voltage measured in the electrical cabinet. | samotics.com samotics.com linkedin.com |
| Timeseer.AI | L0 LX | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 (a search snippet said 17, not opened) | Sensor data quality platform that detects, validates and fixes bad OT time-series before it reaches analytics or AI. | timeseer.ai linkedin.com |
| Viking Analytics | L0 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 | MultiViz: unsupervised-ML vibration analysis and condition monitoring that triages machines for maintenance analysts. | vikinganalytics.se vikinganalytics.se linkedin.com |
| AISing (株式会社エイシング) | L1 L0 | writes to the machine | 31 search snippet citing INITIAL profile: 31 (page itself shows headcount blurred); mynavi-agent.jp band 11-100 opened 2026-11-03 | Sells the AiiR edge-AI algorithm library: AiirMST learns and corrects machine control in real time on microcontrollers, AiirMSAT does on-device anomaly detection for predictive maintenance. | aising.jp initial.inc mynavi-agent.jp |
| Aixtal (アイクリスタル株式会社) | L1 LX | reads MES / ERP | ? not found on company pages; a search snippet said 'approximately 40 employees' without a traceable source, so left null | Nagoya University process-informatics startup whose MetaFactory digital twin and cloud service analyse line data with physics simulation plus ML to propose optimal manufacturing conditions across whole process chains. | aixtal.com aixtal.com thebridge.jp |
| All Sense Technology / Quanying (全应科技) | L1 L0 | writes to the machine | ? not found | 热电云 cloud-edge platform and ADMC intelligent control that optimises and automatically regulates industrial cogeneration and thermal power plants that supply steam and heat to industrial parks. | 36kr.com pitchhub.36kr.com |
| AMESA (formerly Composabl) | L1 | writes to the machine | 6 LinkedIn 2-10 range, checked 2026-11-03 (midpoint used) | Platform to build, train in simulation, validate and deploy autonomous agents that control industrial processes such as blending, reactors and CNC lines. | amesa.com linkedin.com |
| Basetwo AI | L1 L0 L3 | reads sensors | 30 LinkedIn 11-50 band (midpoint used), checked 2026-11-03 | AI agent platform that builds hybrid physics-plus-data digital twins of pharma and chemical processes to simulate, optimise setpoints, soft-sense quality and find root causes. | basetwo.ai basetwo.ai thelogic.co |
| Deep.Meta | L1 L2 | reads sensors | 9 LinkedIn 2-10 range, 9 members listed, checked 2026-11-03 | Explainable AI and a physics digital twin for steel mills that predict slab temperatures and metallurgical defects from existing sensor data and suggest parameter adjustments. | linkedin.com imperial.ac.uk |
| Fero Labs | L1 L3 | reads sensors | 29 LinkedIn 11-50 band, about 29 employees listed, checked 2026-11-03 | Explainable ML software (Diagnostics, Simulator, Production) that finds root causes and recommends optimised setpoints for continuous processes in steel, chemicals and cement. | ferolabs.com alleywatch.com linkedin.com |
| Imubit | L1 L3 | writes to the machine | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Closed-loop reinforcement-learning process control that writes moves through the plant's existing APC or DCS within engineer-set limits. | imubit.com linkedin.com |
| Intelecy | L1 L0 L3 | writes to the machine | 25 LinkedIn 11-50 range, checked 2026-11-03 (search snippet said about 34) | No-code industrial ML platform where process engineers build anomaly, forecast and recommendation models on live plant data, optionally streaming predictions back to control systems. | intelecy.com linkedin.com |
| Intellegens | L1 | documents only | 19 LinkedIn 11-50 range, 19 members listed, checked 2026-11-03 | Alchemite, machine learning for sparse experimental data that proposes formulations and process parameters and cuts the experiments needed in R&D and process development. | intellegens.com linkedin.com |
| INTERX (인터엑스) | L1 L2 L5 | reads sensors | 175 THE VC profile (thevc.kr/interx): 175 employees as of 2026-09; Wanted company page lists 169; read 2026-11-03 | Manufacturing AI that uses data collected from production equipment to compute appropriate operating conditions (process-condition optimisation) and to auto-judge product defects, sold as projects and a platform to automotive, semiconductor and battery makers. | bloter.net wanted.co.kr thevc.kr |
| Kelvin | L1 LX | writes to the machine | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Software that turns models and agents into governed actions on existing control systems, up to changing machine setpoints in closed loop within guardrails. | kelvin.ai linkedin.com |
| Limitless Labs (formerly LimitlessCNC) | L1 | writes to the machine | 46 LinkedIn 11-50 range, 46 members listed, checked 2026-11-03 | An AI CAM agent inside Mastercam, NX and Creo that generates CNC toolpaths, picks tools and sequences operations, which the programmer reviews and posts to the machine. | limitless-labs.ai linkedin.com calcalistech.com |
| Mogulinker / Mushroom IoT (蘑菇物联) | L1 L0 L3 | writes to the machine | ? not found | AIoT SaaS that connects factory utility stations (air compressors, chillers, pumps) and its 云智控 AI Cloud Control runs them unmanned with 10-30 percent energy savings. | mogulinker.com pitchhub.36kr.com |
| Oden Technologies | L1 L3 LX | reads sensors | 45 LinkedIn 11-50 band, 'View all 45 employees', checked 2026-11-03 (slug oden-technologies-ltd-) | Process AI that ingests machine and sensor data from continuous and discrete lines (extrusion, cable, ink, plastics) and gives front-line operators real-time parameter recommendations and factory analytics. | oden.io oden.io linkedin.com |
| plus10 | L1 L3 L2 | reads sensors | 25 startbase 11-50 band (https://www.startbase.com/organization/plus10/), checked 2026-11-03 | Shopfloor AI (Shannon, Darwin, Hopper) that reads high-frequency machine data and recommends parameter settings and fixes for availability, bottlenecks and injection-moulding scrap. | plus10.de startbase.com |
| Profet AI (杰倫智能科技) | L1 L3 L0 | reads MES / ERP | 50 press: https://www.technice.com.tw/uncategorized/36975/ (Feb 2023: 'from the current 50 to 100 people'); no newer figure found | No-code AutoML and the Domain Twin platform that let factory process, quality and supply-chain engineers build predictive models from structured production data for process optimisation, parameter recommendation, anomaly detection and predictive maintenance. | profetai.com profetai.com technice.com.tw |
| Smart Steel Technologies | L1 L4 L2 | reads sensors | 100 LinkedIn 51-200 range, checked 2026-11-03 (search snippets said 44 to 50+) | AI process-control and scheduling software for steel mills: temperature control, surface quality, casting, material allocation and schedulers. | smart-steel-technologies.com heraeus.com linkedin.com |
| TVARIT | L1 L2 | reads sensors | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Industrial AI (TiA) that predicts quality and energy outcomes in foundries and metal plants and recommends process parameter changes to operators. | tvarit.com linkedin.com |
| 24 Vision System | L2 | reads sensors | ? not found (LinkedIn page is a numeric id behind a login wall) | Neural-network visual inspection that checks every product for defects and configuration errors on just-in-sequence automotive lines. | 24vs.io magazin.positiv.cz |
| Adacotech (株式会社アダコテック) | L2 | reads sensors | 20 paiza.jp recruiter page: 20名 (2025/03現在), opened 2026-11-03; a search snippet claimed insured headcount fell to 13 by 2026-07, not verified | Non-deep-learning visual inspection AI (AdaInspector learns from good parts only, POKAMIRU assists manual inspection) that runs on CPU with few samples, from AIST image-analysis research. | adacotech.co.jp adacotech.co.jp paiza.jp |
| Deltaray | L2 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 | Inline 3D X-ray (3DAX) inspection of every part at production speed, with AI analysis of internal and external defects. | imec-int.com linkedin.com |
| Elementary | L2 L3 | reads sensors | 30 LinkedIn 11-50 band at linkedin.com/company/elementary-robotics (legacy slug, page shows current Elementary branding), midpoint used, checked 2026-11-03 | Sells AI-powered camera inspection stations and a QA platform that learn a product's quality standard and automatically flag defects on production lines. | elementaryml.com elementaryml.com linkedin.com |
| Frinks AI | L2 | reads sensors | 30 LinkedIn 11-50 band at in.linkedin.com/company/frinks, midpoint used, checked 2026-11-03 | Sells a no-code machine vision AI platform that automates inspection, compliance checks and shipment validation on production lines. | india.entrepreneur.com in.linkedin.com |
| Instrumental | L2 L3 | reads sensors | 75 directory: cyberdb.co estimate 51-100 employees, midpoint used (https://www.cyberdb.co/ai-vendor/instrumental/) | Sells a manufacturing data and AI platform for electronics assembly that combines line camera images with test and assembly data to intercept defects and find yield and throughput issues. | instrumental.com instrumental.com techcrunch.com |
| Kitov.ai | L2 | reads sensors | 26 LinkedIn 11-50 range, 26 members listed, checked 2026-11-03 | CAD-based automated visual inspection planning that combines AI with classical 3D vision to find cosmetic and assembly defects on production lines. | kitov.ai kitov.ai linkedin.com |
| Maddox AI | L2 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 (search snippet said about 30) | Machine-vision quality control software with collaborative annotation and model training, claiming over 80% fewer false rejects than classic vision. | maddox.ai linkedin.com |
| NeuroBot (矩视智能) | L2 | reads sensors | ? not found | Browser-based low-code machine vision platform (600+ algorithms, 2D/3D/video) for defect detection, OCR, measurement and localisation that engineers train and run without writing code. | nb-ai.com |
| Neurocle (뉴로클) | L2 | reads sensors | 41 THE VC profile (thevc.kr/neurocle): 41 employees as of 2026-09; read 2026-11-03 | No-code auto-deep-learning vision inspection software (Neuro-T to train models from images, Neuro-R runtime, Neuro-T Engine for MLOps integration) used by manufacturers to classify and locate product defects. | neuro-cle.com venturesquare.net thevc.kr |
| Polysense | L2 L1 | writes to the machine | ? not found (LinkedIn page not located; 2025 article names 3 founders only) | Inline camera quality inspection for food lines whose AutoControl module adjusts machine parameters when defects drift. | polysense.ai freshplaza.com tech.eu |
| RESONIKS | L2 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 | QCFlex: acoustic-fingerprint end-of-line testing cell that gives pass/fail on surface and internal defects in seconds. | resoniks.com linkedin.com |
| Ripik.ai | L2 L3 L1 | reads sensors | 92 Inc42 directory: 92 employees (https://inc42.com/company/ripikai/); LinkedIn 51-200 band, checked 2026-11-03 | Computer vision AI agents that watch kilns, furnaces, conveyors, molten metal and raw material feeds in heavy industry to flag anomalies and support process optimisation. | ripik.ai inc42.com inc42.com |
| sentin | L2 | reads sensors | 6 LinkedIn 2-10 range, checked 2026-11-03; startbase 1-10 | AI image analysis (sentin EXPLORER) for non-destructive testing: weld defects on X-ray and RGB images, wall thickness, surface inspection. | sentin.ai startbase.com linkedin.com |
| SixSense | L2 L3 | reads sensors | 35 sixsense.ai/about recruiter testimonial says team of 35 (undated); LinkedIn 11-50 band at linkedin.com/company/sixsenseai, checked 2026-11-03 | Sells an AI platform for semiconductor fabs and OSATs that classifies AOI defect images, predicts failures and links defects to tool and recipe data for root cause and yield. | sixsense.ai sixsense.ai techcrunch.com |
| Smartex | L2 | reads sensors | 80 LinkedIn 51-200 range, 80 members listed, checked 2026-11-03 | Camera hardware and computer vision fitted to knitting machines that detect fabric defects as they are produced. | smartex.ai linkedin.com tech.eu |
| Visionairy | L2 | reads sensors | 15 press: https://www.universite-paris-saclay.fr/en/news/visionairy-artificial-intelligence-improve-quality (team of 15, July 2024); LinkedIn 11-50 range with 9 members listed | End-to-end AI visual inspection platform (industrial PC, edge software, central analytics) using patented unsupervised models that learn from about 50 good-part images. | visionairy.io linkedin.com universite-paris-saclay.fr |
| Arch Systems | L3 L4 | reads sensors | 85 directory: builtin.com 85 total employees (https://builtin.com/company/arch-systems-inc/offices); LinkedIn 51-200 band, checked 2026-11-03 | Sells a factory intelligence platform that pulls machine, MES, ERP and SCADA data from electronics lines and uses AI agents to turn it into prescriptive actions for operators and managers. | archsys.io archsys.io linkedin.com |
| Braincube | L3 L1 | reads sensors | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Industrial AI platform for real-time process optimisation that finds best operating conditions from plant data and guides operators toward them. | braincube.com braincube.com linkedin.com |
| ENLYZE | L3 LX L2 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 (search snippet said about 15) | Manufacturing data platform with retrofit connectivity and a unified namespace, delivering automated OEE, parameter-level traceability and root-cause analysis. | enlyze.com linkedin.com |
| Fabriq | L3 L5 | documents only | 148 LinkedIn 51-200 range, 148 members listed, checked 2026-11-03 | Connected workforce platform that digitises Lean rituals (tier meetings, SQCDP, Gemba walks, audits, problem solving) and uses AI to turn field reports into standardised operational knowledge. | fabriq.tech linkedin.com |
| Factbird | L3 L5 LX | reads sensors | ? not found (LinkedIn page not located; snippets conflict between 11 and 50-100) | Plug-and-play edge sensors and PLC links feeding OEE and downtime analytics, plus connected-operations workflows and skills modules. | factbird.com iotforall.com |
| FourJaw Manufacturing Analytics | L3 L0 | reads sensors | 29 press: own about page https://fourjaw.com/about states 29 employees; LinkedIn 11-50 range with 35 members listed | Plug-and-play MachineLink hardware and software that turn live machine activity plus operator and job data into OEE, downtime and capacity-loss analytics. | fourjaw.com fourjaw.com linkedin.com |
| Guidewheel | L3 L0 | reads sensors | 125 LinkedIn 51-200 band, midpoint used, checked 2026-11-03 | Sells a FactoryOps platform that clips power sensors onto any machine to track real-time utilisation and downtime and alerts teams to unusual machine patterns. | guidewheel.com guidewheel.com linkedin.com |
| MIPU Predictive Hub | L3 L0 | reads sensors | 87 LinkedIn 51-200 range, 87 members listed, checked 2026-11-03 | Industrial AI platform for machine health, process intelligence and energy optimisation with digital twins and real-time energy monitoring for plants. | mipu.eu linkedin.com |
| Raven.ai | L3 | reads sensors | 125 LinkedIn 51-200 band at ca.linkedin.com/company/ravenai, midpoint used, checked 2026-11-03 | Sells OEE and downtime software that pulls machine data to auto-tag downtime and prompts operators on HMIs to label the cause, giving supervisors a contextualised production timeline. | raven.ai raven.ai ca.linkedin.com |
| Shuyi Link (数益工联) | L3 | reads sensors | ? not found | Digital lean factory software for discrete manufacturers that captures shop-floor and machine data to drive value-stream, efficiency and quality improvement. | pitchhub.36kr.com shuyilink.com |
| Smart Craft (株式会社Smart Craft) | L3 L5 L4 | reads MES / ERP | 11 HERP careers company profile: 11名, opened 2026-11-03 (INITIAL snippet said 9) | Cloud MES for small and mid-size factories: tablet work instructions, actuals and defect capture, progress and OEE-style monitoring, quality records, inventory, IoT equipment data, and a natural-language AI assistant over the collected data. | smartcraft.jp herp.careers thebridge.jp |
| TEEPTRAK | L3 | reads sensors | 37 LinkedIn 11-50 range, 37 members listed, checked 2026-11-03 | PerfTrak, QualTrak and PaceTrak boxes with plug-and-play sensors that capture stops, micro-stops, cycle times and scrap for real-time OEE, with operators logging stop causes. | teeptrak.com linkedin.com |
| akeno | L4 | reads MES / ERP | 25 LinkedIn 11-50 range, checked 2026-11-03 | Autonomous production planning and scheduling for process manufacturers, reading and writing back to ERP and MES. | akeno.ai thesaasnews.com linkedin.com |
| ALGO ARTIS (株式会社ALGO ARTIS) | L4 | reads MES / ERP | 109 HERP careers company profile: 109名, opened 2026-11-03 (INITIAL snippet said 103) | Optimisation-algorithm company (DeNA spin-off) selling Optium Plant production-planning optimisation and Planium, a packaged production scheduler for the chemical industry, plus ship and fuel planning for energy. | algo-artis.com initial.inc herp.careers |
| Aolin Technology (傲林科技) | L4 L3 | reads MES / ERP | ? not found | Enterprise-level digital twin that simulates a whole plant's operations to recommend procurement timing and volumes, inventory levels and production plans, aimed at steel and other process industries. | pitchhub.36kr.com tmtpost.com |
| Archlet | L4 | documents only | 100 LinkedIn 51-200 range, checked 2026-11-03 (search snippet said about 32) | AI-native eSourcing with scenario optimisation and Spark agents for RFQs, auctions and bid analysis. | archlet.io linkedin.com |
| Arkestro supply network | L4 L6 | reads MES / ERP | 125 LinkedIn 51-200 band (about 140 profiles listed), checked 2026-11-03; Welcome to the Jungle says 101-200, bouncewatch says 102; 125 is the band midpoint | Predictive procurement platform that uses machine learning, game theory and behavioural science to simulate sourcing events, suggest prices and automate supplier negotiations. | arkestro.com arkestro.com linkedin.com |
| Flexciton | L4 | reads MES / ERP | 36 LinkedIn 11-50 range, 36 members listed, checked 2026-11-03 | Mathematical-optimisation planning and scheduling for semiconductor fabs (Flex Planner, Flex Local, Flex Global) that autonomously schedules lots to tools. | flexciton.com flexciton.com linkedin.com |
| Flowlity supply network | L4 L6 | reads MES / ERP | 28 LinkedIn 11-50 range, 28 members listed, checked 2026-11-03 | AI supply chain planning (demand forecasting, inventory optimisation, supply and S&OP planning, collaborative planning with suppliers) built to be resilient to uncertainty. | flowlity.com linkedin.com |
| Fulcrum | L4 L3 | reads MES / ERP | 75 LinkedIn 51-200 band, checked 2026-11-03; FundedIQ says 51-100; 75 is the midpoint of the narrower band | Cloud ERP/MES for small and mid-sized job shops with Autoschedule production scheduling, job costing, quality and inventory, plus Archie, a built-in AI chat, builder and agent layer. | fulcrumpro.com linkedin.com vcnewsdaily.com |
| Impactive AI (임팩티브AI) | L4 L6 | reads MES / ERP | 36 THE VC profile (thevc.kr/impactiveai): 36 employees as of 2026-09; read 2026-11-03 | Deepflow, a SaaS demand-forecasting and inventory tool for manufacturers that predicts 6-12 month product demand and raw-material prices, flags stock-out and overstock items, and proposes order quantities for purchasing teams. | impactive-ai.com thevc.kr hellot.net |
| InterAgent (塔台智能網絡) supply network | L4 L6 | reads MES / ERP | ? not found | SARA, a cloud AI production-scheduling (APS) system for made-to-order SME factories that takes orders, inventory and capacity from Excel, ERP (SAP, Oracle, Digiwin) or MES and generates and re-generates schedules, delivery-date answers and shortage and bottleneck warnings in minutes. | interagent.io interagent.io meet.bnext.com.tw |
| Keelvar | L4 | reads MES / ERP | 120 LinkedIn 51-200 range, 120 members listed, checked 2026-11-03 | Agentic sourcing software that runs procurement events end to end, from intake to award, with optimisation across direct materials, MRO, packaging and logistics. | keelvar.com keelvar.com linkedin.com |
| Oplit | L4 | reads MES / ERP | 38 LinkedIn 11-50 range, 38 members listed, checked 2026-11-03 | Production planning and scheduling platform (Oplit Core and the AI-first Oplit Kosmos with optimisation and reinforcement learning) fed by native ERP and MES connectors. | oplit.com linkedin.com |
| paretos | L4 | reads MES / ERP | ? not found on an opened page (search snippets: 11-50 or about 40; LinkedIn slug resolved to an unrelated nonprofit) | Socrates decision-intelligence system: demand forecasting, stock optimisation and production scheduling recommendations on ERP, CRM and BI data for the Mittelstand. | paretos.com startbase.de |
| Skydisc (株式会社スカイディスク) | L4 | reads MES / ERP | 27 HERP careers company profile: 27名, opened 2026-11-03 (INITIAL snippet said 32) | AI SaaS production scheduler 最適ワークス that turns order data into optimal machine and staff allocation plans for factories. | skydisc.jp herp.careers thebridge.jp |
| SupplyET | L4 L6 | reads MES / ERP | 4 press: https://startupcity.hamburg/news-events/stories/supplyet-uses-ai-to-future-proof-production-networks (four core team members) | AI capacity and scenario planning across a multi-site production network, flagging bottlenecks before they occur. | startupcity.hamburg |
| Tacto supply network | L4 L6 | reads MES / ERP | 137 press: https://getlatka.com/companies/en.tacto.ai (137 employees, Sep 2025, citing LinkedIn) | Procurement AI for industrial mid-caps: supplier, spend, sourcing and compliance intelligence with RFQ, should-cost and negotiation agents over ERP data and external signals. | tacto.ai tacto.ai getlatka.com |
| Yuanmu Intelligence (圆木智能) | L4 | reads MES / ERP | ? not found | 今天排, an LLM-based scheduling agent for job shops that reads orders, routings, machine calendars, subcontracting and live progress and regenerates the production plan in seconds when orders or machines change. | jintianpai.com news.pedaily.cn |
| Zentio | L4 L5 L3 | reads MES / ERP | ? not found | Agentic manufacturing OS: specialist agents for scheduling, staffing and master data over ERP and MES data, with an assistant for the shop floor. | zentio.ai invest-in.berlin htgf.de |
| Augmented Industries | L5 | documents only | 25 LinkedIn 11-50 range, checked 2026-11-03 | Flow Tool: AI that turns manuals, videos and technician knowledge into step-by-step work instructions and troubleshooting guides. | tech.eu linkedin.com |
| Augmentir | L5 L3 | reads MES / ERP | 38 LinkedIn 11-50 band, 'View all 38 employees', checked 2026-11-03 | Sells an AI-native connected worker platform: digital work instructions, skills tracking, an industrial GenAI assistant (Augie) and no-code AI agents for frontline manufacturing operations. | augmentir.ai augmentir.ai augmentir.ai |
| Azumuta | L5 L3 | documents only | 25 LinkedIn 11-50 range, checked 2026-11-03 | Connected-worker platform combining digital work instructions, audits, training matrices and quality checks, with AI document conversion. | azumuta.com tools.prnewswire.com linkedin.com |
| Datch | L5 L3 | reads MES / ERP | 33 LinkedIn 11-50 band, 'View all 33 employees', checked 2026-11-03 | Sells an AI Diagnostic Agent and mobile EAM front end that lets maintenance technicians and operators troubleshoot equipment by voice or chat using work orders, manuals and schematics. | datch.io datch.io linkedin.com |
| Dirac | L5 L4 | documents only | 39 LinkedIn 11-50 band, 'View all 39 employees', checked 2026-11-03 | Sells BuildOS, an AI build platform that turns CAD models into animated work instructions, manufacturing BOMs and bills of process executed at the station. | diracinc.com diracinc.com linkedin.com |
| Edmund AI | L5 L3 | reads MES / ERP | 18 LinkedIn 11-50 range, 18 members listed, checked 2026-11-03 | Industrial debugging assistant whose agents read PLC programs, robot programs, ePLAN schematics, SQL/MES data and manuals so maintenance teams find line-stop root causes via chat. | edmundai.com linkedin.com en.ain.ua |
| Glartek | L5 | documents only | 33 LinkedIn 11-50 range, 33 members listed, checked 2026-11-03 | AI-native, mobile-first EHSQ and connected-worker platform with a copilot agent, voice input, risk assessment, root-cause analysis and procedure authoring for frontline teams. | glartek.com linkedin.com |
| HiScene (亮风台) | L5 | documents only | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | AR glasses and software that overlay step-by-step work instructions, equipment information and remote expert guidance on the shop floor, using computer vision. | hiscene.com hiscene.com linkedin.com |
| Intoware (WorkfloPlus) | L5 | documents only | 16 LinkedIn 11-50 range, 16 members listed, checked 2026-11-03 | WorkfloPlus digital work instructions and inspections on mobile, desktop and smart glasses, with FloGenAI drafting instructions from a process description. | intoware.com linkedin.com |
| Kaminashi (株式会社カミナシ) | L5 | documents only | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Frontline-worker SaaS for factories: digital checklists and production records, maintenance, and training with AI-generated manuals and translation. | kaminashi.jp linkedin.com |
| Leucine | L5 L3 | reads MES / ERP | 116 LinkedIn 51-200 band, 'View all 116 employees', checked 2026-11-03 | Sells an AI platform for pharma manufacturing: AI-powered batch execution and electronic batch records, digital logbooks, QMS with an AI deviation investigator, and lab execution. | leucine.io linkedin.com |
| LIGHTz (株式会社LIGHTz) | L5 | documents only | 51 search snippet (INITIAL-style listing): 51; not seen on an opened page, treat as unverified | Structures the tacit know-how of veteran engineers into 'Brain Models' and sells blooplinter, a knowledge platform with 3D-CAD similarity search, automated requirement checks and past-issue matching for design and maintenance teams. | lightz-inc.com lightz-inc.com thebridge.jp |
| Skillnote (株式会社スキルノート) | L5 | documents only | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Skills and training management system for manufacturing that maps worker skills to stations for placement and know-how transfer, now with an AI agent. | skillnote.jp corp.skillnote.jp linkedin.com |
| Squint | L5 L3 | documents only | 79 LinkedIn 51-200 band, 'View all 79 employees', checked 2026-11-03 | Sells an industrial AI platform that captures tribal knowledge and manuals into AR/mobile procedures, answers operator questions, and runs industrial agents such as video-based time-and-motion analysis. | squint.ai squint.ai prnewswire.com |
| Team Solution (팀솔루션) | L5 | documents only | 17 THE VC figure (17 employees, Sep 2026) seen only in a Bing result snippet; the profile page itself could not be opened, so treat as weakly sourced | DXE digital-twin modules for heavy industry, including DXE Manual AI (launched Oct 2026) which turns PDF equipment manuals and 2D drawings into step-aligned 3D work manuals with animations, editable by natural-language prompts, cutting manual creation from about a month to a week. | timsolution.io newswire.co.kr dataeconomy.co.kr |
| Testify | L5 | documents only | 25 LinkedIn 11-50 range, checked 2026-11-03 | Digital checklists, inspections and audit documentation for plants, with an MCP server so LLM assistants can query and report on the data. | testify.io testify.io linkedin.com |
| ChAI supply network | L6 L4 | documents only | 30 LinkedIn 11-50 range, 30 members listed, checked 2026-11-03 | AI raw-material price forecasting and market intelligence, plus price-risk insurance, for manufacturers' procurement teams. | chaipredict.com linkedin.com |
| Cofactr supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Electronics supply-chain platform combining AI component intelligence, procurement execution across many suppliers and distributors, and ITAR-registered warehousing/kitting for hardware builders. | cofactr.com linkedin.com |
| Cosmo Tech supply network | L6 L4 | reads MES / ERP | 70 press: own about page https://cosmotech.com/about-us/ states '+70 Employees'; LinkedIn 51-200 range with 120 members listed | Prescriptive simulation digital twins of whole supply chains and asset networks that let planners test scenarios and pick the best production and investment plan. | cosmotech.com cosmotech.com linkedin.com |
| Didero supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | AI agents on top of a manufacturer's ERP that read supplier emails, POs and packing lists and update orders, confirmations and exceptions automatically. | didero.ai linkedin.com techcrunch.com |
| Enmovil supply network | L6 L4 | reads MES / ERP | 183 directory: Inc42 (https://inc42.com/company/enmovil/) 183 employees; Built In says 150; LinkedIn page not located | AI supply chain planning and visibility platform (CADDIE) that connects to ERP, TMS, WMS and carriers for demand forecasting, inventory planning, dispatch and load optimisation and disruption tracing across suppliers, plants and customers. | enmovil.ai entrackr.com builtin.com |
| Kavida.ai supply network | L6 L4 | reads MES / ERP | 30 LinkedIn 11-50 range, checked 2026-11-03 (midpoint used) | AI procurement agents ('Agent PO') that chase order confirmations, changes and documents with a manufacturer's suppliers after the PO is placed. | kavida.ai linkedin.com |
| Kodiak Hub supply network | L6 L4 | documents only | 15 company about page says 15 employees; LinkedIn shows 51-200 range, checked 2026-11-03 | AI-assisted supplier relationship management: onboarding, performance, risk and audit data across supplier tiers for procurement teams. | kodiakhub.com kodiakhub.com linkedin.com |
| KREATIZE supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Networked procurement platform that pools capacity from 70+ specialist supplier factories and lets industrial OEMs order mechanical parts through their own ERP, with KREATIZE matching, producing and QC-ing across the network. | kreatize.com linkedin.com |
| LevaData supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Direct-materials sourcing platform that benchmarks component prices across buyers and distributors, scores supply risk from lead-time and lifecycle data, and assists supplier negotiations. | levadata.com linkedin.com |
| LingMao SCM (领猫SCM) supply network | L6 L4 | reads MES / ERP | 120 press: 36Kr Aug 2021 article says team grew from ~50 to 120 (https://www.36kr.com/p/1361234622989704); no newer figure found | Apparel supply-chain SaaS that connects brands with their contract factories and fabric suppliers for planning, sourcing, production tracking and settlement across the network. | 36kr.com pitchhub.36kr.com |
| Lytica supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Electronic-component spend analytics and risk intelligence built from anonymised cross-buyer transaction data, with agentic AI (Neo, Rik, Lisa) for negotiation and supplier strategy. | lytica.com linkedin.com |
| Nulogy supply network | L6 L3 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Software that gives brand manufacturers real-time capacity, inventory and production visibility across their external co-packers and contract manufacturers, feeding constraint-aware plans into tools like Kinaxis Maestro. | nulogy.com nulogy.com linkedin.com |
| Pactum supply network | L6 L4 | reads MES / ERP | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Autonomous procurement agents that negotiate deals with suppliers (including a single-source negotiation agent) and place purchase orders on behalf of large enterprises. | pactum.com linkedin.com |
| Partsimony supply network | L6 L4 | documents only | 6 LinkedIn 2-10 range, checked 2026-11-03 (midpoint used) | Manufacturing intelligence layer for hardware supply chains that links engineering decisions to a company's network of contract manufacturers, flagging supplier problems early. | partsimony.com linkedin.com |
| Resilire (株式会社Resilire) supply network | L6 | documents only | 36 search snippet citing INITIAL: 36 (as of 2025-09-30); another snippet said 28; company page fetch failed, so not verified | Supply-chain risk management cloud that maps tier-2 and deeper suppliers, sites and parts as a tree, detects disasters and geopolitical events in real time, and uses AI to assess how macro changes hit the network. | thebridge.jp resilire.jp |
| SourceDay supply network | L6 L4 | reads MES / ERP | 106 press: https://builtinaustin.com/company/sourceday-inc ('106 Total Employees', checked 2026-11-03) | Supplier collaboration platform that syncs purchase-order dates, prices and quantities between a buyer's ERP and its suppliers, with AI handling confirmations and changes. | sourceday.com builtinaustin.com |
| Sourcemap supply network | L6 | documents only | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | N-tier supply chain mapping and traceability software that collects verified supplier data down to raw-material origin at BOM/part level for tariff, forced-labour and ESG due diligence. | sourcemap.com sourcemap.com linkedin.com |
| SyncFab supply network | L6 L4 | documents only | 30 LinkedIn 11-50 range, checked 2026-11-03 (midpoint used) | Procurement platform that connects aerospace, space and defense OEM buyers with a vetted network of compliant US suppliers, managing quotes, orders and compliance across companies. | syncfab.com linkedin.com |
| ThroughPut Inc supply network | L6 L4 | reads MES / ERP | 28 directory: Latka (https://getlatka.com/companies/throughput) 28 employees as of Oct 2024 | AI supply chain analytics software for demand sensing, capacity planning, inventory rebalancing and logistics planning on top of a client's existing ERP and operational data. | throughput.world throughput.world getlatka.com |
| Barbara | LX | reads sensors | 86 LinkedIn 11-50 range (Spanish page: 'De 11 a 50 empleados') but 86 members listed, checked 2026-11-03; band and member count disagree | Industrial edge platform that lets OT teams deploy, run and monitor containerised apps and AI models on plant edge devices, with industrial connectors and MLOps. | barbara.tech es.linkedin.com |
| Cirrus Link Solutions | LX | reads sensors | ? LinkedIn 2-10 range, checked 2026-11-03 (https://www.linkedin.com/company/cirrus-link-solutions) | MQTT and Sparkplug infrastructure for plants: MQTT modules for Ignition (Engine, Distributor, Recorder, Injector), Chariot MQTT server and cloud IoT bridges. | cirrus-link.com linkedin.com |
| craftworks | LX L2 L0 | reads sensors | 25 LinkedIn 11-50 range, checked 2026-11-03 | navio MLOps platform to deploy and monitor ML models in plants, plus navio VISION anomaly inspection and predictive quality and maintenance projects. | craftworks.ai linkedin.com |
| Cybus | LX | reads sensors | ? LinkedIn 51-200 range, checked 2026-11-03 (https://www.linkedin.com/company/cybus) | Cybus Connectware, an on-premise factory data hub that connects OT protocols to IT systems and distributes standardised plant data in real time. | cybus.io linkedin.com |
| Factry | LX L3 | reads sensors | 59 LinkedIn 11-50 range with 59 members listed, checked 2026-11-03 (https://www.linkedin.com/company/factry.io/) | Factry Historian (plant time-series collection and analytics) and FactryOS production execution software for process plants. | factry.io factry.io linkedin.com |
| Flow Software | LX L3 | reads sensors | ? LinkedIn 11-50 range, checked 2026-11-03 (https://www.linkedin.com/company/flow-software-pty-ltd) | Industrial information and DataOps platform: Flow Infohub (KPI calculations and governance), Timebase Historian (free time-series historian) and Timebase Atlas (knowledge modelling). | flow-software.com linkedin.com |
| GoodLinker (谷林運算) | LX L3 | reads sensors | ? not found (a 2018 figure of 20 appeared in a search snippet only) | LESI edge gateways, tower-light sensors and screen-data extractors that retrofit legacy machines, feeding a cloud command-centre dashboard (and since 2024 the GenAIoT analytics platform on AWS) that shows utilisation, downtime and alarms for SME factories. | goodlinker.io goodlinker.io digitimes.com.tw |
| HighByte | LX | reads sensors | ? LinkedIn 11-50 range, checked 2026-11-03 (https://www.linkedin.com/company/highbyte) | HighByte Intelligence Hub, edge-native no-code industrial DataOps software that models, contextualises and delivers OT data to IT and AI systems. | highbyte.com linkedin.com |
| HiveMQ | LX | reads sensors | ? LinkedIn 51-200 range, checked 2026-11-03 (https://www.linkedin.com/company/hivemq-gmbh) | MQTT broker and industrial data platform (HiveMQ Broker, HiveMQ Edge, extensions) used as the backbone of unified-namespace architectures in factories. | hivemq.com linkedin.com |
| i-flow | LX | reads sensors | ? not found on a current page (Techstars profile shows 1-10, likely stale; LinkedIn slug resolved to an Indian firm) | Industrial unified namespace (hub, broker, edge) with 200+ connectors from S7/OPC UA to SAP and MES, plus an AI agent for one-click data modelling. | i-flow.io i-flow.io jobs.techstars.com |
| Idein (Idein株式会社) | LX L0 L2 | reads sensors | 55 company profile deck on speakerdeck: 55名 (2025年10月現在), and HERP profile 55名, both opened 2026-11-03 | Actcast edge-AI platform that deploys, monitors and updates camera and sensor AI on fleets of Raspberry Pi-class devices on site, with factory apps for inspection, stock counting, worker behaviour analysis, anomaly detection and meter reading. | idein.jp idein.jp speakerdeck.com |
| IOTech Systems | LX L0 | reads sensors | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Edge data platform (Edge Central, commercial EdgeX Foundry) that connects industrial devices and makes their data ready for edge AI. | iotechsys.com linkedin.com |
| Litmus | LX | reads sensors | 100 LinkedIn 51-200 range, checked 2026-11-03; litmus.io page metadata states '100 employees' | Industrial edge data platform (Litmus Edge, Edge Manager, Unify UNS, Data Catalog, MCP Server) that connects plant assets, standardises OT data and deploys analytics and AI models at the edge. | litmus.io linkedin.com |
| N3uron | LX | reads sensors | ? LinkedIn 11-50 range, checked 2026-11-03 (https://www.linkedin.com/company/n3uron) | Modular industrial DataOps and edge platform (protocol connectors, MQTT/Sparkplug, historian and web visualisation) linking plant systems, SCADA and cloud. | n3uron.com linkedin.com |
| NeuCloud (寄云科技) | LX L0 L5 | reads sensors | ? not found (one database snippet said 25-100, source page not opened) | NeuSeer industrial data intelligence platform: IIoT connectivity, time-series store and analytics, edge computing, low-code apps and the Neu.ai industrial agent built on an enterprise ontology. | neucloud.cn |
| Prosys OPC | LX | reads sensors | ? LinkedIn 11-50 range, checked 2026-11-03 (https://www.linkedin.com/company/prosys-opc-ltd) | OPC UA software vendor: OPC UA Forge gateway, SDKs for Java, C/C++, .NET and Delphi, Simulation Server and Monitor. | prosysopc.com linkedin.com |
| Quartic.ai | LX L1 L3 | reads sensors | 70 LinkedIn 51-200 band, 'View all 70 employees', checked 2026-11-03 | Sells an industrial DataOps and decision-intelligence platform (iLuminator, eXponence) with ML applications for process optimisation, batch multivariate analysis, reliability and PAT in process manufacturing. | quartic.ai quartic.ai linkedin.com |
| Rhize (formerly Libre) | LX | reads MES / ERP | ? LinkedIn 11-50 range, checked 2026-11-03 (https://www.linkedin.com/company/rhizemdh/) | Rhize Manufacturing Data Hub, a real-time event-driven ISA-95 data platform that gives plants a standardised, application-ready data model (headless, GraphQL). | rhize.com linkedin.com |
| Ulala Lab (울랄라랩) | LX L3 L0 | reads sensors | 18 THE VC profile (thevc.kr/ulalalab): 18 employees as of 2026-09; read 2026-11-03 | WimFactory, a low-cost IoT smart-factory platform for SME plants: WICON gateways and add-on sensors collect equipment status data, a cloud dashboard shows real-time machine and line status, and ML models raise malfunction and emergency alerts. | thevc.kr venturesquare.net venturesquare.net |
| United Manufacturing Hub | LX L3 | reads sensors | 40 website says 11-50, LinkedIn shows 51-200 range, checked 2026-11-03; 40 is a midpoint guess between the two | Open-source, ISA-95 unified-namespace data platform that connects PLCs, sensors and ERP into a real-time hub for KPIs, condition monitoring and AI agents. | umh.app tech.eu linkedin.com |
| ZEDEDA | LX | reads sensors | 125 LinkedIn 51-200 range, checked 2026-11-03 (midpoint used) | Edge orchestration platform to deploy, update and secure AI models and applications on distributed edge hardware, including factory floors. | zededa.com zededa.com linkedin.com |
| Zerynth | LX L3 | reads sensors | 45 LinkedIn 11-50 range, 45 members listed, checked 2026-11-03 | Plug-and-play industrial IoT and AI platform that connects any machine in under three hours and feeds OEE, energy and predictive analytics. | linkedin.com |
the layer outside the factory
How a demand signal travels between companies, and where it breaks
Nvidia tells TSMC and the memory makers what it will need two years out, with money attached. A Mittelstand tier-2 learns about a volume change when the call-off arrives. The space between is the L6 row on the map.
flowchart LR D["End demand
hyperscaler, car buyer"] -->|"orders 12 to 36 months out,
prepayments, LTAs"| O["OEM / brand
company X+1"] O -->|"call-offs, EDI DELFOR / DELJIT,
weeks to months"| X["The plant
company X"] X -->|"purchase orders, MRP,
days to weeks"| U["Tier-2 / material
company X−1"] U -. "capacity, lead time, allocation" .-> X X -. "confirmed capacity" .-> O O -. "forward capacity commitment
(rare below the top tier)" .-> X
How a demand signal travels from end buyer to raw material, and where it breaks
Written 2026-11-03; every source [n] was opened that day and is listed at the end. Section 3 is my reading.
1. The chain as it works today
Capital commitments, 1 to 3 years. At the top of the AI hardware chain, the signal is money. Nvidia's 10-Q for the quarter ended 26 July 2026 reports supply and capacity commitments of $279 billion, up from $119 billion a quarter earlier, about $90bn a year through FY2029, "primarily memory and manufacturing facilities"; many agreements "may be cancelable, rescheduled, or adjustable" before firm orders [1]. Its 10-K confirms it has "paid premiums, provided deposits" to secure capacity [2]; the balance I haven't verified. The receiving end confirms it: SK hynix declared 2026 HBM supply talks complete in October 2025 [3] and on 28 October 2026 reported long-term agreements with 14 customers covering more than 50% of 2027 and 2028 capacity, with demand expected to exceed supply through 2030 [4]. Micron's fiscal Q3 2026 release is likewise headlined by multi-year Strategic Customer Agreements [5] (the $22bn deposits figure is from a secondary transcript, I haven't verified it [6]). TSMC calls the gap "still quite big" [7]; CoWoS wafer figures are analyst estimates [8][9]. A TSMC customer-prepayment mechanism for CoWoS: I haven't verified this.
Rolling schedules, days to months. In automotive the signal is an EDI message. DELFOR is "sent from a party who is planning the use or consumption of products to a party who has to plan for the supply" and "can be used to authorize the commitment of labour and materials resources" [10]. DELJIT adds "precise delivery sequence and Just In Time schedule requirements", daily rather than weekly [11]. Adient's guide shows firm and forecast zones, daily buckets rolling into weekly and monthly over about four months [12]; VDA 4984 and 4985 are the German equivalents [13][14][15]. A general "rolling 12-month horizon" rule: I haven't verified this.
Every DELFOR is bilateral: the OEM sends to tier-1, tier-1 runs its own MRP and emits a new DELFOR to tier-2, and the signal re-forms at every hop with that party's safety stock, lot sizes and lead-time offsets. That re-forming is the bullwhip effect, named by Lee, Padmanabhan and Whang in 1997 from P&G diaper data [16].
Data spaces, new. Catena-X reported on 21 September 2026 that 1,069 OEMs and suppliers exchange data through the network, up from 182 in December 2025; the stated next phase is "smaller and mid-sized suppliers across the deeper tiers" [17]. Cofinity-X, founded January 2023 by ten OEMs and suppliers, operates it through 2028 [18][19]. The Demand and Capacity Management KIT covers "a tactical horizon of 1-2 years", exchanging WeekBasedMaterialDemand downstream and WeekBasedCapacityGroup upstream under standard CX-0128, explicitly to tackle "bullwhip effects and in the worst case even temporary production shutdowns" [20]. PURIS covers the short term: item stock, planned output, deliveries, days of supply [21], over the Tractus-X EDC connector [22].
2. Where it broke: chips, 2021 to 2023
Auto-chip buyers cut foundry orders in early 2020; when they returned the capacity had gone to consumer electronics, which orders ten times the units [23]. OEMs buy chips "through the automotive electronics suppliers and chip distributors, rather than direct orders from chipmakers", on mature nodes with 26-week lead times [24]. The US Commerce Department's January 2022 RFI found median buyer inventory fell from 40 days in 2019 to under five in 2021 while demand rose 17% [25] (secondary, commerce.gov blocked). AlixPartners put the 2021 cost at $210bn of revenue and 7.7 million vehicles [26]. The European Court of Auditors found in 2025 that EU crisis monitoring was "still in the very early stages" [27].
Visibility figures explain why: McKinsey's 2025 pulse found 95% with tier-1 risk visibility but only 42% into tier 2 or beyond [28], after deeper-tier visibility fell in 2024 [29]; BCI found 17.1% mapping to tier 4 and beyond [30].
So the breaks are: a DELFOR travels one hop while the chipmaker is three hops and 26 weeks away; forecast zones are not commitments, so capacity went to whoever committed; capacity information moves only as money, which mid-tier firms cannot offer; and the standard that addresses this (Catena-X DCM) sits at the top tiers, not yet where the gap is.
3. Where a startup could sit (my reading)
The gap is the tier-1 to tier-2 hop in the mid-market, where neither party can buy capacity the way Nvidia does and Catena-X has not yet arrived. The startup needs two feeds: from the tier-1, the DELFOR or VDA 4984 schedules it receives from its OEM and those it emits to its tier-2s; from the tier-2, weekly planned production output, item stock and capacity per line, which are exactly the PURIS and DCM objects. It sends back: to the tier-2, the OEM's forecast zone behind the tier-1's re-formed schedule, with a week-by-week confidence; to the tier-1, a WeekBasedCapacityGroup-style answer and an early flag where demand exceeds capacity 8 to 52 weeks out, which the tier-1 can pass up as a CX-0146 disruption notification. In effect: ship the DCM and PURIS semantics as a product for two companies that will not run an EDC themselves, with the data space as the eventual rail. Risks, both my inference: tier-1s may be contractually unwilling to reveal OEM forecasts downstream, and the top hop is already served by OEM portals and Cofinity-X marketplace apps.
Sources
[1] https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000075/nvda-20260726.htm [2] https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm [3] https://news.skhynix.com/sk-hynix-announces-3q25-financial-results/ [4] https://news.skhynix.com/en/q3-2026-business-results/ [5] https://micron.gcs-web.com/node/50671 [6] https://www.moomoo.com/community/feed/micron-q3-fy2026-earnings-conference-call-116806979420565 [7] https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC%203Q26%20Transcript.pdf [8] https://www.trendforce.com/news/?p=65758 [9] https://www.tipranks.com/news/tsmc-expands-ai-chip-packaging-as-supply-limits-shift [10] https://www.stylusstudio.com/edifact/D04B/DELFOR.htm (UN/EDIFACT D.04B mirror; unece.org blocked) [11] https://www.stylusstudio.com/edifact/D04B/DELJIT.htm [12] https://www.adient.com/wp-content/uploads/2021/11/AdientEDI-Implementation-Guide-EDIFACT-96ADELFOR-V30.pdf [13] https://www.seeburger.com/resources/good-to-know/edi-vda-4984-message [14] https://seeburger.com/resources/good-to-know/edi-vda-4985-message [15] https://www.gefeg.com/wp-content/uploads/2024/01/Training-VDA-message-4984-T2-Global-DELFOR-GEFEG.pdf [16] https://sloanreview.mit.edu/article/the-bullwhip-effect-in-supply-chains/ [17] https://catena-x.net/news/catena-x-network-grows-by-more-than-a-factor-of-five-to-over-1000-companies/ [18] https://www.henkel.com/resource/blob/1799832/61792db23f2c88312cb0b21af0f6bf44/data/2023-01-31-henkel-press-release-joint-venture-cofinity-x.pdf [19] https://www.cofinity-x.com/blog/cofinity-x-extends-mandate-as-operating-company-for-catena-x-through-2028 [20] https://eclipse-tractusx.github.io/docs-kits/kits/demand-and-capacity-management-kit/adoption-view/overview [21] https://github.com/eclipse-tractusx/puris/wiki [22] https://github.com/eclipse-tractusx/tractusx-edc [23] https://www.infobae.com/en/2021/01/11/missing-chips-snarl-car-production-at-factories-worldwide (Bloomberg, syndicated) [24] https://semiengineering.com/automotive-ic-shortage-drags-on/ [25] https://www.theregister.com/2022/01/25/doc_chip_shortages/ [26] https://www.alixpartners.com/media-center/press-releases/press-release-shortages-related-to-semiconductors-to-cost-the-auto-industry-210-billion-in-revenues-this-year-says-new-alixpartners-forecast/ [27] https://www.eca.europa.eu/ECAPublications/SR-2025-12/SR-2025-12_EN.pdf [28] https://karriere.mckinsey.de/capabilities/operations/our-insights/supply-chain-risk-survey [29] https://karriere.mckinsey.de/capabilities/operations/our-insights/supply-chain-risk-survey-2024 [30] https://thebci.org/news/supply-chain-disruptions-drive-increased-tier-mapping-and-insurance-uptake.html
how this page was made
Method and sources
Built 2026-11-03. The map is a taxonomy set by hand and filled by Claude: eight layers ordered by distance from the machine, plus two axes per company, how the software touches the machine and who closes the loop.
Companies came from six research passes run the same day: our own notes on the four studied companies, then scouts for Germany, Austria, Switzerland, Benelux and the Nordics; the rest of Europe and Israel; China, Japan, Korea and Taiwan; the US, Canada, India and other; and one pass for the supply-network and infrastructure layers. Each entry carries the URL the scout actually read and a line stating what was checked. Headcounts are LinkedIn bands or press figures where found; a question mark means not found, not small.
Layer placement is a judgement from the public product description. A company that reads sensors but sells a report to a plant manager is placed by what it sells, not by what it ingests; the marker shows the ingest side separately.
| count | what |
|---|---|
| 142 | companies on the map |
| 30 | in L0 Machine & sensor |
| 25 | in L1 Machine configuration & control |
| 27 | in L2 Output & product |
| 48 | in L3 Line & plant monitoring, causal analysis |
| 40 | in L4 Planning |
| 23 | in L5 Knowledge & people |
| 25 | in L6 Supply network |
| 29 | in LX Infrastructure & data |
| 61 | from Europe |
| 36 | from US / Canada |
| 11 | from Japan |
| 10 | from China |
| 9 | from Korea / Taiwan |
| 5 | from other |
| 4 | from studied up close |
| 3 | from Israel |
| 3 | from India |
| 116 | with a sourced headcount figure |
Schema the scouts filled
# Shared schema for the AI/ML-in-manufacturing value map (2026-11-03)
Write your output as a JSON array to the file named in your brief. One object per company:
{
"name": "Company",
"hq_city": "Munich", "country": "DE", "region": "europe|china|japan|korea_taiwan|us|israel|india|other",
"founded": 2019, // integer or null
"headcount": 45, // best integer estimate, or null
"headcount_source": "LinkedIn 11-50 range, checked 2026-11-03" | "press: <url>" | "not found",
"funding_note": "Series A EUR 12m 2025 (url)" | "",
"one_liner": "what they sell, in one plain sentence",
"layers": ["L0"], // one or more of the layer codes below, primary first
"machine_link": "reads_sensors|writes_to_machine|reads_mes_erp|documents_only",
"human_in_loop": "closed_loop|recommends_human_acts|human_reads_report|human_only",
"ai_kind": ["ml_timeseries","computer_vision","optimisation","causal","llm_rag","agents","digital_twin","physics_ml"],
"customers_or_verticals": "automotive tier-1, chemicals ...",
"supply_network": false, // true if they model upstream/downstream companies, not just one plant
"source_urls": ["https://..."], // at least one; the page you actually read
"verified": "what you actually confirmed, in one line; say 'headcount not verified' when so"
}
Layer codes (distance from the machine, innermost first):
L0 machine & sensor: raw signals, sensor correction, soft sensors, edge anomaly detection
L1 machine configuration & control: parameter setting, recipe optimisation, closed-loop control
L2 output & product: vision inspection, quality prediction, scrap and yield
L3 line & plant monitoring, causal analysis: OEE, root cause, downtime reports (human reads)
L4 planning: scheduling, machine allocation, inventory, MRP, ERP ordering, procurement
L5 knowledge & people: SOPs, work instructions, audit, training, RAG over documents, connected worker
L6 supply network: multi-tier demand and capacity signals between companies, 1-3 year horizon
LX infrastructure & data: factory edge/on-prem data centre, cloud, UNS/historian, MLOps for plants
Scope rule: headcount at most ~200 (use the LinkedIn range band or a press figure; if only "51-200"
is known, say so). Skip public giants and their divisions (Siemens, Rockwell, SAP, PTC, Dassault,
Palantir, C3, Uptake-size). Include anything from 2 to 200 people, seed to Series B/C.
Honesty: never invent a founding year, headcount or URL. null plus a note beats a guess.
Load notes
- duplicate dropped: Flowlity (network_and_infra.json, kept europe_rest_israel.json)
- duplicate dropped: United Manufacturing Hub (UMH Systems) (network_and_infra.json, kept europe_dach_nordics.json)
- duplicate dropped: Barbara (network_and_infra.json, kept europe_rest_israel.json)
- duplicate dropped: Cofactr (us_india_other.json, kept network_and_infra.json)
- duplicate dropped: HighByte (us_india_other.json, kept network_and_infra.json)