Fraunhofer Institute for Industrial Mathematics ITWMFrom software to hardware execution: together with our project partners, we are developing an open European framework for Generative AI and Large Language Models (LLM) in the »OptimAIse« project. This creates the foundation for efficiently developing, deploying, and operating powerful, sustainable, and trustworthy AI solutions.
Illustration eines transparenten Roboters auf einem leuchtenden Mikrochip, umgeben von einer dunklen Platine mit türkisen Leiterbahnen.
How can generative AI and Large Language Models be developed and operated efficiently, sustainably, and in a trustworthy manner? In the »OptimAIse« project, together with European partners, we are developing an open framework that connects modern AI software with European hardware, thereby strengthening Europe’s digital sovereignty.
© iStockphoto
European Framework for Trustworthy Generative Artificial Intelligence
Project »OptimAIse«: Generative AI, Large Language Models and European AI Hardware
Generative AI (GenAI) and Large Language Models form the foundation of many applications today – from web search and coding assistance to customer service. So far, there is no end-to-end open technology platform for developing, deploying, and operating GenAI and LLM applications. Existing cloud solutions often do not meet the specific requirements of GenAI applications. This makes deployment and operation complex, non-transparent, and inefficient.
Open Architecture for Generative AI
In the »OptimAIse« project, we are developing a scalable, modular, and interoperable reference architecture for Generative AI. It leverages European hardware platforms and enables the efficient use of large language models. As LLMs introduce new technical challenges, existing software engineering methods often reach their limits. »OptimAIse« therefore establishes a common foundation for the development and operation of modern AI systems. Our goal is to support developers in building scalable, sustainable, and trustworthy applications based on generative AI. In doing so, »OptimAIse« makes an important contribution to advancing high-performance and responsible AI in Europe.
We pursue the following goals:
- Strengthening Europe’s technological independence: The AI accelerator market is currently dominated by a small number of non-European providers. »OptimAIse« integrates European hardware solutions as a compatible alternative, thereby promoting digital sovereignty and sustainability.
- Establishing a unified system architecture: A shared reference architecture for LLM applications facilitates component reuse, reduces development costs, and lowers security risks.
- Easing compliance with regulatory requirements: The European AI Act imposes strict requirements for documentation, transparency, robustness, and human oversight. An open standard architecture helps implement these requirements efficiently.
- Making sustainability measurable: In addition to performance and cost, »OptimAIse« also considers energy consumption, CO₂ emissions, and water usage as optimization targets. This improves planning, efficiency, and resource conservation.
Efficient Operation of AI Systems
At the core of the framework is an adaptive orchestration platform. It continuously optimizes the operation of AI applications in terms of performance, energy efficiency, and sustainability. In addition, »OptimAIse« provides tools that enable AI systems to be designed and evaluated in a transparent, robust, and fair manner. To achieve this, we develop open-source libraries and integration tools that enable European hardware solutions to be easily integrated into established AI frameworks.
Together with European chip developers, we transfer insights from real-world applications directly to the next generation of processors. This strengthens Europe’s technological infrastructure, digital sovereignty, and competitiveness.
Our Contribution: Efficient AI Algorithms for European Hardware Platforms
Our core task is to connect modern European hardware with scalable and efficient AI software. Our focus is on enabling the optimal execution of generative AI on new, energy-efficient European AI accelerators.
To achieve this, we use, among others, the STX Stencil and Tensor Accelerator from Fraunhofer as well as the SpiNNaker2 system from TU Dresden. These platforms enable detailed analyses of performance bottlenecks in modern chip architectures. Based on this, we develop algorithms and methods that reduce communication overhead and make optimal use of available computing resources.
Our Projectpartners
»OptimAIse« brings together seven small and medium-sized enterprises (SMEs), three research centers, one university, and two public administrations. At Fraunhofer ITWM, we work closely with our partners at TU Dresden, in particular with the Chair of Highly-Parallel VLSI Systems and Neuro-Microelectronics.
- [Coordinator] Tree Technology SA (SME), Spain
- Hiro Microdatacenters B.V. (SME), Netherlands
- Tages (SME), France
- Barcelona Supercomputing Center, Spain
- CyberEthics Lab SRLS (SME), Italy
- Biti Innovations AB (SME), Sweden
- Multiverse Computing SL (SME), Spain
- Technical University of Dresden, Germany
- Servicio Andaluz de la Salud, Spain
- [Associated] Fundación para la Gestión de la Investigación en Salud de Sevilla, Spain
- OFFIS e.V., Germany
- F6S Network Ireland Limited (SME), Ireland
Project Duration and Funding
The project runs from June 1, 2026, to May 31, 2029, and is funded as a »Research and Innovation Action« under the HORIZON Europe program, Grant Agreement No. 101296742, »Software Engineering for AI and Generative AI«.
