Executive Insights - OFC 2025 Video Interview

Exploring the Future of Scalable AI Infrastructure with Keysight at OFC 2025

In this exclusive OFC 2025 interview, Anna Berry from Lightwave sits down with Ram Periakaruppan, VP / GM, Network Applications and Security Products at Keysight Technologies, to discuss the growing challenges—and emerging solutions—facing enterprise teams as they scale AI infrastructure.

Ram opens the conversation by identifying three primary hurdles confronting AI infrastructure today: power limitations, memory bottlenecks, and the rising demand for AI reasoning and inference. As compute requirements increase, so does the strain on available power and memory architecture, making it essential to innovate both at the silicon and systems level. Keysight sees reasoning as the next frontier for AI — requiring not just more compute, but smarter, faster infrastructure.

This is where KAI, Keysight’s AI workload emulation platform, comes in. Developed in response to the complexity of large-scale, distributed AI systems, KAI enables teams to simulate real-world AI traffic patterns without relying on scarce GPUs or exposing proprietary data. KAI helps developers isolate and troubleshoot issues within AI networks while maintaining high fidelity, all without disrupting live environments. The platform supports validation earlier in the design cycle and offers critical visibility into performance at scale.

Ram also highlights the tangible business benefits of Keysight’s AI solutions: significantly reducing AI training times, optimizing resource usage, and helping customers validate custom silicon in pre- and post-silicon phases. These advancements reduce both CapEx and OpEx, allowing organizations to bring models like ChatGPT to market faster while improving infrastructure efficiency.

A key differentiator for Keysight, Ram explains, is the company’s design-to-deployment continuum — a unique ability to support AI systems from the earliest design stages through real-world deployment. This continuity ensures consistency, accelerates development, and minimizes rework. In addition, Keysight works with industry bodies like ML Commons to help standardize AI benchmarking and testing methodologies, further reinforcing its leadership in the space.

Looking ahead, Ram predicts a shift in customer needs over the next 2–3 years. More enterprises are moving away from vertically integrated, proprietary stacks toward modular, standards-based infrastructure. At the same time, the focus is shifting from just AI training to inference at scale, as more end users interact with AI in real-time applications.

To meet these evolving needs, Keysight is partnering closely with leading companies to co-design and validate bespoke infrastructure solutions, while ensuring flexibility across hardware, virtualized platforms, and open ecosystems.

This insightful discussion underscores Keysight’s strategic role in enabling the future of AI infrastructure — delivering tools, platforms, and expertise that allow customers to build scalable, efficient, and future-proof AI data center systems.