EuroTPC at ISC 2026 | Bridging AI and HPC Infrastructure

Convergence of AI and Scientific Computing at ISC 2026

The ISC High Performance Conference concluded in Hamburg, Germany, after four days of intensive focus on the intersection of high-performance computing, artificial intelligence, and quantum systems. Held from June 23 to June 26, 2026, the event served as a central meeting point for the European research community and global experts. EuroTPC used this stage to push for a coordinated strategy, emphasizing how the union of traditional scientific computing and AI represents the next frontier for large-scale research. Fabrizio Gagliardi, representing the Barcelona Supercomputing Center, presented this vision at the EuroHPC Joint Undertaking booth. He argued that the current moment demands clear European action to align infrastructure capabilities with the needs of emerging AI models.

This shift is necessary because the requirements of frontier-scale AI often clash with the structural design of existing supercomputers. Scientific workloads require precision and reliability, while current AI models prioritize throughput and distributed training. Bridging this gap occupied the focus of the 3rd Trillion Parameter Consortium Workshop, which took place on the final day of the conference. This workshop examined the logistics of training trillion-parameter models, the use of synthetic data pipelines as standard HPC tasks, and the practical challenges of integrating these massive workflows into existing facility operations. Industry leaders and academic researchers debated whether the rise of dedicated AI factories suggests an evolution of current HPC centers or points toward a more fundamental change in their role.

Insights from the Trillion Parameter Consortium Workshop

The TPC workshop featured a keynote from Rio Yokota of the Institute of Science Tokyo. His address detailed the technical requirements for pre-training large language models, covering sparse expert models and the role of synthetic data in maintaining performance at scale. Following this, a roundtable moderated by Gokcen Kestor of the Barcelona Supercomputing Center explored the future of these facilities. Panelists included representatives from Oak Ridge National Laboratory, NVIDIA, and C-DAC. They questioned if AI infrastructure is a natural next step for HPC sites or a potential threat to their traditional scientific mission.

The afternoon sessions provided concrete examples of this convergence in practice. Presentations covered diverse topics such as generating synthetic biological reasoning problems on heterogeneous systems and the development of energy profiling frameworks for frontier-scale AI. Researchers from Argonne National Laboratory presented work on scaling structured tensor algebra for foundation model training on the Aurora system. Another presentation by Y. K. Singh of C-DAC highlighted how convergence between HPC and AI can tangibly improve disaster response. These sessions demonstrated that while the technical barriers are real, the progress toward integrated frameworks is already underway across global research institutions.

Strategic Implications for European Infrastructure

Beyond technical presentations, the conference addressed the governance of shared infrastructure. Europe is actively aligning its development goals with the broader Trillion Parameter Consortium and the EU’s AI Continent Action Plan. Reproducibility and the responsible use of computing resources remain top priorities for the consortium. Participants engaged in open discussions about how to standardize these practices across a diverse set of facilities and international partners. The event was co-organized by leaders from the Barcelona Supercomputing Center and Argonne National Laboratory, ensuring a collaborative approach to these complex questions.

These discussions point toward a future where AI and HPC are no longer treated as distinct silos. Instead, they are becoming increasingly unified through shared data management, unified software stacks, and cross-disciplinary infrastructure projects. The industry should watch for how upcoming grant cycles and national investments evolve in response to these technical findings. The alignment of AI development with scientific HPC infrastructure will likely determine the pace of progress in fields ranging from drug discovery to climate modeling. Looking ahead, the focus shifts to how these frameworks will scale outside of controlled workshop environments.