Medical device manufacturing is changing on two fronts at once: the technology on the production floor and what regulators expect to see in the records. For a US manufacturer there is a third front that most trend lists skip: how each of these technologies shows up in the quality system and in the FDA file. Since February 2, 2026, the quality records behind that file fall under FDA’s Quality Management System Regulation (QMSR), which amended 21 CFR part 820 to incorporate ISO 13485:2016 by reference.
These are the seven shifts that carry the most weight now, with what each one means in practice for a team that builds devices or outsources production.
Table of contents
1. AI and machine learning on the production floor
AI in medical device manufacturing is used today on three concrete jobs: quality control, predictive maintenance and process optimization. Camera systems with trained models inspect parts for defects a human inspector can miss, and models that read equipment sensor data predict failures before they stop the line. The NIST Manufacturing Extension Partnership lists both among the main uses of AI in US manufacturing.
For the file, the rule is simple. If an algorithm takes part in accepting or rejecting a part, it is part of the quality system and has to be validated as such. ISO 13485 requires validation of software used in the quality system and in production (subclauses 4.1.6 and 7.5.6), and FDA’s guidance on computer software assurance for production and quality management system software, revised on February 3, 2026, applies its risk-based approach to AI and machine learning tools used there. Keep records of what the model checks, how it was tested against known good and bad parts, and what triggers a retest. A model that works on the line but cannot show how it decides is hard to defend in an inspection.
2. 3D printing and additive manufacturing
3D printing earns its place where the geometry is complex or the device is matched to one patient, as with implants and surgical instruments. FDA’s guidance Technical Considerations for Additive Manufactured Medical Devices (December 5, 2017) names both advantages: devices matched to the anatomy in a patient’s own imaging, and porous structures and internal channels that are hard to make with traditional manufacturing. On the line, three practices come with it:
- In-process monitoring of the build, such as melt pool data and build chamber conditions, which the guidance lists among the methods for process validation.
- Non-destructive checks of finished parts with CT, micro-CT or ultrasound for geometry and internal structure.
- Printing at the point of care, inside the hospital, for which FDA published a discussion paper with possible oversight approaches in December 2021.
On the regulatory side, a patient-matched device is usually not a “custom device”. The guidance says these devices qualify for the custom device exemption only if they meet every criterion of section 520(b) of the FD&C Act, and most go through the existing pathway for their device type. What the submission has to describe is the design envelope: the minimum and maximum dimensions and the performance limits inside which each patient’s version is made. Settle that envelope with your regulatory lead before you scale production.
3. The connected factory (Industry 4.0)
Connected sensors, industrial IoT devices and analytics add up to a production line that reports on itself, with real-time performance data, early warning of breakdowns and bottlenecks, and a cleaner record of what happened to each lot.
This matters more in medtech than in most industries, because continuous traceability is exactly what an FDA inspection asks for. Since February 2, 2026, FDA inspects device makers under the updated compliance program 7382.850 and no longer uses the old QSIT technique. The QMSR also gives FDA the authority to inspect management review, quality audit and supplier audit reports, which the old regulation kept out of reach (FDA QMSR questions and answers). A plant with well-recorded process data walks into that inspection with half the work done. Our summary of what changed under the QMSR lists the rest.
4. Biocompatible materials and miniaturization
Material work follows the device. Polymers have to survive the sterilization method the plant can use, implant alloys have to resist wear, and every surface that touches the patient has to pass biological evaluation. Every change of material, supplier or process reopens the biocompatibility evaluation. FDA clears the finished device and does not clear individual materials, so its ISO 10993-1 guidance asks the risk assessment to cover processing, manufacturing methods and sterilization as well. It is the line item teams underestimate most often in the schedule.
The standard itself has moved. FDA partly recognized the sixth edition, ISO 10993-1:2025, in May 2026, and it accepts declarations of conformity to the 2018 edition in premarket submissions until July 1, 2029. A biological evaluation planned now has to hold up after that date, so it is worth planning it against the new edition.
Devices also keep getting smaller. Micro injection molding, micro-machining and precision assembly make parts for minimally invasive instruments, drug delivery and implants, and the tolerances leave little room for a late material change. Small wearables and implants often carry a radio as well. If yours can connect to the internet, section 524B of the FD&C Act adds cybersecurity content to the premarket submission, and the cyber device check shows whether it applies.
5. Next-generation quality control and validation
- Non-destructive testing with imaging, which FDA’s additive manufacturing guidance accepts in acceptance activities when the choice of technique is discussed and documented.
- Real-time monitoring of critical process parameters, with alarms tied to the validated ranges.
- Validation supported by automated tools, assured with the same risk-based approach as the rest of your quality system software.
- Digital twins and computational models that test a change virtually before anyone touches the line.
If model results go into a submission, FDA’s final guidance on assessing the credibility of computational modeling and simulation (November 2023) describes the evidence it expects. Plan that evidence when you build the model, because it is much harder to add afterward.
6. RegTech: compliance gets automated
Compliance technology pays back most for a manufacturer that sells the same device in more than one market, because the file is built several times for several regulators. In practice it covers four things:
- records and documentation generated by the process itself;
- compliance monitoring in real time instead of once a year at audit;
- traceability that covers the whole supply chain;
- submissions assembled with less manual work; FDA already requires the eSTAR electronic template for 510(k) and De Novo submissions, so structured data has a place to go.
Take a US company that also sells in Canada and the EU. It keeps a 510(k) file and the QMSR records FDA can inspect. It goes through MDSAP audits: Health Canada requires an MDSAP certificate for a new class II, III or IV license, and FDA may accept MDSAP audit reports in place of a routine inspection. And it maintains EU MDR technical documentation for its notified body. If the product is a legacy device still sold under a directive certificate, Regulation (EU) 2023/607 lets it stay on the EU market until December 31, 2027 (class III and class IIb implantables, with exceptions) or December 31, 2028 (other classes), if the conditions are met; the EU MDR transition deadline check gives the date for your device.
That is three or four sets of documents fed by the same plant data, so each point of automatic capture pays back several times. The saving is real, and few trend lists mention it. The paperwork is also a large share of the work: in OVA’s hours benchmark, documentation and verification were estimated at 15 to 31% of the engineering hours in four proposals.
7. Cleaner production and a digital supply chain
Cleaner production
In US device manufacturing, sterilization is where environmental pressure and regulation meet. Ethylene oxide is still the most common sterilization method for medical devices in the US, and FDA presents its work on alternatives as a way to reduce ethylene oxide use and strengthen the supply chain. On January 8, 2024, it added vaporized hydrogen peroxide to Established Category A, which reduces the sterility information a 510(k) needs for that method. A design whose polymers, adhesives and packaging tolerate hydrogen peroxide keeps more sterilization options open. Energy use, raw material and scrap are the other levers, and most of them depend on the same process data as the connected factory in trend 3.
Digital supply chain
- Demand forecasting from historical data and inventory tracked in real time, both on NIST’s list of AI uses in manufacturing.
- Analytics that flag a supplier disruption before it reaches the line.
- Traceability built on FDA’s Unique Device Identification system: a UDI on the label and device data in the GUDID database, so a device can be identified from manufacturing through distribution to patient use.
Regional production and resilience
Investment in local capacity, regional supplier networks and distributed manufacturing all answer the same lesson: one long, single-source chain is a business risk. FDA asks for early warning too. During or in advance of a public health emergency, section 506J of the FD&C Act requires makers of certain devices to notify FDA of a permanent discontinuance or an interruption in manufacturing, and FDA encourages voluntary notices at other times. Trade policy adds pressure: in September 2025 the Commerce Department opened a Section 232 national security investigation into imports of medical equipment and devices.
For the US market, nearshoring usually means Mexico, which the International Trade Administration describes as the largest medical device exporter in Latin America and the eighth largest in the world, with most of its production going to the United States. For a manufacturer, the decision is which part of the process to keep close and which part can sit farther away.
Where to start
None of these seven shifts is adopted all at once. The order that wastes the least money is almost always the same. First comes traceability of process data, because it feeds quality, predictive maintenance and the regulatory file at the same time. Compliance automation comes next, and only then the production technologies that depend on the specific product. Buying a 3D printer before the data is in order is the most expensive mistake and the most common one.
The same logic applies before production. Our guide to the medical device product development process walks through the nine phases from intended use to post-market surveillance, and the development cost calculator shows what the engineering is likely to cost, phase by phase.
OVA Solutions is a medical device engineering group with 62 engineers and 220 devices developed, working under an ISO 13485:2016 certified quality system. We design electronics, firmware and mechanics and take a device from concept to design for manufacturing. The regulatory strategy and the submission stay with you or your regulatory consultant. If you are deciding which of these shifts makes sense for your product, book a 30-minute engineering call.
More on this topic: Manufacturing & DFM