By industry
What manufacturing owners and operators have put into production, what it cost and what went wrong.
Collection: Manufacturing AI use cases (Verified and Reported only) →
This page covers AI in manufacturing and the factory changes around it: inspection, castings that cut parts, 3D printing and software-run plants. Humanoid-robot and robot-dog pilots in car plants are collected separately in Manufacturing AI use cases.
Fewer parts is the first lever. Ford says it has cast more than 5,000 unicastings for its Fathom electric truck, with two replacing 146 structural parts. They are feeding prototypes ahead of a 2027 launch. [1] Honda cut one battery case from more than 60 parts to five, then cancelled the three EVs it had planned to build in the U.S. [2][3]
Inspection is the second. One of Ford's in-house AI vision tools runs at nearly 700 North American workstations, Automotive News reported, though Ford has published no savings. [4] Microsoft says AI that points Rolls-Royce inspectors at likely turbine-blade defects raised machine utilization 30%, a vendor claim. [5]
Whole plants are being benchmarked too. The World Economic Forum says GlobalFoundries' Singapore fab raised labour productivity 40% with more than 60 AI and digital use cases. [6] And the U.S. Navy says 3D printing cut lead times 70% in its headline 2025 example. [7]
Each case below covers what was deployed, what it cost where disclosed, and what went wrong.
© 2026 Real World Wire. Written October 29, 2026.
The castings feed prototype builds of the Fathom electric pickup at Louisville Assembly Plant, which Ford is retooling for about $1.9 billion. Launch is set for 2027, so there are no production results yet.
Tesla's filings list more than 125,000 a year of installed Cybercab capacity at Gigafactory Texas. Tesla hasn't said how many it has built, and says battery supply is limiting vehicle output.
Synthetic images of circuit boards let the plant prepare inspection before the first physical prototype exists, a step that used to take several weeks. Siemens' false-call and yield figures are still targets.
Hexagon's AEON robot is delivering materials and testing battery work at BMW's first humanoid pilot in Europe. BMW hasn't published any results from Leipzig.
BMW says, in its own words, that its Figure 02 pilot supported production of more than 30,000 X3s over ten months. Its successor, Figure 03, has started a parts-sorting demonstration. There's still no cost data, and no word on whether the first robot hit its speed target.
Slate says its pickup has fewer than half the parts of a typical truck, and it plans to build it in a former printing plant in Warsaw, Indiana, with no paint shop. First deliveries are expected in Q4 2026.
Boston Dynamics' four-legged robot inspects car bodies at the end of an almost fully automated weld shop. Hyundai owns both the plant and the robot maker, and it hasn't published results.
Rivian says it designed the R2 and its Normal, Illinois, production line together, cutting wiring, welds and fasteners. Capacity is 155,000 a year; output and cost figures are not public.
Honda built a megacasting process for the Honda 0 Series battery case, then cancelled three U.S.-built EVs in March 2026 and booked operating expenses of up to ¥1.12 trillion. Fewer parts didn't fix weak demand.
At Lynn, Mass., blasting media settling in a wet-blast machine cost more than an hour of setup per batch and three to five hours a day. GE says fixing it, with other small changes, cut lead time on the F404 turbine shaft 49%.
The air-conditioner maker and Hitachi built a chat tool that reads equipment drawings and suggests causes and fixes. Hitachi now sells it, and the only test results come from the two companies.
A supplier printed copper-nickel Virginia-class components in under five weeks against a typical 17-month wait, the shipbuilder installed its first metal-printed submarine part, and a destroyer valve's 29-week lead time fell 70%.
Audi's Edge Cloud 4 Production now runs a high-volume body shop at Neckarsulm, where about 100 robots take their commands from the cloud. Audi published no cost savings.
In early U.S. pilots with Siemens and NVIDIA, PepsiCo reports a 20% throughput increase and 10–15% lower capital spending from finding hidden capacity in plants it already owns. The sites and baselines aren't named.
The automated precision-parts maker says F3 is already producing aerospace and defense components. It hasn't published cycle-time, cost or output figures for the plant.
Figure says its Figure 02 robot ran 10-hour weekday shifts on a BMW body-shop line and loaded more than 90,000 parts before being retired. It published its speed and accuracy targets but not whether the robot met them.
Three kaizen rounds on a turbine-manifold line in India moved it to single-piece flow, cut changeover 20%, put one operator where two had worked, and took one step's defects from five a day to zero, GE says.
Short of talent and facing surging demand, Fab 7 used machine-learning maintenance, remote support, ML quality control and digital workflows, and cut prototyping time 30%. It earned Global Lighthouse status.
Faced with jetty congestion and more complex crude and products, the refinery joined up planning, inventory and logistics with AI forecasting. On-time delivery rose from 85% to 95%.
BMW says every operation at its new Hungarian plant was tested in a digital twin first, including the position of nearly 1,000 robots. It also says Debrecen is its first car plant to run without fossil fuels.
Printing harness paths and attachment points into each panel replaces dozens of parts, long-lead tooling and bonding steps, Boeing says. The hardware is still in qualification and hasn't flown.
Two in-house systems, AiTriz and MAIVS, flag misfitted connectors and missing parts at the station, before end-of-line checks. Ford hasn't published what they've saved.
GE's Celma overhaul shop was holding more than 24,000 parts in work-in-process against a standard of 19,000. Sending one blade at a time through the whole repair route cut turnaround time 58%, GE says.
Using its production system that 3D-prints structures and assembles them automatically, Divergent says it readied a decades-old Raytheon design for production in under five months, without the original tooling, and cut another airframe's part count by 80%.
Mercedes ramped up the new CLA at Rastatt in a hall it reconfigured virtually within weeks. AI now runs paint sub-processes that conventional controls used to handle. Investment: a mid triple-digit million euros.
BMW now runs digital twins of more than 30 plants and simulates every new model's path down the line before building anything. It projects planning costs will fall by up to 30%. That figure is a forecast.
A BMW spokesperson told Fortune a single Figure robot worked at Spartanburg at any given time, and only outside production hours until March. Figure's CEO had described a fleet doing "end-to-end operations."
With no hard tooling, Divergent printed and robot-assembled fuselages, wings and fins for CoAspire's low-cost RAACM missile by week 10, and test rounds entered ground testing by week 14, the companies say.
Rolls-Royce used to check about 2 million turbine-blade cooling holes a month by hand. Its Signature Analyzer uses machine vibration data and generative AI to tell inspectors where to look, Microsoft says.
Technicians photograph turbine blades and AI picks which images they review. GE says the tool cut inspection time in half on GEnx engines over about three years, and is now going to more than a dozen MRO sites for narrowbody LEAP engines.
Toyota built an in-house platform so shop-floor staff, not AI specialists, create the machine-learning models. Models made rose from 8,000 in 2023 to 10,000 in 2024, and the platform runs in all 10 of its car and unit factories.
A Figure 02 robot placed sheet-metal parts into welding fixtures during a trial at Plant Spartanburg. BMW calls it a success and says it's still working out where humanoids fit. No cost, speed or uptime figures were released.