iPronics and the Barcelona Supercomputing Center are collaborating to address a critical bottleneck in artificial intelligence infrastructure. As GPU clusters expand to support demanding workloads like Large Language Models and Mixture-of-Experts, networking must evolve beyond static designs, according to iPronics CEO Christian Dupont: “AI infrastructure is reaching a point where scaling compute requires rethinking the network.” The two-year strategic technology collaboration will focus on programmable optical networking, integrating iPronics ONE, a rack-ready platform designed to bypass complex software development, into BSC’s high-performance computing environment.
iPronics ONE Platform Streamlines Optical Integration for AI Workloads
iPronics ONE bypasses historical limitations in data center deployment by delivering a rack-ready, plug-and-play platform for programmable optics, eliminating the need for complex, custom software development. This approach directly addresses a critical bottleneck as artificial intelligence infrastructure scales, demanding more dynamic networking aligned with the behavior of modern AI workloads. The platform’s design intends to integrate seamlessly into existing software stacks, allowing cloud providers and hyperscalers to maximize GPU utilization without extensive system overhauls.
BSC’s expertise is particularly relevant as Spain’s national supercomputing centre; it leads Quantum Spain, the national quantum computing project, and operates at the intersection of conventional and quantum computing with the MareNostrum supercomputer. This position allows for unique system-level insights, as the BSC team, led by Antonio J. Peña, will develop the software intelligence that sits above the iPronics ONE switch-management layer.
The partnership aims to connect the communication requirements of Large Language Models (LLMs) and Mixture-of-Experts (MoE) applications directly with dynamically reconfigurable optical connectivity, the company says. iPronics provides the programmable optical switching platform, low-level control software, and open APIs, while BSC focuses on the higher-level software that orchestrates the network based on AI workload demands. This full-stack innovation is intended to overcome the communication bottlenecks created by static interconnects at the scale required for these demanding applications, a problem that has become increasingly acute as GPU clusters expand. The collaboration seeks to demonstrate workload-aware network control across both hardware and software, ultimately lowering latency and reducing energy consumption.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
