iPronics Partners with BSC to Cut AI Network Latency
iPronics and the Barcelona Supercomputing Center are testing programmable optical switching to boost GPU efficiency in AI clusters.
Key points
- iPronics and BSC are collaborating for two years to integrate optical switching with AI workload management.
- The partnership aims to reduce latency and energy use in GPU clusters without replacing existing software.
- iPronics recently raised $125 million in Series B funding to expand its AI infrastructure product strategy.
iPronics has entered a two-year strategic partnership with the Barcelona Supercomputing Center (BSC) to develop programmable optical networking for high-performance computing. The collaboration aims to integrate optical switching hardware with software to support dynamic connectivity across GPU workloads, addressing the growing bottleneck in AI infrastructure.
Traditionally, adding programmable optics to data centers required custom software development, creating a barrier for cloud providers. This partnership seeks to remove that friction by offering a rack-ready architecture that fits into existing environments. The goal is to improve GPU utilization and reduce energy consumption without forcing operators to replace their current software stacks.
Integrated hardware and software stack
Under the agreement, iPronics will supply its programmable optical switching platform along with low-level software and application programming interfaces. BSC will focus on developing the higher-level software that manages the switch. By working together, the teams are creating a unified architecture that aligns the communication needs of AI applications with dynamically reconfigurable optical connections.
This integration allows for workload-aware networking, where the network adapts in real-time to the specific demands of different AI tasks. The system is designed to handle both training and inference loads, including complex models like Large Language Models and Mixture-of-Experts architectures. This approach aims to maximize the value of expensive GPU hardware by ensuring data moves as efficiently as possible.
Testing in research-class infrastructure
The new platform will be deployed within BSC’s research-class GPU and high-performance computing infrastructure. This setting provides a realistic environment for testing how optical networking performs under heavy AI workloads. According to engineering.com, the collaboration will generate system-level insights that help inform the design of future AI data centers.
The move comes as iPronics expands its market presence following a $125 million Series B funding round. The company, which has raised a total of $177 million, recently opened a U.S. office in Santa Clara, California. This expansion signals a broader push to integrate optical switching into the standard architecture of AI infrastructure.
Trade-offs and practical limitations
While the plug-and-play promise is attractive, the technology is still in the development phase. The partnership is focused on advancing the architecture, meaning widespread availability and standardization are not yet guaranteed. Operators must weigh the potential for reduced latency against the current need for specialized integration efforts.
The primary trade-off is that this solution targets high-end, research-grade environments first. Smaller enterprises may not immediately benefit from the optimized workflows demonstrated at BSC. However, the insights gained from this collaboration will likely shape the product roadmap for future commercial deployments, potentially lowering the barrier to entry over time.
