German AI Patent Growth Puts Technical Character in Focus
The German Patent and Trade Mark Office (DPMA) has put digital technologies back at the centre of the patent debate. Its recent trend analysis shows continued growth in patent applications linked to digital key technologies, with particularly visible movement in computer technology, audiovisual technology and related fields. Within those categories, generative AI, machine learning, virtual modelling and industrial digital twins are becoming harder to separate from one another.
The point is not simply that more AI-related applications are being filed. The more practical issue is that DPMA’s approach to computer-implemented and AI-related inventions keeps drawing attention to “technical character”. Applicants seeking protection in Germany for generative AI or digital twin inventions will need to show more than model capability, business usefulness or data value. The application should explain the technical problem, the technical means used and the technical effect that can be assessed in examination.
Growth is coming from use cases, not just model enthusiasm
DPMA’s digital technology analysis looks at areas such as computer technology, digital communication, semiconductors, audiovisual technology and data processing methods for business administration. Computer technology remains one of the most active areas, covering image data processing, speech recognition, information and communication technologies and many AI-related developments. Audiovisual technology is also becoming more relevant, especially where virtual reality, augmented reality and virtual representations of products, machines or industrial facilities support digital twin applications.
This tells us something important about the German market. AI patent competition is moving away from broad claims about general model capability and towards the question of how models are embedded in technical processes. Digital twins are a clear example. They are rarely just a software feature. In industrial settings, they may combine sensing, simulation, control, predictive maintenance and production decisions. Patentability is more likely to rest on how the system reduces error, shortens simulation time, improves control or makes equipment-state prediction more reliable.
Technical character is becoming the filter for AI patent quality
Under German practice, a computer program as such is not protected merely because it is expressed in code. For computer-implemented inventions and AI-related inventions, DPMA’s examination framework focuses first on whether the invention belongs to a technical field, then on whether excluded subject matter is involved, and finally on which features can count for novelty and inventive step because they solve a specific technical problem through technical means.
That distinction matters for generative AI filings. A training-data choice, model architecture, prompt workflow, business rule or generated output will not automatically be treated as a technical contribution for inventive-step purposes. The application needs to connect the algorithm to hardware, data processing architecture, controlled equipment, sensor inputs, network resources, storage behaviour or computing efficiency. The more the filing reads like a business concept, the more fragile it becomes. A filing grounded in technical constraints and measurable technical effects gives the applicant more room in prosecution.
Digital twin filings should not stop at a virtual mirror
Digital twin applications often run into a familiar weakness: they describe a virtual model, visual interface or management platform in detail, but say too little about how the virtual model technically interacts with the physical object. Examiners are likely to ask where the data comes from, how it is corrected, and how model outputs influence machine operation, testing, maintenance planning or production control. Without that chain, a digital twin can look like little more than a polished representation of information.
For manufacturing, automotive, semiconductor and energy companies, a stronger drafting strategy is to separate the invention into technical layers: sensing and data acquisition, model updating and error control, simulation or prediction algorithms, and closed-loop feedback to equipment or production processes. This will not guarantee grant. It can, however, reduce the risk that the invention is treated as abstract data processing or mere presentation of information.
Public-interest balance will be tested at the claim boundary
DPMA’s emphasis on technical character also performs a boundary-setting function. Patent law should reward real technical contributions without locking up general algorithmic ideas, ordinary business logic or overly broad uses of data. Clear claim boundaries matter to the public interest. They tell the applicant what is protected and give later developers a more reliable view of which technical paths remain open.
Applicants should adjust drafting strategy before filing, not during the first office action. The centre of the application should not be a broad statement of what generative AI can do. It should be a precise account of how, in a defined industrial or digital system, the invention uses technical means to solve a technical problem. That affects prior-art searching, priority drafting, experimental support, the number of embodiments and the likely rhythm of prosecution. Rising filing activity and stricter attention to verifiable technical contribution are part of the same story.
