Global manufacturing is undergoing a seismic change as U.S. and European companies move to increase production closer to home, prompted by a range of economic and geopolitical factors.

In the United States, Apple, TSMC, Nvidia, IBM, and others have recently announced plans to invest hundreds of billions of dollars in American manufacturing facilities. One study found that 58% of U.K. manufacturing firms are reshoring their supply chains. Several EU initiatives, including the European industrial strategy, focus on making the manufacturing industry’s supply chain more reliable and less dependent on distant suppliers — weaknesses that were exposed by the COVID pandemic.

Manufacturers’ success in reshoring production — after decades of outsourcing it to lower-cost countries — will depend heavily on how smartly they can innovate to run their operations more efficiently. Not surprisingly, many are looking to generative AI (GenAI) for that productivity boost. According to a study released in early 2025, the market for AI in manufacturing is expected to grow over 31% a year until 2034, when it will be worth $61 billion.

The sector’s interest in AI is rising quickly, with 34% of manufacturers now anticipating that it will have a significant impact on their operations, a big jump from only 10% last year.

Despite this potential, AI deployment in manufacturing faces a range of challenges from compliance issues, unique legacy systems and processes, and workforce safety and training.

Let’s look at five core challenges manufacturers face in adopting AI and how they can overcome them.

1. Ensuring compliance with regulations

AI can help manufacturing firms align with emerging rules governing how organizations gather, analyze, and store data.

The challenge

Because they operate plants and supply chains across multiple jurisdictions, global manufacturers face enormous complexity when it comes to ensuring compliance to data privacy rules like GDPR and CCPA.

The need to tighten up data protection has been further complicated by yet more rules governing how AI can use data. The EU AI Act, for example, has potentially major implications for manufacturing companies. Any firm using AI in a way that impacts worker safety, quality assurance, or workforce management, for instance, could come under the act’s “high-risk” category, which imposes stricter standards on risk management, traceability, and data governance.

The solution

AI itself can be a powerful tool to help manufacturers get and stay in compliance with regulations targeting AI.

AI tools can scan internal systems and flag potential high-risk applications. They can also streamline documentation, ensuring it is audit-ready, and provide the traceability that regulators are demanding through features like built-in citations and data-lineage tracking. AI can also be used to monitor systems, detect anomalies, and highlight any compliance risks.

Retrieval-augmented generation (RAG) frameworks allow manufacturers to draw responses from pre-vetted company data and sharply reduce the risk of hallucinations, in line with regulatory standards.

GenAI can be carefully tailored to align with the needs of a local plant and its regulatory environment. With a private deployment, for example, an AI model can use on-premises servers or a dedicated private cloud to ring fence enterprise data and provide an additional layer of security.

2. Addressing real-time, mission-critical safety and maintenance

Manufacturers want to improve safety and maintenance for complex facilities without compromising sensitive data.

The challenge

Manufacturing plants are dynamic environments with complicated machinery that demands high safety and maintenance standards. U.S. manufacturers reported 355,800 worker injuries in 2023, while a global survey found that two-thirds of industrial companies suffer unscheduled downtime at an average cost of $125,000 an hour due to maintenance issues at least once a month.

Secure real-time monitoring for safety and maintenance requires sensors located on-site and away from central data centers. This means an AI solution may need to be self-sufficient and able to process data locally without relying on constant cloud connectivity.

The solution

Increasingly efficient AI models can achieve high performance with less energy and fewer specialized chips. This paves the way for systems that can be deployed at the edge to look for defects and predict failures, while remaining air-gapped from external connections that might imperil data security.

Edge AI can enhance worker safety by monitoring for hazardous conditions and providing immediate alerts. Interactive GenAI-powered agents that accompany workers can offer constant and comprehensive assistance wherever they are on the factory floor, a particularly useful solution for first-year employees who account for 30% of all injuries in manufacturing.

Use cases requiring edge, air-gapped solutions means that manufacturers will require AI models that can be deployed privately. Running AI models on their own servers helps manufacturers ensure data security while reducing latency (crucial for real-time, safety-related decisions) and making them less vulnerable to connectivity issues or third-party failures.

3. Integrating unique systems and software

Manufacturing is often built on a patchwork of specialized systems and software — requiring AI models that can access, understand, and integrate with them.

The challenge

The unique software systems that manufacturers rely on don’t have out-of-the-box integrations with most AI solutions (unlike more common platforms like SharePoint or Salesforce). And most of the software that manufacturers rely on is programmed in less common programming languages in which the majority of LLMs aren’t proficient.

For example, systems such as a manufacturing execution system (MES), programmable logic controller, computer-aided manufacturing, and others will frequently use proprietary formats and restricted APIs.

Meanwhile, specialized programming languages such as the Structured Control Language (SCL) used to control assembly lines and conveyor belts, or Hardware Description Language used to design integrated circuits and IoT devices, aren’t easily integrated with Python-based AI models.

Together, these incompatibilities can prevent manufacturers from realizing the same benefits from GenAI enjoyed by other industries, and hold back their plans to improve coding and debugging while increasing the overall optimization of their plants.

The solution

Tailored AI models can help manufacturers bridge this “integration gap.” By customizing, or training, these models on their unique data and programming languages, manufacturers can create a GenAI solution that truly understands their world, delivering the high levels of accuracy and reliability they need.

For example, an AI model could help convert high-level SCL logic into G-code for machining. Customized AI can also be deployed to assess and improve code that will optimize workflows, such as aligning the programming of a production-line robot with the programming that controls a conveyor belt’s speed.

Foundation models can also be trained on a manufacturer’s unique terminology, equipment, and workflows. A RAG system connected to internal documentation allows for easier troubleshooting and decision-making support across diverse systems.

Multimodal AI capabilities are also developing rapidly, giving organizations more powerful search capabilities that can access and understand a much wider range of data formats — including unstructured data that contains a trove of previously hidden value.

Take the example of an engineer trying to find the cause of a production stoppage. Instead of manually digging through a range of chaotic data and document types — from PDF manuals to handwritten inspection notes to raw sensor logs — to identify a problem, they could query a multimodal-equipped AI in natural language and receive an answer that understands and cross-references all of that diverse data.

4. Planning for scale

The siloed systems and software discussed earlier directly contribute to another challenge for manufacturers: their ability to scale with new technologies like AI.

The challenge

Factories often have a patchwork of legacy equipment, proprietary software, and siloed data that don’t easily communicate with one another. For instance, sensors on a production line may not talk to an inventory management tool, which is in turn isolated from software that governs quality control.

According to a study by MIT last year, 54% of manufacturing executives believe weak data integration hampers their ability to scale — with just 1 in 5 saying their data is currently ready for existing AI models.

The solution

While this lack of integration poses a problem for many conventional AI models, it can be resolved by newer models that are designed to knit these disparate systems together — helping manufacturers optimize their operations and ultimately unlock scale.

Another step in this evolution is the emergence of agentic AI. Autonomous AI agents can assess, plan, and act across a variety of languages, code, systems, images, and video to accelerate integration and drive optimization. Fujitsu, for example, expects AI agents will help orchestrate processes throughout its manufacturing value chain.

5. Overcoming talent and training gaps

A lack of expertise and training on AI tools can be an obstacle to faster adoption, especially in an industry like manufacturing that involves unique and complex processes.

The challenge

According to the Manufacturing Institute, U.S. manufacturers could have as many as 2.1 million unfilled jobs by 2030. And with half of manufacturers pointing to insufficient training as the reason they cannot unlock the potential of automation, robotics, IoT, and other innovations, having a productive workforce could become an even greater challenge now that leading manufacturers are stepping up their investment in new plants and technology.

The solution

GenAI is already helping manufacturers boost the productivity of current employees. Airbus, for example, used AI to allow their engineers to run thousands more tests, while AstraZeneca says GenAI is reducing development lead times by 50%.

GenAI has enormous potential for improving and accelerating training as well — with a recent study highlighting increased optimism among manufacturers that GenAI will help their employees identify problems, detect trends, and improve tasks like monitoring video of the factory floor.

Manufacturers can further meet this challenge by integrating AI into their onboarding programs and taking advantage of partnerships that offer AI training. They can also benefit from low-to-no-code AI applications that reduce the need for deep technical expertise. North, for example, allows enterprises to build and deploy AI agents to help employees carry out specific workflows.

Catching the next productivity wave

The smart, strategic adoption of AI is essential if manufacturers are to adapt successfully to the sweeping global trends and massive new investments reshaping the industry. To capture the full potential of the technology, manufacturers need solutions that are fully attuned to their unique needs — which means working with expert partners to develop custom, secure AI that truly understands their world, allowing them to accelerate and scale adoption.

Building enterprise AI agents

A comprehensive blueprint for developing scalable AI agents in regulated industries.

Download the eBook

Five ways to modernize manufacturing with AI