AI Adoption in the Chemical Industry: What the Data Shows

The adoption of AI in the chemical industry is progressing more slowly than in some other sectors – yet the growth curve is clearly turning upwards. According to a study by the Fraunhofer ISI Institute, only eight per cent of companies in the chemical and pharmaceutical industries were using artificial intelligence in 2022. A further 13 per cent are expected to join them by 2025. The study also shows that companies which have once overcome the initial hurdle are consistently expanding the scope of application. The technology is thus becoming a permanent part of operational strategy, rather than a one-off experiment.

A structural pattern is striking: Companies with more than 500 employees find it easier to get started than smaller companies, and AI applications are more frequently concentrated in sectors where complex products are manufactured. This is precisely why the process industry is particularly well-suited as a field of application – numerous production and reaction processes cannot be fully modelled using rigid rules, making data-based models a useful complementary tool.

Fields of Application in the Process Industry

In process monitoring and control optimisation, sensor data from running plants is processed in real time to automatically adjust parameters such as temperatures, flow rates or dosing quantities – approximately, when raw material properties fluctuate or reaction conditions change. The control adjustment continues until optimal conditions for the respective reaction are achieved, without the need for manual intervention.

Another area of application concerns molecular simulation in product development. AI systems are capable of simulating a very large number of molecular variants and mathematically predicting their interactions. In the pharmaceutical sector, this can significantly shorten the development time for new active ingredients. This method is also used for plant protection products and detergent formulations to identify suitable compounds with good environmental compatibility in the early stages of development. The computational effort required for this is considerable: an industrially used supercomputer achieves a computing power of three petaflops, which is equivalent to around 20,000 standard laptops.

AI-supported anomaly detection systems can detect irregularities in reaction processes earlier than traditional monitoring methods. Leaky valves or pipes, poor electrical contacts, as well as vibration and temperature deviations that indicate impending defects can thus be identified and rectified at an early stage. In the coatings industry, imaging AI technology is used to assess corrosion defects, a task previously carried out through subjective visual inspections. By capturing images in a standardised format and enriching them with experimental metadata, inconsistencies arising from individual human perception are systematically eliminated.

In Advanced Process Control (APC), mathematical models are combined with real plant data to continuously adapt control strategies. A documented case study from the chemical industry demonstrates that a batch plant has been operating fully automatically since mid-2024 – from production planning through to the filling line and packaging – although the project was originally designed for just ten test batches. All plant data flows into a cloud infrastructure, where it is optimised by AI algorithms and then fed back into the system. The plant manager in charge summed up the effort required to qualify the model by comparing it to the idea that AI must be treated like a student who needs to be trained until he becomes a professor.

Requirements and Barriers to AI implementation

The biggest hurdles to AI adoption do not lie primarily in the technology itself, but in the basic requirements that must be met for successful operation. Data-driven models work exclusively with what they can recognise in the training data. It follows that a significant amount of high-quality, structured production data must be available before a model can be trained effectively – a prerequisite that is not yet met in many organisations.

Successful AI implementations also require employees in production environments to develop specific digital skills. In the case study mentioned, production staff were empowered to build up the relevant skills, and accompanying digitalisation initiatives were launched in parallel. The conclusion drawn from operational practice is that AI deployment does not work without active human support.

On the technical side, the development of generative AI assistance systems for automation engineers shows that the barrier to entry for programming is falling significantly. At the 2025 Hannover Messe, it was demonstrated that a system can now be automated in a matter of minutes without requiring knowledge of specific engineering tools. Such systems can speed up SCL (Structured Control Language) code generation by up to 60 per cent whilst simultaneously reducing error rates.

One of the structural drivers for the use of AI is demographic change in the skilled workforce. Many skilled workers from the baby boomer generation are committed to retirement and cannot be fully replaced. A higher degree of automation is not a panacea, but it can be an effective lever, provided that the growing digital skills gap between employee groups is countered through targeted further training measures.

The carbon footprint of the AI infrastructure itself must not be overlooked. The software industry is already responsible for around three to four per cent of global CO₂ emissions – a share equivalent to that of the aviation sector and one that continues to rise. For the process industry, which is committed to improving sustainability, this aspect must be factored into the overall assessment when strategically planning AI projects.

At European level, in February 2025, sixty leading companies launched the EU AI Champions initiative, which aims to establish a comprehensive AI strategy as well as a simplified legal framework for Europe. Market developments underscore the scale of the transformation: according to Statista, the German market for generative AI alone is set to grow by a total of 14.1 billion US dollars between 2024 and 2030, representing an increase of 887.5 per cent. The forecast market size for 2030 is 15.7 billion US dollars.

Source: Trade journal ‘PROCESS’

Photo: unikyluckk